Infants at high and low likelihood for autism show different EEG developmental trajectories in speech tracking and statistical learning
- Department of Psychiatry, University of Geneva School of Medicine, Switzerland
- Division of Adult Psychiatry, Department of Psychiatry, University Hospitals of Geneva, Switzerland
- Cognitive Neuroimaging Unit, CNRS ERL 9003, INSERM U992, CEA, Université Paris-Saclay, NeuroSpin Center, France
- Département d’étude Cognitives, École Normale Supérieure, France
- Aix Marseille Univ, INSERM, INS, Inst Neurosci Syst, France
Figures
Figure 1
Experimental procedure and multivariate statistical analyses.
(A) The learning part was sandwiched by a silent resting state (RS) and a random stream (RND) with even transition probabilities between syllables. This design accounted for the potential effect of time during the experiment and changes in vigilance state on neural entrainment measures. The learning segment consisted of a long structured (STR) stream where syllables were organized into four three-syllable words presented in random order with no repetition. Following this, six test-blocks were presented, each comprising eight triplets from the words and part-words conditions with 2-s silences interleaved between items. To sustain learning, 30-s short STR streams were interspersed between test blocks. A 4.5-s fade-in/out at the borders of each stream was included to minimize any perceptual anchor effect. The full procedure lasted ~17 min. Arrows’ width schematically represents the transition probability magnitude. (B) Pipeline for longitudinal partial least squares correlation (PLS-c) analysis. Details are provided in the Methods section.
Figure 2 with 3 supplements
Neural entrainment to syllable rate (4 Hz): main effect across all participants (top row); group differences (bottom row).
(A) Design salience (left) and brain salience (right topography) derived from the significant latent component for neural entrainment to the syllable rate (4 Hz) using the targeted frequency (4 Hz) vs. adjacent frequencies as a contrast. Significance (i.e., salience) was established through bootstrapping. Bars represent the mean of 500 random salience samples with replacement bootstrapping, and error bars indicate the 95% confidence interval. Yellow shading highlights variables that significantly contribute to the latent component, defined by a bootstrap ratio (BSR; mean of bootstrapping divided by standard deviation) >2.3. The topography of the BSR values shows electrodes significantly contributing to the latent component (indicated by black dots, BSR >2.3). (B) Individual raw phase locking values (PLVs) extracted from the salient electrodes identified by the latent component (black dots in the topography on (A)) are displayed. A fitted curve is included for visualization purposes only, produced by a mixed-effects model with a 95% confidence interval. This curve is intended solely to aid visualization, as the statistical relationships between EEG and behavioral variables are determined by the PLS-c analysis. (C) PLS-c analysis of the differences between HL and LL groups is presented following the same format as in A. (D) Individual raw PLVs extracted from the salient electrodes (black dots in (C)) are shown for visualization purposes only. At every age, a verbal developmental quotient (DQ) of 100 is expected in the general population.
Figure 2—figure supplement 1
Follow-up analyses restricted to each stream for syllable entrainment.
(A) For RND, the latent component was significant (p < 0.001; r = 0.67; 87.3% explained covariance) with the following BSRs: contrast: 70.3*; mean age: –1.6; contrast*mean age: –0.7; delta age: 3.6*; contrast*delta age: 8.0*; age2: 1.2; contrast*age2: –4.1*. (B) For STR, the latent component was significant (p < 0.001; r = 0.73; 91.1% explained covariance) with the following BSRs: contrast: 91.4*; mean age: –.9; contrast*mean age: 0.7; delta age: 4.2*; contrast*delta age: 4.7*; age2: –2.4; contrast*age2: –3.8*. Yellow shading on left panels and black dots on middle panels indicate BSR >2.3. BSR >2.3 is considered significant.
Figure 2—figure supplement 2
Neural entrainment to syllable rate in sleeping participants (n = 25 recordings).
(A) Design salience (left) and brain salience (right topography) derived from the significant latent component for neural entrainment to the syllable rate using frequency (4 Hz vs. adjacent frequencies) as contrast. Significance (i.e., salience) was established through bootstrapping. (B) Individual raw phase locking values (PLVs) extracted from the salient electrodes identified by the latent component (black dots in the topography on Figure 2A) are displayed. A fitted curve is included for visualization purposes only, produced by a mixed-effects model with a 95% confidence interval.
Figure 2—figure supplement 3
Neural entrainment to syllable rate (4 Hz), excluding the final visit (n = 54 recordings).
(A) Design salience (left) and brain salience (right topography) derived from the significant latent component for neural entrainment to the syllable rate (4 Hz) using group (low vs. high-likelihood for autism) as contrast. Significance (i.e., salience) was established through bootstrapping. Bars represent the mean of 500 random salience samples with replacement bootstrapping, and error bars indicate the 95% confidence interval. Yellow shading highlights variables that significantly contribute to the latent component, defined by a bootstrap ratio (BSR; mean of bootstrapping divided by standard deviation) >2.3. The topography of the BSR values shows electrodes significantly contributing to the component (indicated by black dots, BSR >2.3). (B) Individual raw phase locking values (PLVs) extracted from the salient electrodes identified by the latent component (black dots in the topography on Figure 2A) are displayed. At every age, a verbal developmental quotient (DQ) of 100 is expected in the general population. A fitted curve is included for visualization purposes only, produced by a mixed-effects model with a 95% confidence interval. This curve is intended solely to aid visualization, as the statistical relationships between EEG and behavioral variables are determined by the PLS-c analysis.
Figure 3 with 5 supplements
Neural entrainment to word rate (1.3 Hz): main effect across all participants (top row); group differences (bottom row).
(A) Design salience (left) and brain salience (right) derived (through bootstrapping) from the significant latent component for neural entrainment to word rate. (B) Individual raw phase locking values (PLVs) extracted from the salient electrodes given by the latent component (black dots on A) with a fitted curve, for visualization purposes only. (C) PLS-c applied on PLV at 1.3 Hz with adjacent frequencies subtracted, using group as contrast, and verbal outcome (verbal developmental quotient [DQ] collected at 18–21 months) added as a design variable. (D) Raw individual data extracted from the salient electrodes of the latent component (black dots on Figure 2C) with a linear regression curve fitted for illustration purposes only.
Figure 3—figure supplement 1
Neural entrainment to word rate in sleeping participants (n = 25 recordings).
(A) Design salience (left) and brain salience (right topography) derived from the significant latent component for neural entrainment to the word rate using frequency (1.3 Hz vs. adjacent frequencies) as contrast. Significance (i.e., salience) was established through bootstrapping. (B) Individual raw phase locking values (PLVs) extracted from the salient electrodes identified by the latent component (black dots in the topography on A) are displayed. A fitted curve is included for visualization purposes only, produced by a mixed-effects model with a 95% confidence interval.
Figure 3—figure supplement 2
Word entrainment within HL participants (n = 25; 44 recordings).
(A) Design salience (left) and brain salience (right topography) derived from the significant latent component for neural entrainment to the word rate using frequency (1.3 Hz vs. adjacent frequencies) as contrast. Significance (i.e., salience) was established through bootstrapping. BSRs: contrast: 12.9*; mean age: –3.2*; contrast*mean age: –5.3*; delta age: 5.9*; contrast*delta age: 6.7*; age2: 3.6*; contrast*age2: 1.9. (B) Individual raw phase locking values (PLVs) extracted from the salient electrodes identified by the latent component (black dots in the topography on A) are displayed. A fitted curve is included for visualization purposes only, produced by a mixed-effects model with a 95% confidence interval.
Figure 3—figure supplement 3
Neural entrainment time course over the experimental session considering all participants.
The plain squares under the plots correspond to the time samples with phase locking values (PLVs) significantly greater than 0 (p < 0.05).
Figure 3—figure supplement 4
Group differences in the time course of syllable neural entrainment (4 Hz).
Gray shading on right panel indicates BSR <2.3; BSR >2.3 is considered significant. Vertical dashed lines indicate the transitions between random and structured streams.
Figure 3—figure supplement 5
Group differences in syllable entrainment within RND (A, B) and STR (C, D).
Figure 4 with 1 supplement
Early evoked response potential (ERP) to part-words compared to words across all participants.
(A) Design and brain saliences derived from the significant latent component. Brain topographies of bootstrap ratios (BSRs) are displayed at 250-ms intervals. Black dots indicate BSR >2.3. (B) Participants’ brain scores for part-word and word conditions, as a function of age. Brain scores are participants’ raw voltage data projected onto electrode saliencies. Brain scores illustrate how individual EEG data fit the saliences derived from the latent component. Linear fitting is used for illustrative purposes only. (C) Voltage grand averaged (left) and differential (right) responses to part-word and word conditions at each age bin: 3 months (n = 18), 6–9 months (n = 20), 12–15 months (n = 20), and 18–21 months (n = 25).
Figure 4—figure supplement 1
Evoked response potential (ERP) topographies across age bins and participants.
Figure 5
Late evoked response potential (ERP) to part-words compared to words across all participants.
(A) Design and brain saliences derived from the significant latent component. Brain topographies of bootstrap ratios (BSRs) are displayed at 250-ms intervals. Black dots indicate BSR >2.3. (B) Participants’ brain scores for part-word and word conditions, as a function of age. For details, see Figure 4. (C) Voltage averaged (left) and differential (right) responses to part-word and word conditions at each age bin: 3 months (n = 18), 6–9 months (n = 20), 12–15 months (n = 20), and 18–21 months (n = 25).
Figure 6 with 5 supplements
Late evoked response potential (ERP) to novelty in infants at high and low likelihood for autism.
(A) Design and brain saliences derived from the significant latent component. Brain topographies of bootstrap ratios (BSRs) are displayed at 250-ms intervals. Black dots indicate BSR >2.3. (B) Participants’ brain scores for part-word and word conditions, as a function of age (left) and of verbal outcome (right). Detailed information about brain scores is provided in Figure 4. Linear fitting is used for illustrative purposes only. (C) Voltage differential responses to part-word and word conditions at each age bin and within each group: 3 months (11 LL and 7 HL), 6–9 months (9 LL and 11 HL), 12–15 months (10 LL and 10 HL), and 18–21 months (9 LL and 16 HL). DQ: developmental quotient; HL: high likelihood for autism; LL: low likelihood for autism.
Figure 6—figure supplement 1
Evoked response potential (ERP) topographies in each group at 3 and 6–9 months.
Figure 6—figure supplement 2
Evoked response potential (ERP) topographies in each group at 12–15 and 18–21 months.
Figure 6—figure supplement 3
Late evoked response potential (ERP) to word novelty, excluding the final visit (n = 54 recordings).
(A) Design and brain saliences derived from the significant latent component. Brain topographies of bootstrap ratios (BSRs) are displayed at 250-ms intervals. Black dots indicate BSR >2.3. (B) Participants’ brain scores for part-word and word conditions, as a function of verbal outcome. Brain scores are participants’ raw voltage data projected onto electrode saliencies. Brain scores illustrate how individual EEG data fit the saliences derived from the latent component. Linear fitting is used for illustrative purposes only. HL: high likelihood for autism; LL: low likelihood for autism.
Figure 6—figure supplement 4
Late evoked response potential (ERP) to word novelty within low likelihood infants (n = 19; 39 recordings).
(A) Design and brain saliences derived from the significant latent component. Brain topographies of bootstrap ratios (BSRs) are displayed at 250-ms intervals. Black dots indicate BSR >2.3. (B) Participants’ brain scores for part-word and word conditions, as a function of age. Brain scores are participants’ raw voltage data projected onto electrode saliencies. Brain scores illustrate how individual EEG data fit the saliences derived from the latent component. Linear fitting is used for illustrative purposes only. LL: low likelihood for autism.
Figure 6—figure supplement 5
Late evoked response potential (ERP) to word novelty within high likelihood infants (n = 25; 44 recordings).
(A) Design and brain saliences derived from the significant latent component. Brain topographies of bootstrap ratios (BSRs) are displayed at 250-ms intervals. Black dots indicate BSR >2.3. (B) Participants’ brain scores for part-word and word conditions, as a function of age. Brain scores are participants’ raw voltage data projected onto electrode saliencies. Brain scores illustrate how individual EEG data fit the saliences derived from the latent component. Linear fitting is used for illustrative purposes only. HL: high likelihood for autism.
Author response image 1
Late evoked response potential (ERP) to word novelty in awake participants.
A. Design and brain saliences derived from the significant latent component. Brain topographies of bootstrap ratios (BSR) are displayed at 250ms intervals. Black dots indicate BSR > 2.3. B. Participants’ brain scores for part-word and word conditions, as a function of age (left panel) and verbal DQ (right panel). For details on brain scores, see Figure 6 -figure supplement 1. Linear fitting is used for illustrative purposes only. HL: high likelihood for autism; LL: low likelihood for autism.
Author response image 2
Syllable entrainment within RND (A-B) and STR (C-D).
Tables
Table 1
Sample characteristics.
Statistical comparison between LL and HL samples. For categorical variables, chi-square (χ2) was applied. For continuous variables, we used two-tailed independent T-tests. p-values <0.05 are highlighted in bold.
| Measure [mean (SD)] | Low-likelihood | High-likelihood | p-value |
|---|---|---|---|
| Number of participants (n = 44) | 19 | 25 | |
| Number of EEG recordings (n = 83) | 39 | 44 | |
| 3 months EEG age (n = 18) | 3.3 ± 0.6 (n = 11) | 3.5 ± 0.3 (n = 7) | 0.309 |
| 6–9 months EEG age (n = 20) | 6.6 ± 0.8 (n = 9) | 7.1 ± 1.3 (n = 11) | 0.320 |
| 12–15 months EEG age (n = 20) | 13.0 ± 1.0 (n = 10) | 13.5 ± 1.4 (n = 10) | 0.371 |
| 18–21 months EEG age (n = 25) | 19.0 ± 1.1 (n = 9) | 18.7 ± 1.6 (n = 16) | 0.438 |
| Female biological sex | 4 (21.1%) | 12 (48.0%) | 0.066 (χ2) |
| Age at verbal outcome | 19.6 ± 2.4 (n = 16) | 19.7 ± 2.0 (n = 24) | 0.886 |
| Verbal outcome [DQ] | 104.3 ± 16.2 | 77.9 ± 22.2 | 0.001 |
Additional files
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)
Infants at high and low likelihood for autism show different EEG developmental trajectories in speech tracking and statistical learning
eLife 14:RP109901.
https://doi.org/10.7554/eLife.109901.3
