The role of autism and alexithymia traits in behavioral and neural indicators of self-concept and self-esteem in adolescence

Alexithymia is a sub-clinical condition characterized by difficulties in recognizing and describing one’s own emotions (Sifneos, 1973). Whereas prevalence in the general population is around 10%–15%, alexithymia co-occurs in 50%–55% of people with ASC (Hill et al., 2004; Kinnaird et al., 2019; Milosavljevic et al., 2016). The alexithymia hypothesis (Cook et al., 2013) suggests that emotion-related difficulties in individuals with autism originate from alexithymia rather than representing a core feature of autism. This raises the question whether alexithymia rather than autism traits may play a role in self-concept positivity and self-esteem of autistic adolescents.

In addition, alexithymia traits have been related to decreased mentalizing and perspective-taking abilities (Moriguchi et al., 2006), which was also reflected in neuroimaging studies showing that alexithymia is related to reduced brain activation during empathizing (insula), mentalizing (mPFC), and emotion processing (precuneus) (Bird et al., 2010; Moriguchi et al., 2006; van der Velde et al., 2013). In addition, a structural MRI study in autistic adults has related alexithymia to reduced covariance in social-emotional (frontal-insular) but not social-cognitive (dorsal mPFC, TPJ) networks (Bernhardt et al., 2014).

The potential role of alexithymia in behavioral and neural measures of reflected self-evaluation has not yet been examined.

The goal of this study was to provide a comprehensive investigation of self-concept in autistic adolescents by examining (1) self-concept across domains and (2) direct versus reflected self-concept, and using behavioral and neural measurements. Participants evaluated their traits in the academic, physical appearance, and prosocial domain from their own perspective and the reflected perspective of their peers, while undergoing MRI scans. Participants were adolescent males with a clinically established autism diagnosis and typically developing adolescent males aged 12–16 years. Since autistic traits are represented on a spectrum between individuals, we tested for group differences as well as for relationships with autism traits across all participants.

We expected (1a) that self-concept positivity, especially in the prosocial domain, and self-esteem would be lower in autistic compared to non-autistic adolescents, and would be negatively related to the number of autism traits (Bauminger et al., 2004; McCauley et al., 2019; van der Cruijsen & Boyer, 2021; Williamson et al., 2008). Regarding reflected versus direct self-concept, we expected (1b) higher similarity in non-autistic compared to autistic adolescents, or in participants with fewer autism traits (Pfeifer et al., 2009; Pfeifer & Peake, 2012). As alexithymia may explain emotion-related problems in autistic individuals (Cook et al., 2013), and it has been found to be related to decreased perspective-taking skills (Moriguchi et al., 2006), we exploratively tested (1c) whether alexithymia would explain lowered self-concept, lower self-esteem, or larger differences between direct and reflected self-concepts above autism traits.

On the neural level, we expected (2a) self-related mPFC activation in adolescents with and without autism (Burrows et al., 2016; Cygan et al., 2018). Previous studies were conflicted regarding potential differences in this activation between adolescents with and without autism. Therefore, here we tested exploratively for group differences and relationships with autism traits across both groups. Next, we expected (2b) that TPJ activation for reflected self-evaluations, and differentiation in TPJ activation for reflected versus direct self-evaluations would be stronger in non-autistic adolescents compared to adolescents with autism, or in participants with fewer autism traits (Kana et al., 2015; Kennedy & Courchesne, 2008; Lombardo et al., 2011; Murdaugh et al., 2014). Potentially, given the difficulties with social skills adolescents with autism often face, differences in neural activation between autistic and non-autistic adolescents mainly become apparent in the prosocial domain. Last, as alexithymia traits have been related to reduced neural activation for affective and mentalizing processes (Bird et al., 2010; Moriguchi et al., 2006; van der Velde et al., 2013), we tested (2c) whether alexithymia explained lowered mPFC and TPJ activation above autism traits.

Participants were 40 adolescent autistic males and 37 non-autistic peers aged between 12.1 and 16.9 years. In total, five participants with and three participants without autism were excluded due to not completing the MRI scans (N_Autism = 1), or excessive head movements during the MRI scans (N_Autism = 4, N_Non-Autism = 3), resulting in a final sample of 35 adolescent males with autism and 34 non-autistic peers (see Table 1). IQ was estimated using the two subtests “vocabulary” and “block design” of the Dutch Wechsler Intelligence Scale for Children (WISC-III-NL; Kort et al., 2005), which are known to correlate strongly to full-scale IQ (M = 100, SD = 15; Sattler, 2001). Estimated IQ scores ranged from 80 to 135 and did not differ between groups (t(67) = 1.18, p = 0.241). See Table 1 for information on ethnicity and gross annual income.

Participants with ASC were recruited by sending an email to parents of boys aged 12 to 16 years, who were registered at the Netherlands Autism Register (NAR; https://nar.vu.nl/english/what-is-the-nar). A clinical Diagnostic and Statistical Manual of Mental Disorders (4th ed., text rev.; DSM-IV-TR; White, 2012) or Diagnostic and Statistical Manual of Mental Disorders (5th ed., DSM-5) diagnosis of autism spectrum disorder (ASD; American Psychiatric Association, 2013) was previously, independent of this study, determined in all participants by a psychiatrist or certified psychologist. Twenty autistic adolescents had co-occurring diagnoses, of whom six adolescents had two. Comorbid diagnoses were dyslexia (9), ADHD (8), post traumatic stress disorder (PTSD) (2), sensory integration disorder (2), Gilles de la Tourette (1), anxiety disorder (1), behavioral disorder not otherwise specified (1), and nonverbal learning disability (1). Fifteen adolescents used medication. Twelve participants used methylphenidate, of which one additionally used citalopram and four additionally used antipsychotics (aripiprazole, risperidone (2x), pipamperone). In addition, one participant only used dexamphetamine, one only used atomoxetine, and one only used aripiprazole. Participants who used medication were asked to take medication as usual before the lab visit to minimize possible influences of co-occurring problems such as attention problems on task performance.

Non-autistic participants were 12- to 16-year-old males (selected on matching age and gender) who participated in the first timepoint of a larger study (van der Cruijsen et al., 2018). None of the participants had any clinical diagnosis, or used medication, as was reported by parents over the phone during study inclusion, and self-reported by the adolescent in the questionnaire.

Data were analyzed using SPM8 (Wellcome Department of Cognitive Neurology, London) for comparison with previously published studies (Van der Cruijsen et al., 2019). Functional scans were corrected for slice-timing acquisition and rigid body movement differences. Structural and functional volumes were spatially normalized to T1 templates by an algorithm using a 12-parameter affine transformation together with a nonlinear transformation involving cosine basis functions, resampling the volumes to 3-mm cubic voxels. Templates were based on the MNI305 stereotaxic space (Cocosco et al., 1996). Functional volumes were spatially smoothed with a 6-mm full width at half maximum (FWHM) isotropic Gaussian kernel.

Task effects for each participant were estimated using the general linear model (GLM) in SPM8. The fMRI data were modeled as a series of zero duration events convolved with the hemodynamic response function (HRF). Modeled events of interest for the direct condition were “Direct-Academic-Positive,” “Direct-Academic-Negative,” “Direct-Physical-Positive,” “Direct-Physical-Negative,” “Direct-Prosocial-Positive,” and “Direct-Prosocial-Negative.” The same events were modeled for the reflected condition. For the control condition, only one event of interest was modeled: “Control” (collapsed across domains and valences). Trials for which participants failed to respond in time were modeled as events of no interest. The events were used as covariates in a GLM, together with a basic set of cosine functions that high-pass filtered the data. Six motion regressors were added to the model. Participants who moved more than 3 mm in any direction were excluded from the analyses (n = 4 autistic adolescents and n = 3 non-autistic adolescents). The resulting contrast images, computed on a subject-by-subject basis, were submitted to group analyses.

For motion differences between groups, see Table 1. Both across groups and within both groups separately, motion was not related to autism traits, alexithymia traits, or self-esteem (all p > 0.063). Controlling for motion in all analyses did not change the results.

See supplement for fMRI data acquisition. To investigate our hypotheses, we first performed whole-brain one-sample t tests for the contrast Self (Direct + Reflected) > Control, separately for both groups. Subsequently, we performed a whole-brain two-sample t test for the difference in this same contrast between the groups. Family-wise error (FWE) cluster correction was applied in these analyses. To further investigate our hypotheses regarding mPFC and TPJ activation, we extracted parameter estimates from 3 regions of interest (ROIs; 8-mm spheres) using the MarsBaR ROI toolbox: mPFC (x = −6, y = 50, z = 4), right TPJ (x = −53, y = −59, z = 20), and left TPJ (x = 56, y = −56, z = 18). These ROIs were based on previous meta-analyses (Denny et al., 2012; Schurz et al., 2014) and have been used in our early study on self-concept development in adolescence (Van der Cruijsen et al., 2019).

Repeated-measures analyses of variance (ANOVAs) were conducted to examine group differences in behavior and neural activation in the three ROIs. Next, hierarchical regression analyses were conducted for two purposes. First, with these analyses we examined whether behavior and neural activation related to autism traits regardless of diagnosis. Second, by adding alexithymia traits in the next step of the regression, we tested whether alexithymia would explain additional variance in self-concept and self-related neural activation above autism traits.

Results were corrected for multiple comparisons using a Bonferroni method adjusting for correlated variables (http://www.quantitativeskills.com/sisa/calculations/bonfer.htm; Perneger, 1998; Sankoh et al., 1997). For the hierarchical regressions including behavioral measures, the correlation between the seven outcome variables was r = 0.314, which resulted in an adjusted significance level (two-sided) of α = 0.013. For the hierarchical regressions including the 12 ROI measures in the contrasts Self > Control, the average correlation was r = 0.503, resulting in an adjusted significance level of α = 0.0145. Last, for the hierarchical regressions including the 12 ROI measures in the contrasts Reflected > Direct, the average correlation was r = 0.4659, resulting in an adjusted significance level of α = 0.013. Even though all hierarchical regression analyses answered one overarching question (whether alexithymia rather than autism traits were related to indicators of self-concept and self-esteem), there were three variables of interest in these analyses. Therefore, we have reported when results would not survive an additional Bonferroni correction (α = 0.013/3 = α = 0.0043; α = 0.0145/3 = α = 0.0048; and α = 0.013/3 = α = 0.0043, respectively).

Community members were not actively involved in the construction of this study. However, every year, the NAR exchanges ideas on relevant research topics with stakeholders such as autistic adults and parents of children with autism. The NAR also has several autistic team members.

First, we examined group differences in self-concept positivity (perspective, domain), self-esteem, and perspective similarity (reflected–directed). Second, we tested for relationships of these behavioral indicators with autism and alexithymia traits using hierarchical regression analyses.

First, whole-brain contrasts for Self > Control conditions for both groups (autistic and non-autistic) were constructed. Next, ROI analyses examined the main hypotheses concerning neural activation for self-concept across domains and for perspective similarity. These analyses were first performed to compare groups using repeated-measures analyses. Second, we tested for relationships of neural activation with autism and alexithymia traits using hierarchical regression analyses.

There was no evidence for a significant difference in self-concept positivity between adolescents with and without autism. We confirmed, however, the hypothesis of lower self-concept positivity in adolescents with more autism traits, across groups (Bauminger et al., 2004; McCauley et al., 2019; van der Cruijsen & Boyer, 2021; Williamson et al., 2008). We expected that this effect would be specific for the prosocial domain, but the negative relation was also observed for physical appearance traits. In contrast, no relation between autism traits and self-concept positivity was observed for academic traits, consistent with prior studies showing that academic competence is similar between groups with and without autism (Bauminger et al., 2004; Vickerstaff et al., 2007; Williamson et al., 2008).

There was no evidence for a difference in self-esteem in adolescents with and without autism. However, this study suggests that adolescents with more autism traits, across groups, may have lower general self-esteem, consistent with prior studies (McCauley et al., 2019; van der Cruijsen & Boyer, 2021). Interestingly, alexithymia (specifically DIF) explained additional variance in lower self-esteem (but not self-concept) above autism traits. This was not the case for self-concept which despite the inherently evaluative component can be seen as a relatively cognitive construct, whereas self-esteem is a more affective or emotional indicator of positivity about the self (Crone et al., 2022). This adds to the alexithymia theory such that in addition to previously studied emotional difficulties in autistic individuals, alexithymia plays a role in adolescents’ feelings of self-worth. Prior research showed that self-esteem is related to internalizing problems that also have been related to alexithymia, such as depression and anxiety (Bloch et al., 2021; McCauley et al., 2019; Milosavljevic et al., 2016; van der Cruijsen & Boyer, 2021). Therefore, self-esteem may be an important link in the development of adolescents’ mental well-being.

Autistic and non-autistic adolescents activated the same brain regions when evaluating self-traits, specifically the mPFC. This is consistent with recent studies in adolescents and early adults showing that autistic adolescents recruit similar underlying neural networks (Burrows et al., 2016; Cygan et al., 2018). Activation in mPFC for self-evaluations was negatively and positively related to different components of alexithymia. First, self-related mPFC activation was less pronounced for individuals with more DIF. Activation in mPFC underlying evaluation of self-traits proposedly reflects self-relevance or personal value (D’Argembeau, 2013; van der Cruijsen et al., 2017) and activation has been related to self-evaluations of positive compared to negative traits (D’Argembeau et al., 2008; van der Cruijsen et al., 2017, 2018). Having trouble identifying one’s feelings about the self may make the evaluation of one’s traits feel relatively less relevant or of less value to the self, associated with attenuated mPFC activation for self-evaluations. Second, DDF was positively related to mPFC activation, such that activation was stronger for individuals with more DDF. Even though DIF and DDF are correlated within individuals (Loas et al., 2001), this study highlights the importance of separately evaluating these components of alexithymia since they are differentially related to self-esteem and neural activation underlying self-evaluations. Future studies may aim to replicate and further investigate these distinct relationships of alexithymia components DIF and DDF with self-concept and related neural activation.

A second goal of this study was to investigate perspective similarity in self-evaluation. It has been theorized that the perceived opinions of others, especially peers, about the self (reflected self-concept), become internalized into one’s own opinion about the self (direct self-concept) (Dudovitz et al., 2017; Harter, 2012). This study showed that autistic and non-autistic adolescents were equally positive about themselves from the direct and reflected perspective. In addition, within-person similarity between direct and reflected self-evaluations did not differ between groups (Cage et al., 2016). However, similarity in the prosocial domain was related to autism traits across groups, such that self-evaluations of the same prosocial traits from the direct and perceived peers’ perspective were less aligned in adolescents with more autism traits (Hobson et al., 2006; Lee & Hobson, 1998). Possibly, adolescents with more autism traits are explicitly aware that their (pro)social traits are generally not well appreciated or understood by others (Carrington et al., 2003). Alternatively, adolescents with more autism traits may have more trouble estimating the opinions of others about their prosocial traits. Indeed, adolescents with autism generally struggle more to understand others’ thoughts and intentions, despite being able to form (ToM; Begeer & Scheeren, 2021).

This pattern was further investigated using neural ROI analyses. Neural activation in the TPJ for taking others’ versus own perspectives in self-evaluations did not differ between groups and was not related to autism traits across groups. This contradicts previous studies showing attenuated TPJ activation in individuals with ASC in processes involving ToM in different paradigms (Kana et al., 2015; Murdaugh et al., 2014). A prior study also showed more selective involvement of TPJ in mentalizing in individuals without autism (Lombardo et al., 2011).

There was, however, a relationship between right TPJ activation and alexithymia. That is, right TPJ activation was higher for reflected compared to direct self-evaluations in individuals with more DIF across domains and for the academic domain specifically. This does not directly align with previous studies that linked alexithymia to attenuated neural activation in regions involved in self-processing and mentalizing (Bird et al., 2010; Moriguchi et al., 2006; Van der Velde et al., 2013). Possibly, larger differentiation in TPJ activation between perspectives reflects either increased neural effort in aiming to reason from others’ perspectives or relatively lowered mentalizing activation during direct self-evaluations. Future studies may aim to replicate and break down this pattern.

This study had several limitations that should be addressed in future studies. First, it is noticeable that, although the design of this study was optimized to test for group differences, the results were mainly associated with severity of autism and alexithymia traits. Future studies should optimize their design to test for dimensional differences in autism traits by applying a population-based approach, in order to obtain samples reflecting the epidemiological prevalence of autism, alexithymia, or other psychological diagnoses (Abu-Akel et al., 2019). Second, this study was cross-sectional and therefore limited in allowing inferences on development. Future studies in adolescents should use a longitudinal design in order to better understand the origin and development of relationships between self-evaluation, underlying neural mechanisms, during childhood and adolescence. Third, unlike the participants with autism, participants without autism had no (additional) mental struggles and did not use any medication. However, this strengthens our null results regarding group differences since overlapping additional issues in both groups would make it even more difficult to strictly differentiate between groups. Fourth, due to practical limitations, this study was limited to the inclusion of adolescent males, even though the male-to-female ratio among children and adolescents with ASC is about 3 to 1 (Loomes et al., 2017). It should be kept in mind that results may be different for autistic adolescent females, especially since adolescent girls are generally lower on self-esteem (Chung et al., 2014; Robins & Trzesniewski, 2005), and self-positivity may differ between the sexes depending on the domain (Cole et al., 2001; Gentile et al., 2009; Marsh & Ayotte, 2003). Therefore, future studies should aim to also include adolescent females. Fifth, although autistic traits as measured with the AQ-short significantly differed between groups, there was a participant in the autism group who scored very low (3), and some participants in the non-autistic group scored rather high (up to 22). Future studies may aim to use an additional measure to get an indication of autism traits. Sixth, future studies may consider investigating group differences using Bayesian inference to calculate the probability of the null hypothesis being true.

This study did not observe group differences between autistic and non-autistic adolescents, but confirmed that self-concept positivity (specifically physical appearance and prosocial) and self-esteem were lower in adolescent males with more autism traits. These findings fit with recent theories showing that a dimensional approach is more suitable in mental health research (Conway et al., 2019) as autism traits may also be present along a continuum in the general non-diagnosed population. In addition, evaluating self-traits largely relies on the same neural activation in adolescents with and without autism (Cage et al., 2016), suggesting that adolescents make use of the same underlying neural processes when evaluating self-traits.

Last, this study confirmed that alexithymia is predominantly related to the social-emotional facets of evaluating the self (Bernhardt et al., 2014; Cook et al., 2013; Milosavljevic et al., 2016; Moriguchi et al., 2006). Several social-emotional problems traditionally associated with ASCs such as empathic deficits, emotion recognition difficulties, mentalizing impairments, and less interoceptive awareness have been shown to be related to alexithymia rather than autism (i.e. alexithymia hypothesis; Cook et al., 2013; Lombardo et al., 2007; Milosavljevic et al., 2016; Moriguchi et al., 2006; Shah et al., 2016). Internalizing problems often co-occurring with ASCs such as depression and anxiety may be explained by alexithymia as well (Bloch et al., 2021; Milosavljevic et al., 2016). Since negative self-concept and low self-esteem are risk factors for developing internalizing problems that increase during adolescence (Giedd et al., 2008; Kessler et al., 2005; Van Tuijl et al., 2014) and that often co-occur with ASC (Simonoff et al., 2008), future studies should examine negative self-views in adolescents with and without autism in more detail.

Together, this study informs future studies by providing a comprehensive view of adolescent self-concept and self-esteem in relation to autism and alexithymia traits, and it applies to clinical practice where it may be important to take both autism and alexithymia into account in treatments of clients with low self-esteem.