Learning from errors: neural evidence for altered mechanisms in Autism Spectrum Disorder

376 

Abstract Submission 

Maria Camila Dias^1,2, Teresa Sousa^1,2, Susana Mouga^1,3, Miguel Castelo-Branco^1,2,4

¹CIBIT - Coimbra Institute for Biomedical Imaging and Translational Research, University of Coimbra, Coimbra, Portugal, ²ICNAS - Institute for Nuclear Sciences Applied to Health, University of Coimbra, Coimbra, Portugal, ³ICNAS PHARMA Unipessoal, Lda, Ed. ICNAS, Pólo das Ciências da Saúde, University of Coimbra, Coimbra, Portugal, ⁴FMUC - Faculty of Medicine, University of Coimbra, Coimbra, Portugal

Maria Camila Dias  
CIBIT - Coimbra Institute for Biomedical Imaging and Translational Research, University of Coimbra|ICNAS - Institute for Nuclear Sciences Applied to Health, University of Coimbra
Coimbra, Portugal|Coimbra, Portugal

Teresa Sousa  
CIBIT - Coimbra Institute for Biomedical Imaging and Translational Research, University of Coimbra|ICNAS - Institute for Nuclear Sciences Applied to Health, University of Coimbra
Coimbra, Portugal|Coimbra, Portugal

Susana Mouga  
CIBIT - Coimbra Institute for Biomedical Imaging and Translational Research, University of Coimbra|ICNAS PHARMA Unipessoal, Lda, Ed. ICNAS, Pólo das Ciências da Saúde, University of Coimbra
Coimbra, Portugal|Coimbra, Portugal

Miguel Castelo-Branco  
CIBIT - Coimbra Institute for Biomedical Imaging and Translational Research, University of Coimbra|ICNAS - Institute for Nuclear Sciences Applied to Health, University of Coimbra|FMUC - Faculty of Medicine, University of Coimbra
Coimbra, Portugal|Coimbra, Portugal|Coimbra, Portugal

Error monitoring involves detecting, processing, and signaling errors to prevent future mistakes. Perceiving the outcomes of our actions is vital for regulating behavior and learning (Ullsperger, Danielmeier, & Jocham, 2014). Accordingly, impaired error monitoring skills have been demonstrated in some conditions, such as Autism Spectrum Disorder (ASD) (Kim, Grammer, Benrey, Morrison, & Lord, 2018; Santesso et al., 2011). However, the alterations in the error monitoring neural circuitry in ASD remain to be fully understood. Here, we tested the hypothesis that the evolution of learning associated with error monitoring is altered in ASD.

In this study, 15 non-ASD (mean age 30.7 ± 6.1 years) and 15 ASD (mean age 29.1 ± 8.7 years) male participants performed a challenging 3T functional magnetic resonance imaging (fMRI) task based on Estiveira et al. (2022). The task had multiple concurrent cues, namely emotional facial expression and gaze cues, that signaled the appropriate action and needed to be integrated to make a correct response. To understand how neural learning processes progress in error monitoring, we conducted linear mixed-effects analyses to assess the impact of group, performance, learning, and their respective interactions on key error monitoring regions, such as the dorsal anterior cingulate cortex (dACC) and anterior insula (AI) (Dali et al., 2023; Neta et al., 2015). We also evaluated the influence of these variables on the putamen, a region linked to trial and error learning (Ashby, Turner, & Horvitz, 2010; Patterson & Knowlton, 2018). We controlled our analyses for age and Full-Scale Intelligence Quotient by including them as covariates in the linear mixed-effects models.

Behaviorally, we found a significant effect of group (F(1, 29.02) = 20.59, p = 9.10 × 10-5) and learning stage (F(6,764.19) = 34.94, p = 2.05 × 10-37) on error rate. Both groups showed clear learning curves with decreasing error rates from the beginning to the end of the session. During the entire task, the ASD group had increased error rates compared to the non-ASD group.
The fMRI results revealed that activity in the dACC and AI was modulated by the interaction between group and performance (dACC: F(2,53.23) = 10.80, p = 1.16 × 10-4; AI: F(2,54.21) = 25.48, p = 1.58 × 10-8), and learning and performance (dACC: F(12,7809.34) = 5.38, p = 3.55 × 10-9; AI: F(6,7810.24) = 7.83, p = 2.03 × 10-8). The activity in these regions was similar for correct and erroneous responses in an initial learning stage but, as learning progressed, the differences became evident. This happened due to a simultaneous decrease in activity in correct responses and an increase following errors. Although this pattern did not differ between groups, the differences between correct responses and errors were attenuated in ASD (differences being only significant in the non-ASD group). Moreover, for the ASD group, we found an inverse correlation between the difference in dACC activity between correct and erroneous responses and autistic traits (r(11) = -0.62, p = 0.023) measured by Autism Diagnostic Observation Schedule (ADOS) total scores (Lord et al., 1989).
The putamen response was influenced by the interaction between group and learning (F(6, 7808.01) = 6.84, p = 2.99 × 10-7), and learning and performance (F(6, 7810.18) = 4.84, p = 6.10 × 10-5). In the non-ASD group, its activity increased both in correct responses and errors with learning. Nonetheless, in the ASD group, it only increased in response to errors: the activity associated with correct actions was approximately constant throughout learning.

These findings suggest that, in ASD, error monitoring mechanisms are impaired, and the learning process is altered, possibly leading to higher error rates when integrating social cues.

Emotional Perception

Emotion and Motivation Other ¹

Learning and Memory Other

BOLD fMRI

Perception: Visual ²

Addictions

Cognition

Emotions

FUNCTIONAL MRI

Learning

Limbic Systems

Perception

Other - visual perceptions