Weaker neural suppression in autism

The study investigates whether atypical sensory experiences in autism spectrum disorder (ASD), particularly in visual motion perception, arise from differences in neural suppression within visual cortex. Prior behavioral studies reported mixed findings in ASD: superior motion discrimination for large, high-contrast stimuli in some studies, and higher thresholds in others. The authors hypothesize that spatial (surround) suppression—an established property of visual cortex—may be weaker in ASD and aim to test this using convergent behavioral psychophysics, fMRI in motion-selective human MT complex (hMT+), MR spectroscopy (MRS) of inhibitory and excitatory metabolites, and computational modeling within a divisive normalization framework that includes top-down gain modulation. The purpose is to clarify the neural basis of sensory differences in ASD and reconcile divergent findings via modeling.

• Sensory anomalies are common in ASD, with reports of both enhanced and diminished motion perception. Prior work (Foss-Feig et al.) showed enhanced motion discrimination for large stimuli in ASD; other studies (Schauder et al.; Sysoeva et al.) reported higher thresholds or abnormal size-dependent modulation.
• Spatial suppression in motion perception is linked to center–surround antagonism and divisive normalization in visual cortex (e.g., MT/V5), supported by psychophysics, fMRI, and neurophysiology.
• Computational accounts include: (1) weaker divisive normalization in ASD predicting lower thresholds across tasks, and (2) larger excitatory spatial filters explaining higher thresholds.
• Prior neurochemical hypotheses suggest excitation–inhibition imbalance in ASD, but MRS studies of visual cortex GABA+ have shown mixed or normal levels.
• Top-down processes such as spatial attention modulate visual responses via normalization; evidence suggests a sharper spatial attention gradient (narrower focus) in ASD, potentially influencing surround suppression.

Participants: 28 young adults with ASD (18 male) and 35 neurotypical (NT) controls (21 male), ages 18–30, matched on age, sex, handedness, and non-verbal IQ (WASI). ASD diagnoses confirmed via ADOS-2, ADI-R, and clinical judgment (DSM-5). Inclusion criteria included normal/corrected vision, no neurological disease, seizure history, or recent substance use; additional MRI safety criteria. One ASD and two NT participants excluded for behavioral catch trial performance; one ASD excluded from MRI/MRS for excessive motion. Final samples varied by analysis due to ROI criteria.

Stimuli and display: Sinusoidal gratings at low (3%) and high (98%) contrast. Psychophysics on a 120 Hz CRT; fMRI projection at 60 Hz. Viewing distance 66 cm; luminance linearized.

Psychophysics (spatial suppression task): Direction discrimination (left/right) of vertically oriented drifting gratings (4 cpd/s) within Gaussian-windowed apertures (sizes: 0.84°, 1.7°, 10°; spatial frequency 1.2 cpd). Adaptive Psi staircases targeted 80% accuracy to estimate minimum duration thresholds (range 6.7–333 ms). Six interleaved staircases (3 sizes × 2 contrasts) per run; 10 catch trials (large, high-contrast, 333 ms) per run; four runs (~30 min). Size index computed as log10(medium) − log10(big); more negative indicates stronger suppression.

fMRI paradigm: Alternating 10 s blocks of smaller (2°) and larger (12°) foveal drifting gratings (16 stimuli/block, each 400 ms, ISI 225 ms), at 3% or 98% contrast in separate runs; no rest blocks. Fixation task: color-shape conjunction detection (respond to green circle) to minimize eye movements and divert attention from gratings.

MRI acquisition: Philips 3T. T1 anatomical (1 mm isotropic). BOLD fMRI: 3 mm isotropic voxels, 30 slices (0.5 mm gap), TR 2 s, TE 25 ms, flip 79°, AP phase encoding; reversed PE scan (PA) for distortion compensation.

Functional localizers: (1) hMT+ localizer: blocks of drifting vs static gratings (2° dia., 15% contrast), to define motion-selective voxels; (2) center vs surround retinotopic localizer: phase-reversing checkerboards (8 Hz) in central 2° vs annulus (2–12°) to select foveal-biased voxels. ROIs defined bilaterally as the intersection (for hMT+) of motion > static and center > surround; EVC ROIs near occipital pole (V1/V2/V3 foveal confluence) from center > surround. Top 20 voxels per hemisphere (relaxed threshold if needed), minimum one-tailed p < 0.006; center-selective hMT+ required (participants without were excluded from hMT+ analyses).

fMRI preprocessing/analysis: Brain Voyager: motion correction, distortion compensation, high-pass filtering (>2 cycles/run), aligned to in-session anatomy; no spatial smoothing or template normalization. Event-related epochs extracted from −4 to +12 s around transitions from smaller to larger blocks; baseline: mean signal 0 to 4 s before transition; responses converted to % signal change. Suppression magnitude: average signal 8–12 s post transition. Primary regions: foveal hMT+ and foveal EVC.

MR spectroscopy: 1H MEGA-PRESS (3 cm isotropic voxel; 320 averages; TR 2 s; TE 68 ms; 2048 points; 2 kHz width). Editing pulses at 1.9 ppm (on) and 7.5 ppm (off), 16-step phase cycle; ~11 min per run. Voxels in bilateral hMT+ (runs left then right) guided by in-session hMT+ fMRI localizer (moving > static, t ≥ 3.0), and in EVC (anatomically placed). Participants watched a movie to improve compliance. GABA+ and Glx quantified in institutional units relative to water; GABA+ tissue-corrected for GM/WM/CSF composition; Glx explored via control analyses. Data processed with Gannet 2.0 (frequency/phase correction, artifact rejection, 3 Hz line broadening; Gaussian fits at 3 ppm for GABA+ and 3.75 ppm for Glx).

Computational modeling: Divisive normalization model (Reynolds & Heeger) adapted to predict motion duration thresholds: R = (E × M) / (S × S_s + σ). E: excitatory drive (Gaussian spatial selectivity, width x_we). M: top-down gain field (Gaussian width). S: suppressive drive (broader than E; normalization pool). S_s: suppressive gain (normalization strength). σ: semi-saturation. Duration thresholds predicted as T = C / R_peak. Three model variants compared: (1) weaker normalization (25% reduction in S_s), (2) larger excitatory spatial filters (x_we +25%), (3) narrower top-down gain (width 6 vs 14 arbitrary units). Models qualitatively matched against observed threshold patterns and size indices.

Clinical measures and covariates: ADOS-2 total comparison scores; Sensory Profile sensitivity and avoiding subscales (summed). Control analyses included age, sex, IQ as covariates (linear mixed-effects), eye movement metrics, and MRS data quality comparisons.

Statistics: Mixed repeated-measures ANOVAs with participants as random effects, stimulus size (continuous) and contrast as within-subject factors; Pearson r with permutation-based p-values; Bonferroni correction where applicable. Predefined exclusion criteria for psychophysics and ROIs.

• Psychophysics: Participants with ASD showed weaker spatial suppression during motion discrimination. Size indices (log10(medium) − log10(big); more negative = stronger suppression) were less negative in ASD vs NT (main effect of group: F1,61 = 9.37, p = 0.003). Suppression was stronger at low contrast (main effect of contrast: F1,61 = 19.7, p = 4×10^-5). Group × contrast interaction was not significant (F1,61 = 1.55, p = 0.2). ASD participants achieved lower duration thresholds particularly for large stimuli, indicating weaker suppression.
• fMRI hMT+: Robust suppression of fMRI responses to larger vs smaller stimuli in foveal hMT+ for both contrasts (paired t > 4.13, p < 0.003, Bonferroni-corrected). Suppression was significantly weaker in ASD vs NT (main effect of group: F1,47 = 5.66, p = 0.022). Suppression stronger at high vs low contrast (F1,47 = 5.52, p = 0.023); no group × contrast interaction (F1,47 = 1.15, p = 0.3). No correlation between hMT+ fMRI suppression and psychophysical suppression across individuals (r = 0.02, p = 0.9).
• fMRI EVC: Significant suppression for larger vs smaller stimuli in foveal EVC at both contrasts (paired t > 9.04, p < 4×10^-11). No group difference in suppression (F1,55 = 1.61, p = 0.2). Suppression stronger at high vs low contrast (F1,55 ≈ 56.5, p = 7×10^-10); no group × contrast interaction (F1,53 = 1.50, p = 0.2).
• MRS: No significant group differences in GABA+ or Glx concentrations in hMT+ or EVC. No significant correlations of metabolite levels with behavioral or fMRI suppression metrics.
• Clinical associations: No significant correlations between ADOS-2 total comparison scores and suppression metrics (|r| < 0.18, p > 0.3). A trend-level association between higher Sensory Profile sensitivity+avoiding scores and weaker hMT+ fMRI suppression across groups (r = 0.34; uncorrected p = 0.019; Bonferroni-corrected p = 0.074); no relationship with behavioral size indices (r = 0.004, p = 0.98).
• Computational modeling: Weaker normalization (−25% S_s) reduced thresholds globally without altering size indices, failing to explain weaker spatial suppression. Larger excitatory spatial filters predicted higher thresholds for small/medium stimuli and less negative size indices, inconsistent with observed ASD pattern. Narrower top-down gain (width 6 vs 14) predicted lower thresholds especially for large stimuli and less negative size indices, closely matching ASD results. Varying top-down width could also account for divergent prior findings (both superior and reduced performance) depending on decision sampling region size.
• Controls: Group differences in suppression persisted after covarying age, sex, and IQ. No group differences in eye movements; no correlations between eye metrics and suppression. MRS data quality metrics were comparable across groups.

The study demonstrates that people with ASD exhibit weaker spatial (surround) suppression in visual motion perception, reflected behaviorally as lower motion duration thresholds for large stimuli and neurally as reduced suppression of fMRI responses in foveal hMT+. Early visual cortex (EVC) showed suppression overall but no group differences, suggesting that differences in ASD may be more prominent in later, motion-selective areas where top-down influences are stronger. The absence of group differences in MRS measures of GABA+ and Glx in visual cortex suggests that gross differences in inhibitory or excitatory metabolite levels do not account for the suppression differences observed. A divisive normalization model incorporating narrower top-down gain (consistent with a narrower spatial attention field) accounts for the behavioral and hMT+ fMRI findings and can reconcile previous contradictory behavioral results in ASD by considering the relative width of top-down modulation and the neural sampling region used for perceptual decisions. This points to a higher-level mechanism (top-down modulation of sensory processing) rather than solely lower-level changes in normalization strength or receptive field size. The findings advance understanding of the neural basis of sensory processing differences in ASD and suggest that altered top-down spatial modulation may contribute to atypical sensory experiences, with potential links to self-reported sensory sensitivity.

Main contributions: (1) Convergent evidence from psychophysics and fMRI indicates weaker spatial suppression in ASD, particularly in hMT+. (2) MRS in visual cortex revealed no group differences in GABA+ or Glx, arguing against a simple excitation–inhibition imbalance as measured by these metabolites. (3) A normalization-based computational model with narrower top-down gain provides a unified account of the observed weaker suppression in ASD and can reconcile prior divergent behavioral findings. Future directions: Directly manipulate and measure spatial attention to test the top-down gain hypothesis; extend to other visual specializations (e.g., faces) and other sensory modalities to assess generality of weaker suppression; investigate developmental and clinical heterogeneity (including lower-functioning ASD) and relationships with sensory symptoms; refine neurochemical assessments (e.g., higher-field MRS, macromolecule-suppressed GABA, dynamic measures) and circuit-level mechanisms of feedback modulation.

• Participant sample comprised high-functioning young adults with ASD, potentially limiting generalizability to broader ASD populations.
• Spatial attention/top-down modulation was inferred via modeling rather than directly manipulated or measured.
• MRS measures (GABA+ including macromolecules; Glx) are spatially coarse and may mask subtle neurotransmitter differences; static measures may not reflect dynamic inhibition.
• Behavioral and fMRI were conducted in separate sessions with slightly different stimuli and attentional demands, which may contribute to the lack of individual-level correlations.
• EVC localizer stimuli may have been suboptimal for hMT+ foveal selection, leading to exclusions; ROI definitions and alternating-block design precluded independent size-specific response estimates.