The research team, led by Daniel C. Javitt of the Nathan Kline Institute for Psychiatric Research and Columbia University, together with Antigona Martinez, Pejman Sehatpour, Pamela D. Butler, and colleagues, simultaneously recorded eye movements and continuous EEG from 23 adults diagnosed with schizophrenia, 21 autistic adults, and 24 neurotypical controls. Participants viewed intact emotional faces drawn from the Karolinska Directed Emotional Faces collection as well as chimeric faces, in which the top and bottom halves expressed different emotions. Because the top half contains the eye region and the bottom half contains the mouth region, chimeric stimuli allowed the researchers to determine, on a face-by-face basis, whether a person’s responses were driven by eye information or mouth information. Symptoms were rated with the Positive and Negative Symptom Scale in schizophrenia and the Autism Diagnostic Observation Schedule, Second Edition, in autism.

As predicted, both clinical groups showed similar levels of reduced face emotion recognition accuracy relative to neurotypical controls, with large between-group effect sizes on both the novel stimulus set and the standardized Penn Emotion Recognition Task. Eye tracking revealed that both schizophrenia and autism were associated with fewer fixations to the eye region and more fixations to the mouth region. Yet when the researchers statistically controlled for fixation location, the two groups diverged sharply. In schizophrenia, the reduced tendency to endorse the emotion of the top half of chimeric faces, and the increased tendency to endorse the bottom half, persisted even after accounting for where participants actually looked. In autism, these differences disappeared once fixation location was considered, indicating that gaze patterns themselves accounted for the behavioral differences in that group.

The researchers quantified this eye-versus-mouth weighting with a congruency index, defined as the percentage of responses matching the top half of each chimeric stimulus relative to the percentage matching the bottom half. This index differed significantly across groups with a large effect size and correlated with face emotion recognition performance to intact faces across the entire sample, remaining significant after control for group status. Notably, for intact faces, the ratio of eye to mouth fixations did not influence performance in any group, consistent with the idea that face emotion processing operates holistically rather than by parts. The results argue strongly against gaze patterns as the primary drivers of impaired emotion recognition in either condition, a conclusion that contradicts a widely held assumption in the field.

Traditional time-domain EEG analyses, which focus on sequential P1 and N170 event-related potentials, revealed only modest between-group differences. Autistic participants showed the expected prolongation of N170 latency, but this measure did not correlate with emotion recognition accuracy or clinical symptoms. The real mechanistic insights emerged from time-frequency analyses, which decompose the EEG signal into its constituent spectral frequencies and preserve information that averaging discards. These analyses separated inter-trial coherence, reflecting the consistency of neural responses across trials, from single-trial power, which includes both phase-locked and non-phase-locked components.

In schizophrenia, reduced face emotion recognition accuracy correlated with reduced inter-trial coherence and increased total power of the initial sensory response in the theta band, reflecting diminished engagement of primary visual regions. The researchers interpret this pattern as consistent with magnocellular visual hypoactivity, a failure of the subcortical retinogeniculate pathway that normally provides rapid, low-spatial-frequency input to primary visual cortex. In autism, by contrast, reduced accuracy correlated with increased initial inter-trial coherence and increased alpha-frequency event-related desynchronization, a signature of hyper-engagement of secondary visual regions. Beamformer source localization mapped the increased desynchronization in autism to early extrastriate cortex consistent with area V2, a region that normally processes complex shapes rather than faces. The increased V2 response in autism correlated with both reduced emotion recognition accuracy and ADOS Social Interaction domain scores, linking the physiological measure directly to real-world social difficulties.

The mechanistic interpretation draws on the two major subcortical visual pathways. Visual information reaches cortex via the retinogeniculate system, which engages the lateral geniculate nucleus and is divided into magnocellular and parvocellular divisions, and via the retinotectal system, which engages the pulvinar nucleus of the thalamus and contributes to rapid transmission of threat-related information to the amygdala. The authors propose that in autism, a loss of the normal developmental attenuation of the retinotectal pathway leads to hyperactivity of the pulvinar and, through alpha-band oscillatory synchrony, hyper-engagement of V2 by face stimuli. This non-face-specific hyperactivation may interfere with more specialized processing in regions dedicated to social information extraction, such as the posterior superior temporal sulcus or the recently described third visual pathway for social perception.

Perhaps the most striking result is the categorical discrimination between the two conditions. A discriminant function combining eye-tracking and time-frequency measures differentiated schizophrenia from autism with greater than 90 percent accuracy, and this classification held up in leave-one-out cross-validation at 84.1 percent. The critical measures included the inter-trial coherence and power of the initial theta-frequency response, N170 latency, late event-related desynchronization amplitude, and the congruency index. These findings suggest that physiological biomarkers could serve as objective tools for patient stratification, potentially guiding diagnosis-specific interventions such as non-invasive brain stimulation or targeted cognitive remediation.

The study carries important implications for how clinicians and researchers conceptualize the relationship between schizophrenia and autism. Rather than occupying a single continuum in which divergence in either direction from the neurotypical pattern produces the same deficit, the two conditions appear to involve different subcortical visual systems altogether. Schizophrenia reflects hypoactivity of the magnocellular retinogeniculate pathway, while autism reflects hyperactivity of the retinotectal pathway leading to pulvinar-driven hyper-engagement of non-face regions by face stimuli. The authors suggest that these alterations may serve as targets for future intervention studies related to social cognition, and that the increased V2 response in autism, if confirmed during formal qualification studies, may prove useful as a biomarker for etiological and interventional research.

Limitations remain. All participants with schizophrenia were receiving stable doses of antipsychotic medication, whereas autistic participants were not, so medication effects cannot be entirely excluded, although no correlations were observed with antipsychotic dose. The sample size was relatively small, allowing detection only of statistically large effects, and the spatial resolution of the beamformer method may not fully exclude involvement of other early visual areas such as V3 or V4. Future studies with larger samples and more stimuli would be needed to determine how well these findings generalize and to create separate averages by fixation location. Nevertheless, the convergence of eye-tracking, time-frequency EEG, and source localization evidence provides a compelling case that two conditions once considered to share a single social-cognitive deficit are, at the level of the early visual system, truly opposite sides of different coins.

Subject of Research: Neuroscience

Article Title: Opposite sides of different coins: near-diametrical opposition of physiological indices of reduced accuracy of face emotion recognition in schizophrenia and autism spectrum disorders

Article References: Javitt, D. C., Martinez, A., Sehatpour, P., Butler, P. D., Dias, E., Micceri, K., Breland, M., & Tobe, R. H. (2026). Opposite sides of different coins: near-diametrical opposition of physiological indices of reduced accuracy of face emotion recognition in schizophrenia and autism spectrum disorders. Frontiers in Neuroimaging, 5, Article 1771087. https://doi.org/10.3389/fnimg.2026.1771087

Image Credits: AI Generated

DOI: 10.3389/fnimg.2026.1771087

Keywords: schizophrenia, autism spectrum disorder, face emotion recognition, EEG, eye tracking, time-frequency analysis, visual neuroscience, social cognition, biomarkers, Opposite, sides, different