Inflexible neurobiological signatures precede atypical development in infants at high risk for autism

Variability in neurobiological signatures is ubiquitous in early life but the link to adverse developmental milestones in humans is unknown. We examined how levels of signal and noise in movement signatures during the 1st year of life constrain early development in 71 healthy typically developing infants, either at High or Low familial Risk (HR or LR, respectively) for developing Autism Spectrum Disorders (ASD). Delays in early learning developmental trajectories in HR infants (validated in an analysis of 1,445 infants from representative inf﻿ant-sibling studies) were predicted by worse stochastic patterns in their spontaneous head movements as early as 1–2 months after birth, relative to HR infants who showed more rapid developmental progress, as well as relative to all LR infants. While LR 1–2 mo-old infants’ movements were significantly different during a language listening task compared to during sleep, HR infants’ movements were more similar during both conditions, a striking lack of diversity that reveals context-inflexible experience of ambient information. Contrary to expectation, it is not the level of variability per se that is particularly detrimental in early life. Rather, inflexible sensorimotor systems and/or atypical transition between behavioral states may interfere with the establishment of capacity to extract structure and important cues from sensory input at birth, preceding and contributing to an atypical brain developmental trajectory in toddlerhood.


SUPPLEMENTARY TABLES
Supplementary Table 2. The results of the power curve fits of the form f(x)=a*x^b as well as of the exponential fits of the form f(x) = a*exp(b*x) fitted to the Gamma shape and scale parameters of head movement fluctuations at 9-10 months (raw linear and angular speed), for infants subgrouped by the rapidity of their progress over time on Mullen (MU) Scales of Early Learning trajectories. Note. ELC: Early Learning Composite score. Both equations fit the data well. Power fits yield higher RSquare and lower RMSE (root mean square error), indicating a slightly better fit for these data. Consistent with the results presented in the Main text using T-scores, the b-slope is consistently higher (steeper) for the LR group.

SUPPLEMENTARY FIGURES
Supplementary Figure 1. Sample datasets from two infants in this study, a 1-2 mo-old HR and a 1-2 mo-old LR infant, during a resting-state scan. The top rows plot 6 movement parameter output from SPM, separately for linear (mm) and rotational (radians) values. The raw values are greater in magnitude during the entire scan for the HR infant. The bottom row shows each infant's EPI image in sagittal, coronal, and axial views displayed using MRIcron (http://people.cas.sc.edu/rorden/mricron/index.html). Note the difference in quality between the images of the two infants; the image data of the HR infant (who had greater movement during the scan) reveals visible striped lines (artefacts). (Note. Such "striping" artefacts can stem from subject movement during acquisition of brain slices in an interleaved manner; http://imaging.mrc-cbu.cam.ac.uk/imaging/CommonArtefacts).

Supplementary Figure 2.
Parameter estimates on the Gamma plane for subgroups of infants across all datasets and time points, crosssectionally and longitudinally, for angular and linear speeds. (a) Parameter estimates on the Gamma plane for High and Low risk infants' distributions (total N=93: N=49HR and N=44LR). (b) High and Low risk infants' cross-sectional data shown separately by age subgroups: for 1-2 month-olds (N=28HR and N=28LR) and for 9-10 month olds (N=21HR and N=16LR). The bottom panel presents the same data as the top panels from different age subgroups on the same plot, for angular and linear speeds. (c) High and Low risk infants' longitudinal subset data (N=22) are shown by age subgroups: for 1-2 month-olds (N=11HR and N=11LR) and for 9-10 month olds (N=11HR and N=11LR). The bottom panel presents the same data as the top panels, presenting data from different age subgroups on the same plot for angular and linear speeds. Regardless of the age of the infant, HR infants' parameter estimates are consistently located towards higher noise-to-signal levels (scale, b parameter) and towards the left on the x-axis, away from the more normative, Gaussian shape (shape, a parameter). Error bars denote 95% CIs.
Supplementary Figure 3. Inclusion criteria at the ACE UCLA site included full-term birth. Here we confirm the main finding of our study: HR infants, specifically those with known full-term status, have increased noise-to-signal levels and decreased symmetry relative to full-term LR infants. (a) presents data for 1-2 mo-olds and (b) presents data for 9-10 mo-olds. Note. Data presented includes some subjects who were tested longitudinally. Total N for 1-2 mo-olds, N=24: N=9HR, N=15LR. Total N for 9-10 mo-olds, N=18: N=12HR, N=6LR. Error bars denote 95% CIs.
Supplementary Figure 4. Individual parameter estimates on the Gamma plane for all High and Low risk infants and residual values denoting deviation from the a vs b linear relation. Noise-to-signal levels (scale, b parameter; y-axis) and randomness (shape, a parameter; x-axis) in individual infants' spontaneous head fluctuations on the Gamma plane. Data are presented on log axes to demonstrate linearity. Lower values on the x-axis correspond to higher values on the y-axis. Data are shown for High Risk (red circles) and Low Risk (blue triangles) infants for angular and linear speed. (a) shows the entire sample (N=93 across all time points, including N=22 infants tested longitudinally), (b) shows data only for 1-2 mo-old (N=28HR, N=28LR) infants, and (c) shows data only for 9-10 mo-old (N=21HR, N=16LR) infants. For each a-c, and for angular and linear speeds, Kruskal-Wallis rank-order test on residuals (Whisker plots on the right-hand side) shows significant group differences, with HR infants showing greater deviation from linearity for angular and linear speed relative to LR infants across the entire sample, as well as separately for 1-2 mo-old and 9-10 mo-old age groups (all p<0.001).