Aging in Autism: An Event‐Related Potentials Study of Auditory Change Detection

ABSTRACT

Autistic individuals are often thought to exhibit reduced tolerance for change, which may stem from atypicalities in sensory processing, particularly in the detection of environmental changes. Within the predictive coding framework, this intolerance may reflect altered generation and updating internal predictions in response to sensory input, leading to increased prediction errors. Mismatch negativity and the P3a component are event‐related potentials (ERPs) indexing automatic auditory change detection and involuntary attention shift. While studied in autistic children and young adults, little is known about how these components evolve during aging, an important gap as 21 million autistic adults over 60 are expected by 2050. This study examined mismatch negativity and P3a responses in 43 autistic adults (aged 21–64) and 60 nonautistic adults (aged 20–69) using a passive auditory oddball paradigm with standard (1000 Hz) and deviant (1100 Hz; p = 0.15) tones. Electroencephalography recording revealed no significant differences between autistic and nonautistic individuals on mismatch negativity and P3a amplitude or latency. However, mismatch negativity amplitude decreased with age over frontal site, similarly across both groups. No age‐related effects were found for mismatch negativity latency or P3a measures in either group. These findings suggest that, from adulthood onward, the neural mechanisms underlying automatic change detection and preattentional processes are similar in both autistic and nonautistic adults, and that their overall time course of aging appears comparable, supporting the parallel aging hypothesis. Nonetheless, future research should explore how these preserved neural responses relate to behavioral reactivity to change in autism.

Automatic change detection (MMN) and preattentional processes (P3a) have been studied in autistic children and young adults, but little is known about how they evolve during aging. We examined MMN and P3a in autistic and nonautistic adults aged 20–69 years. Results show that the neural mechanisms underlying automatic change detection and preattentional processes are similar in both groups, and that their overall time course of aging appears comparable, supporting the parallel aging hypothesis.

Abbreviations

ADI‐R

Autism Diagnostic Interview‐Revised

ADOS‐2

Autism Diagnostic Observation Schedule, Second Edition

AIC

Akaike information criterion

AQ

autism‐spectrum quotient

BF

Bayes factor

EEG

electroencephalography

ERP

event‐related potential

GLM

general linear models

ICA

independent component analysis

IQ

intelligence quotient

nVIQ

nonverbal intelligence quotient

MMN

mismatch negativity

MMSE

Mini‐Mental State Examination

SD

standard deviations

SOA

stimulus onset asynchrony

VIQ

verbal intelligence quotient

WAIS‐IV

Wechsler Adult Intelligence Scale, Fourth Edition

1. Introduction

Autism ¹ is a neurodevelopmental disorder characterized by challenges in social communication and the presence of restricted, repetitive, and stereotyped behaviors (American Psychiatric Association ²⁰¹³). While some studies have been conducted in children and young adults, research on aging in autism remains scarce. This gap is particularly concerning, as the aging population grows (World Health Organization ²⁰²²) and, considering a prevalence of autism at 1% (Zeidan et al. 2022), the number of autistic individuals over 60 is projected to reach 21 million by 2050. Moreover, there is evidence for an increased vulnerability to neuroevolutive conditions (e.g., Starkstein et al. ²⁰¹⁵; Vivanti et al. ^{2021, 2025}), further underscoring the importance of studying older populations. However, research involving older autistic populations presents specific challenges. Life expectancy has been reported to be lower in autistic individuals compared to the general population, suggesting that those who reach older age may represent a particular subgroup and may not fully capture the heterogeneity of the broader autistic population. However, recent studies indicate that this reduction may be more modest than previously suggested (Krantz et al. 2023; O’nions et al. ²⁰²⁴). In addition, many older autistic adults receive diagnoses later in life, which may further bias samples toward individuals with fewer support needs. Nevertheless, little is known about how aging affects perceptive and neural processes in autism. To date, three main hypotheses have been proposed to describe neurocognitive aging in autism: the accelerated aging hypothesis, which aligns with the concept of premature aging, proposes that autistic adults may undergo cognitive decline earlier or at a faster rate than nonautistic individuals. In contrast, the safeguard aging hypothesis suggests that early‐life cognitive difficulties could lead to the development of compensatory strategies that help slow the aging process (Bessé et al. 2025; Geurts and Vissers ²⁰¹²; Wang et al. ²⁰²⁴). The parallel aging hypothesis argues that autistic and nonautistic individuals follow similar aging trajectories. This perspective was further elaborated by Wang et al. (2024), who divided parallel aging into two subhypotheses: one proposing that cognitive performance remains similar across the lifespan, and another suggesting that autistic individuals may display lower overall cognitive performance but follow a similar aging pattern, potentially leading to an earlier onset of decline (for a detailed explanation, see Wang et al. ²⁰²⁴). Despite heterogeneous findings across the literature, existing reviews indicate that cross‐sectional studies examining cerebral and cognitive aging in autism tend to support either the parallel or safeguard aging hypotheses. Conversely, the limited number of longitudinal investigations appear more consistent with the parallel or accelerated aging hypotheses (Bessé et al. 2025). To better examine these hypotheses, it is necessary to identify neural markers that are sensitive to aging effects in the general population and can be reliably measured across the lifespan. In this context, mismatch negativity (MMN), a potential index of cognitive and perceptual aging, represents a valuable tool for characterizing age‐related neural changes in autistic individuals. This is supported by recent findings suggesting that MMN alterations may serve as early indicators of age‐related cognitive decline in nonautistic older adults (Yang et al. 2023). Beyond its sensitivity to aging, MMN is highly relevant to autism research because it provides insight into neural mechanisms underlying the automatic detection of unexpected sensory events. Alterations in change‐detection and predictive processing mechanisms indexed by MMN may contribute to core behavioral characteristics of autism, such as resistance to change (American Psychiatric Association ²⁰¹³), which is observed across multiple sensory modalities, including the auditory domain (Boyd et al. 2010; Liss et al. ²⁰⁰⁶; Rogers et al. ²⁰⁰³). This reduced tolerance for change may stem from atypical sensory processing, particularly in the automatic detection of environmental changes (Gomot et al. 2002, 2006, 2011). Difficulties in detecting change may, in turn, be associated with challenges in predicting future events (Gomot and Wicker 2012). MMN is elicited in auditory oddball paradigms, in which rare deviant sounds are embedded within sequences of frequent standard sounds. It is thought to reflect the prediction error generated by the comparison between incoming sensory input and memory‐based predictions (Garrido et al. 2009; Winkler ²⁰⁰⁷; Winkler et al. ¹⁹⁹⁶; Winkler and Czigler ²⁰¹²). Such prediction error has been conceptualized within the framework of Bayesian brain and predictive coding theories and applied to autism, suggesting that autistic individuals generate internal predictions that differ from those of neurotypical individuals, making it harder to update internal memory representations in response to changing sensory inputs (Pellicano and Burr 2012; Van de Cruys et al. ²⁰¹⁴). This mismatch may lead to increased prediction errors (see for review: Rapaport and Sowman ²⁰²⁴) and could contribute to core features of autism such as a preference for sameness, avoidance of unpredictability, and repetitive behaviors (Gomot and Wicker 2012; Rapaport and Sowman ²⁰²⁴). MMN typically manifests as a fronto‐central negative deflection occurring approximately 150–250 ms following stimulus onset (see Näätänen et al. ²⁰⁰⁷ for review) and is often followed by the P3a component, which occurs around 220–280 ms poststimulus and is associated with involuntary attentional orienting toward novel stimuli (Friedman et al. 2001; Squires et al. ¹⁹⁷⁵). While both frequency and duration deviants have been used to investigate MMN in autism, this study focuses on frequency deviants, one of the most widely studied auditory features in autism (Chen et al. 2020) and aging (Cheng et al. 2013), providing a clear baseline for investigating auditory change detection in older autistic adults.

In the general population, MMN amplitude declines with age, sometimes accompanied by longer latencies (Aghamolaei et al. 2018; Alain and Woods ¹⁹⁹⁹; Bertoli et al. ²⁰⁰²; Cooper et al. ²⁰⁰⁶; Czigler et al. ¹⁹⁹²; Gaeta et al. ¹⁹⁹⁸; Rimmele et al. ²⁰¹²; Schiff et al. ²⁰⁰⁸; Tsolaki et al. ²⁰¹⁵; Yang et al. ²⁰²³, see for reviews: Cheng et al. ²⁰¹³; Näätänen et al. ²⁰¹²), and P3a amplitude is reduced with prolonged latency in older adults (Aghamolaei et al. 2018; Gaeta et al. ¹⁹⁹⁸; Knight ¹⁹⁸⁷; Schiff et al. ²⁰⁰⁸; Tsolaki et al. ²⁰¹⁵). Given these age‐related changes in MMN and P3a observed in the general population, these components are particularly important to investigate in autism. Furthermore, understanding the brain's ability to adapt to sensory information in autism and how aging influences the neural mechanisms underlying automatic change detection may also shed light on behavioral manifestations related to intolerance of change and their evolution across the lifespan.

Research on MMN and P3a in autism has mostly focused on children and adolescents. Two recent meta‐analyses produced conflicting results: Chen et al. (2020) found that autistic individuals showed MMN amplitudes similar to nonautistic peers but with shorter response latencies. In contrast, Schwartz et al. (2018) found reduced MMN amplitude but no latency differences. However, they did not distinguish between different types of deviants (e.g., frequency, duration, or phoneme), which may account for the inconsistent findings. A few studies have explored MMN in young autistic adults, but findings remain inconsistent. MMN amplitude has been reported as either reduced (Reisli and Molholm 2023) or similar (Chen et al. 2020; Chien et al. ²⁰¹⁸; Kujala et al. ²⁰⁰⁷; Schwartz et al. ²⁰¹⁸), with latency findings showing both shorter (Kujala et al. 2007) and comparable latencies (Chien et al. 2018; Lepistö et al. ²⁰⁰⁷; Schwartz et al. ²⁰¹⁸) compared to nonautistic young adults. Regarding the P3a component, research has also largely been focused on children and adolescents. Some studies report similar (Cui et al. 2017; Vlaskamp et al. ²⁰¹⁷) or greater P3a amplitude in autistic children (Gomot et al. 2011; Yu et al. ²⁰¹⁵), with similar latency findings (Cui et al. 2017; Gomot et al. ²⁰¹¹; Vlaskamp et al. ²⁰¹⁷; Yu et al. ²⁰¹⁵). In young autistic adults, P3a amplitude results are also mixed. Studies report either no group differences (Chien et al. 2018; Cui et al. ²⁰¹⁷; Kujala et al. ²⁰⁰⁷) or increased amplitudes in autistic individuals (Lepistö et al. 2007), while latency results show either similar (Chien et al. 2018; Cui et al. ²⁰¹⁷; Lepistö et al. ²⁰⁰⁷) or shorter latencies in autism (Kujala et al. 2007). These discrepancies may be explained by the limited number of studies which used different experimental parameters, such as stimulus frequency, duration, and combinations of frequency‐duration deviance within the same experiment. However, the trajectory of ERP components such as MMN and P3a in autism across aging remains largely unknown. In addition to MMN and P3a, early obligatory auditory components such as N1 and P2 provide complementary information about basic sensory encoding, which is particularly relevant given the sensory particularities often reported in autism and given the age‐related effects reported for some of the obligatory sensory responses. In autism, especially among cognitively able individuals, N1 and P2 responses are frequently described as comparable to those of nonautistic peers (e.g., Borgolte et al. ²⁰²¹; Haigh et al. ²⁰²²). In the general population, N1 responses are generally reported as relatively preserved with age (Bertoli et al. 2002; Czigler et al. ¹⁹⁹²; Gaeta et al. ¹⁹⁹⁸; Schiff et al. ²⁰⁰⁸), whereas the P2 component more consistently shows age‐related amplitude reductions and prolonged latencies (Bertoli et al. 2002; Cooper et al. ²⁰⁰⁶; Czigler et al. ¹⁹⁹²). Examining these obligatory sensory responses alongside MMN and P3a therefore makes it possible to determine whether potential group differences in deviance detection reflect alterations in predictive mechanisms rather than differences in low‐level auditory processing. Including these measures is particularly important when comparing groups using the same oddball paradigm, as it ensures that both groups perceive and encode the acoustic stimuli similarly before assessing change detection processes.

This study aims to investigate the evolution of MMN and P3a in autism across the adult lifespan by using a cross‐sectional sample of autistic adults (aged 21 to 64 years), compared to nonautistic individuals (aged 20–69 years), together with obligatory sensory response. We used a passive auditory oddball task with tone‐frequency deviants due to their relevance in autism research. The goal was to determine whether the neural mechanisms underlying change detection evolve with age in autistic individuals and whether these changes follow a similar or distinct pattern compared to nonautistic aging. We hypothesize that both MMN and P3a components will show the typical age‐related reductions in amplitude and increases in latency in nonautistic individuals (e.g., Aghamolaei et al. ²⁰¹⁸; Cheng et al. ²⁰¹³; Tsolaki et al. ²⁰¹⁵). Regarding obligatory components, we expected N1 to remain relatively stable with age, whereas P2 was expected to show age‐related latency increases and possible amplitude modulation (Bertoli et al. 2002; Cooper et al. ²⁰⁰⁶; Czigler et al. ¹⁹⁹²). Examining N1 and P2 responses therefore allowed us to verify whether potential group differences in MMN and P3a reflect alterations in predictive change‐detection mechanisms rather than differences in basic auditory encoding. Furthermore, group comparisons will contribute to understanding whether autistic individuals age similarly to nonautistic individuals or display a distinct aging pattern, in line with the accelerated or the safeguard hypothesis.

2. Materials and Methods

2.1. Participants

Forty‐three autistic adults (34 males, 9 females) aged 21–64 years and 60 nonautistic adults (36 males, 24 females) aged 20–69 years (see Table 1) participated in this study. All participants volunteered for the study and provided written consent, with an additional legal guardian signature obtained when required. Ethical approval for the study was obtained from the Ethics Committee (PROSCEA: IDRCB: 2017‐A00756‐4). Participants received a compensation of 20€ for their involvement.

TABLE 1.

Descriptive statistics of demographic, cognitive and clinical variables across groups.

			t‐tests	Chi‐squared
	Autistic adults n = 43	Nonautistic adults n = 60	t(101)	χ 2
Demographic
Age (years)	40.37 (13.21) [21–64]	45.30 (15.85) [20–69]	−1.67 ns (d = −0.33)
Male %	79%	60%		3.35 ns (V = 0.18)
Education (years)	14.35 (2.88)	14.92 (2.42)	−1.09 ns (d = −0.22)
AQ	35.60 (7.38)	16.38 (7.38)	12.94*** (d = 2.60)
Cognitive
VIQ	122.98 (18.10)	120.48 (14.95)	0.76 ns (d = 0.15)
nVIQ	108.05 (15.79)	106.40 (14.58)	0.55 ns (d = 0.11)
MMSE	28.67 (1.49)	28.93 (0.99)	−1.06 ns (d = −0.21)
Clinical
ADOS‐2 (43/43)	9.65 (4.97)
ADI‐R (28/43)
Reciprocal social interaction	16.46 (7.81)
Communication	11.36 (5.81)
Restricted, repetitive behaviors	4.50 (2.32)

Note: Mean (SD) [range]. Group differences in demographic variables were assessed using independent‐samples t‐tests, while gender differences were analyzed with chi‐square (χ²) tests. Effect sizes were reported using Cohen's d for t‐tests and Cramer's V for chi‐square tests.

Abbreviations: AQ, Autism Spectrum Quotient; MMSE, Mini‐Mental State Examination; nVIQ, estimated nonverbal intelligence quotient based on the cube and matrix subtests of the WAIS‐IV; ns, not significant; VIQ, estimated verbal intelligence quotient based on the vocabulary and similarity subtests of the WAIS‐IV.

***p < 0.001.

Autistic participants were recruited in France through the Autism Resource Center of Centre‐Val de Loire—EXcellence Center for Autism—Tours and the researcher's professional network. Participants were eligible if they had a clinical diagnosis of autism based on DSM‐5 criteria (American Psychiatric Association ²⁰¹³), confirmed by either the Autism Diagnostic Observation Schedule (ADOS‐2, Lord et al. ²⁰¹²; 43/43) and/or the Autism Diagnostic Interview (ADI‐R, Lord et al. ¹⁹⁹⁴; 28/43). Five participants scored slightly below the ADOS‐2 cutoff (< 7). However, they were diagnosed in later adulthood, and their diagnoses were confirmed by a multidisciplinary team of clinicians. Among all autistic individuals, 91% (39/43) were diagnosed during adulthood. Nonautistic participants were recruited via media advertisements and social networks. Selected subtests from the WAIS‐IV (Wechsler 2008) were used to assess verbal IQ (Vocabulary and Similarities subtests) and nonverbal IQ (Cube and Matrix reasoning subtests). This study was part of a larger project involving several cognitively demanding tasks. Participation therefore required the ability to complete cognitively demanding experimental tasks and tolerate EEG procedures. Consequently, the study primarily included autistic adults with cognitive and adaptive functioning allowing to perform the study. All participants also completed the self‐reported 50‐item Autism‐Spectrum Quotient questionnaire to assess traits related to autism (AQ, Baron‐Cohen et al. ²⁰⁰¹). The total score ranges from 0 to 50, with higher scores reflecting greater autistic traits.

Inclusion criteria for all participants were (1) being aged between 18 and 80 and (2) the ability to understand and follow instructions for active tasks. The exclusion criteria included issues with uncorrected visual acuity or abnormal hearing thresholds. For nonautistic participants, additional exclusion criteria were (1) a prior clinical diagnosis of neurodevelopmental disorder or other psychiatric conditions, (2) a close family member with ASD, (3) neurological disorders or a history of traumatic brain injury, (4) significant developmental difficulties in walking, language, or learning, (5) use of any psychotropic medication, and (6) a Mini‐Mental State Examination (MMSE; Folstein et al. ¹⁹⁷⁵) score below 26. This last assessment measures general cognitive ability and has a maximum score of 30. This cutoff of 26 was chosen to reduce the likelihood of including individuals with cognitive impairments.

For autistic participants, one individual reported a history of traumatic brain injury. However, this participant's cognitive scores were within the range of the autistic group, and the pattern of results remained unchanged when analyses were conducted with and without this participant. Therefore, the participant was retained in the final sample. Participants taking psychotropic medications were required to be stabilized for at least 1 month prior to participation, and individuals taking anxiolytic medications were excluded because of potential interference with EEG recordings.

Participants had normal hearing as assessed through subjective audiometry (250 Hz, 1000 and 1500 Hz) prior to the investigation of auditory evoked potentials. 86% of autistic adults and 92% of nonautistic adults were right‐handed (Fisher's exact test, p > 0.05) (Edinburgh Handedness Inventory; Oldfield ¹⁹⁷¹). There were no significant differences in sex ratio, education, global cognitive abilities, verbal and nonverbal IQ between autistic and nonautistic participants. Autistic individuals had significantly higher autistic traits, as assessed through the AQ scale. Participants' characteristics are detailed in Table 1.

2.2. Stimuli and Procedure

To investigate change detection processes, a passive auditory oddball paradigm was employed. The auditory stimuli included standard tones (1000 Hz) and deviant tones (1100 Hz) presented with a deviant probability of 15%, with the constraint that 2 deviant tones were always preceded by at least three consecutive standard tones. A total of 1354 tones were played at 70 dB SPL, each lasting 50 ms with 5‐ms rise and fall time. The tones were delivered through loudspeakers with a consistent SOA of 700 ms. Participants sat 110 cm away from a TV screen and were instructed to watch a silent movie while the 15‐min recording session was conducted. To ensure that participants maintained attention during the silent movie, they were instructed to recount the story at the end of the session. This procedure was applied consistently across all participants to monitor vigilance and minimize potential differences in arousal or engagement between groups.

2.3. EEG Recording

Continuous electroencephalogram (EEG) was recorded using 64 active electrodes arranged according to the 10–20 international system (Jasper 1958) using the BioSemi system (ActiveTwo, BioSemi, Amsterdam, the Netherlands). In addition, the nose served as the reference and Fpz as the ground electrode. Electrode impedance values were maintained below 30 kΩ and homogeneous. For EOG recordings, two electrodes were positioned at the outer corners of the eyes (HEOG), and one was placed beneath the left eye (VEOG). Signals from both EEG and EOG were sampled at 512 Hz and filtered offline with a 0.1‐ to 30‐Hz bandpass filter. Artefacts caused by movements were manually excluded, and ocular activity effects were corrected using a semiautomatic process based on independent component analysis (ICA) in the Brain Vision Analyzer toolbox 2.2 (Brain Products, Gilching, Germany). Baseline correction was applied by referencing activity to the 100‐ms period preceding stimulus onset. Data were segmented into 700‐ms epochs, including a 100‐ms prestimulus baseline, and averaged separately for standard and deviant stimuli. A minimum of 120 responses to deviant tones were included for each participant. Difference waveforms were computed by subtracting responses to standard tones from those to deviant tones, allowing the extraction of MMN and P3a components. For each participant, the MMN peak amplitude and latency were semiautomatically identified as the most negative peak at the frontal site and the most positive peak at the mastoid sites, within a ± 30‐ms window centered on their individual peak latency. Similarly, the P3a peak amplitude and latency were semiautomatically identified as the most positive point within a ± 30‐ms window around each individual's peak. Electrode sites were selected a priori based on the typical scalp distribution of each ERP component. MMN measures were extracted at the frontal‐central Fz electrode and at the mastoid electrodes (left: P9 and right: P10), which served as primary sites for MMN analysis under nose‐referenced conditions (Näätänen et al. 2007). At the mastoids, the MMN typically appears as a polarity‐inverted positivity due to the orientation of the underlying auditory dipole in the superior temporal gyrus. In contrast, P3a was measured exclusively at the Fz electrode, as this frontocentral location typically exhibits the largest P3a amplitude associated with preattentional orienting. N1 and P2 components were also extracted to assess obligatory responses to standard tones. The N1 was defined as the most negative point within a ± 20‐ms window around each participant's individual peak, and the P2 as the most positive point within a ± 30‐ms window. N1 and P2 amplitudes and latencies were measured at the Cz electrode, as this central electrode usually exhibits the largest amplitude.

2.4. Statistical Analysis

Statistical analyses were conducted using RStudio (Version 4.5.0, RStudio Team ²⁰²⁰). To assess early auditory processing, amplitude and latency of the N1 and P2 components in response to standard stimuli were analyzed at electrode Cz using General Linear Models (GLMs; LM function in R), with age as a continuous predictor and group (autistic vs. nonautistic) as a between‐subjects factor. If significant interactions were found, follow‐up group‐wise regression analyses were performed to explore the effects of age separately on amplitude and latency.

To confirm the evolution with age of the MMN and P3a components, we first analyzed age‐related changes within the nonautistic group. For the MMN, amplitude and latency were analyzed at the predefined frontal‐central electrode (Fz) and at bilateral mastoid electrodes (P9/P10) using GLMs (LM function in R) with age as a continuous predictor. Mastoid site (left vs. right) was included as a within‐subject factor to examine potential hemispheric differences. The P3a component was analyzed exclusively at electrode Fz using similar GLM approach.

The age‐related evolution of the MMN and P3a components was then assessed in both autistic and nonautistic individuals using GLMs (LM function in R) with continuous age and group as predictors. For the MMN, two separate analyses were conducted: a first analysis examined MMN amplitude at the Fz electrode, and a second analysis examined MMN amplitude at the mastoid electrodes (P9 and P10), with electrode site (P9 vs. P10) included as a within‐subject factor. For the P3a, a similar model was applied at Fz only. If interactions involving group or electrode site were significant, follow‐up regressions were conducted within each group or site to clarify how age influenced the amplitude and latency of each component.

For each measure, effect sizes were calculated using eta‐squared (η ²). Quadratic age terms (Age²) were tested in each analysis to explore potential nonlinear effects, but Akaike information criterion (AIC) comparisons, which compare model quality, showed no improvement of model fit. Thus, linear models were retained.

Additionally, to assess the strength of evidence for potential null effects, specifically for the nonsignificant age × group interaction terms, Bayesian model comparisons were conducted using the BayesFactor package in R (Morey and Rouder 2012). This method followed the approach described by Torenvliet et al. (2024), who also studied aging in autism. Default priors were used to calculate Bayes factors (BF₁₀) which quantify evidence for the more complex model (e.g., age × group interaction) relative to the corresponding simpler model (e.g., group or age only). A BF₁₀ greater than 3 was interpreted as substantial evidence for the interaction term, while values below 0.3 indicated support in favor of the simpler model (without the interaction). These Bayesian analyses were conducted as a complementary approach to the GLM results. They provide a direct quantification of evidence supporting the absence of age × group interactions.

As a complementary analysis, cluster‐based permutation tests were conducted to identify significant effects across electrodes and time points of the MMN while controlling for multiple comparisons. These analyses were intended to complement the GLM results by assessing whether the observed effects extended beyond the predefined electrodes used in the primary analyses. Analyses were performed in Python (Version 3.13.9) using the MNE‐Python library (Version 1.12.1). For each electrode and time point, a statistical comparison was computed between experimental conditions. Comparisons between the two age groups (young vs. older) and between autistic and nonautistic groups, whatever the age were performed. Within each subgroup (young autistic, older autistic, young nonautistic, older nonautistic), we assessed the presence of a significant MMN by testing whether the deviant–standard difference significantly differed from zero. Adjacent significant samples were grouped into clusters based on their temporal and spatial contiguity and the cluster‐level statistics were calculated as the sum of the statistical values within each cluster. Statistical significance was assessed using a nonparametric permutation procedure with 10,000 permutations (significant cluster‐level p < 0.05).

3. Results

Results are illustrated in Figures 1, 2, 3, 4. Figure 1 shows the standard, deviant, and difference waveforms for each group, upon which the N1, P2, MMN, and P3a components can be observed. Figures 2 and 3 display MMN and P3a waveforms at Fz, left and right mastoids, along with corresponding scalp topographies. Although age was treated as a continuous variable in all statistical analyses, participants were divided into younger (< 40 years) and older (> 40 years) subgroups in Figures solely for visualization purposes and to facilitate graphical interpretation of age‐related trends. Descriptive statistics for these visualization‐based subgroups are provided in the Supporting Information (Tables S1–S3) for transparency. Waveforms have been smoothed over a time window of 20 ms for visualization purposes. Figure 4 illustrates age‐related changes in the amplitude of the MMN component at the frontal site in both autistic and nonautistic individuals.

FIGURE 1.

Standard, deviant, and difference waveforms (MMN) in younger (< 40 years) and older (≥ 40 years) autistic and nonautistic participants. Waveforms are shown separately for younger and older participants, with age groups defined for visualization only, while age was treated as a continuous variable in all analysis.

FIGURE 2.

Group‐averaged MMN and P3a waveforms at Fz, left (P9) and right (P10) mastoids in Younger (< 40 years) and older (≥ 40 years) autistic and nonautistic participants.

FIGURE 3.

Topographic maps of MMN (3a) and P3a (3b) components for younger (< 40 years) and older (≥ 40 years) autistic and nonautistic participants. Topographic maps were plotted on the ±20‐ms time window centered around the peak of the grand average waveform for each group.

FIGURE 4.

Age‐related effects on MMN amplitude at Fz as a function of age and group.

3.1. Analysis of Auditory Components (N1 and P2)

GLMs overall analysis was not significant for the amplitude (F(3,99) = 0.90, p = 0.44, R ² = −0.003) or latency (F(3,99) = 1.49, p = 0.22, R ² = 0.01) of the N1 component, nor on the amplitude of the P2 component (F(3,99) = 0.42, p = 0.73, R ² = −0.02), indicating that the effect of age, group, or their interaction was not significant for any of these measures (all η ² range from 0.03 to < 0.00). In contrast, the overall GLM for P2 latency was significant (F(3,99) = 6.14, p = 0.001, R ² = 0.13). A significant main effect of age was found, with P2 latency increasing with age (β = 0.58, p < 0.001, η ² = 0.16). There was no significant effect of group (β = −0.30, p = 0.94, η ² = 0.00) nor a significant age × group interaction (β = −0.04, p = 0.88, η ² = 0.00), indicating that latency did not differ between groups, and the age‐related increase in latency was consistent across groups. (see Table 2 for descriptive statistics for the N1 and P2 components in each group).

TABLE 2.

Means and standard deviations (SD) of N1 and P2 amplitude (μV) and latency (ms) in autistic and nonautistic individuals.

	Autistic adults	Nonautistic adults
Cz amplitude μV
N1	−1.57 (1.35)	−1.41 (1.12)
P2	1.71 (0.99)	1.94 (1.18)
Cz latency ms
N1	99.4 (16.20)	95.4 (11.61)
P2	169.5 (21.54)	172.0 (22.32)

Bayesian analyses supported these findings, consistently indicating that models including the age × group interaction were not favored compared to simpler models, providing strong evidence against the added value of the interaction term. This pattern was consistent across analyses of N1 and P2 for amplitude and latency, with BF₁₀ values ranging from 0.09 to 0.02.

3.2. Age‐Related Effects in the Nonautistic Group (MMN and P3a)

Age‐related changes were initially examined in nonautistic participants to verify the effects of aging on MMN and P3a components.

In the nonautistic group, a significant negative effect of age was observed for MMN peak amplitude at Fz site (F(1,58) = 3.92, p = 0.05, R ² = 0.04, β = 0.01, η ² = 0.06), indicating reduced amplitude with increasing age. No significant effect of age was found for MMN latency (F(1,58) = 0.11, p = 0.75, R ² = −0.02, β = 0.05, η ² = 0.00). For MMN peak amplitude at mastoid sites, the overall GLM was significant (F(3,116) = 3.78, p = 0.012, R ² = 0.07), driven by a higher amplitude at the left mastoid (P9) compared to the right (P10) (β = 1.30, p = 0.03, η ² = 0.07). All other effects were nonsignificant (p > 0.05, η ² < 0.01). The overall GLM for MMN latency at mastoid sites was not significant (F(3,116) = 0.04, p = 0.99, R ² = −0.02) and neither age (β = −0.01, p = 0.92, η ² = 0.00), electrode site (β = −0.98, p = 0.78, η ² = 0.00), nor the interaction between age and electrode site (β = 0.04, p = 0.87, η ² = 0.00) predicted the outcome.

No significant age effect was observed on P3a peak amplitude (F(1,58) = 1.13, p = 0.29, R ² = 0.002, β = −0.01, η ² = 0.02) or latency (F(1,58) = 0.00, p = 0.99, R ² = −0.02, β = −0.01, η ² = 0.00). (See Table 3 for descriptive statistics of MMN and P3a components in nonautistic individuals).

TABLE 3.

Means and standard deviations (SD) of MMN and P3a amplitude (μV) and latency (ms) in autistic and nonautistic individuals.

	Autistic adults	Nonautistic adults
Amplitude (μV)
MMN
Fz	−2.61 (0.90)	−2.31 (0.91)
Left mastoid (P9)	1.65 (1.15)	1.69 (1.13)
Right mastoid (P10)	1.05 (1.17)	1.12 (1.01)
P3a
Fz	1.77 (1.10)	1.95 (1.14)
Latency (ms)
MMN
Fz	141.8 (22.49)	140.5 (20.15)
Left mastoid (P9)	143.7 (17.33)	139.7 (19.23)
Right mastoid (P10)	143.4 (18.68)	140.6 (19.27)
P3a
Fz	284.3 (38.95)	277.3 (36.45)

3.3. Age‐Related Effects Across Autistic and Nonautistic Groups (MMN and P3a)

The overall GLM model for MMN amplitude at Fz was significant (F(3,99) = 3.09, p = 0.03, R ² = 0.06) (see Figure 4). A significant main effect of age was found, with increasing age associated with smaller MMN amplitude (β = 0.02, p = 0.01, η ² = 0.07). There was no significant effect of group (β = 0.22, p = 0.22, η ² = 0.02) nor a significant age × group interaction (β = −0.002, p = 0.87, η ² = 0.00), indicating that the age‐related decrease in amplitude was similar in autistic and nonautistic individuals. In contrast, the overall GLM for MMN latency at Fz was not significant (F(3,99) = 0.56, p = 0.65, R ² = −0.01), and none of the predictors: age (β = 0.13, p = 0.34, η ² = 0.00), group (β = −1.94, p = 0.65, η ² = 0.00), or their interactions (β = −0.25, p = 0.42, η ² = 0.00), were associated with MMN latency at Fz.

The GLM model for MMN amplitude at mastoids sites was significant (F(7,198) = 2.70, p = 0.01, R ² = 0.05). A main effect of electrode site was observed, indicating larger amplitude at the left (P9) than at the right mastoid (P10) (β = 0.58, p < 0.001, η ² = 0.07). No other predictors reached significance (all p > 0.05 and η ² < 0.01). In contrast, the overall GLM for MMN latency at mastoid sites was not significant (F(7,198) = 0.37, p = 0.92, R ² = −0.02), indicating that neither age (β = 0.03, p = 0.67, η ² = 0.00), group (β = −3.63, p = 0.18, η ² = 0.00), electrode site (β = −0.45, p = 0.86, η ² = 0.00), or any interaction (all p > 0.05 and η ² < 0.00) predicted latency.

The overall GLM for P3a amplitude (F(3,99) = 0.74, p = 0.53, R ² = −0.01) and latency (F(3,99) = 0.53, p = 0.67, R ² = −0.01) were not significant, indicating that neither age (amplitude: β = −0.01, p = 0.21, η ² = 0.01; latency: β = 0.11, p = 0.64, η ² = 0.00), group (amplitude: β = 0.23, p = 0.32, η ² = 0.00; latency: β = −7.64, p = 0.32, η ² = 0.00), or interaction between age and group (amplitude: β = −0.001, p = 0.93, η ² = 0.00; latency: β = −0.37, p = 0.49, η ² = 0.00) predicted the amplitude or latency of the P3a component. (See Table 3 for descriptive statistics of MMN and P3a components in autistic and nonautistic individuals).

Bayesian analyses supported these results, indicating that adding the age × group interaction term was not supported over simpler models. The evidence against the added value of the interaction was substantial to extreme across MMN and P3a amplitude and latency measures, at both frontal‐central and mastoid electrodes, with BF₁₀ values ranging from 0.12 to 0.002.

Finally, complementary cluster‐based permutation analyses were consistent with the GLM results regarding the absence of significant MMN differences between the autistic and nonautistic groups (p > 0.10). For the age comparison, the analysis revealed a fronto‐central cluster spanning approximately 110–210 ms that showed a trend toward significance (cluster p = 0.07). This finding was spatially consistent with the significant age effect observed at Fz in the GLM but did not reach the cluster‐level significance threshold (Figure 5). This result should be interpreted cautiously, as the age groups were defined using a 40‐year cutoff, with several participants having ages close to this threshold, which may have reduced the sensitivity of the categorical age comparison. Finally, one‐sample cluster‐based permutation tests confirmed that the MMN significantly differed from zero in all four subgroups (young autistic, older autistic, young nonautistic, and older nonautistic), with significant fronto‐central clusters in each case (all cluster p < 0.0002).

FIGURE 5.

(A) Spatiotemporal cluster identified by the cluster‐based permutation analysis comparing younger and older adults. (B) Grand‐average MMN waveforms averaged across electrodes included in the cluster for younger (blue) and older (green) adults. The grey shaded region indicates the temporal extent of the cluster (111–211 ms). (C) MMN scalp potential distributions averaged over the 111–211 ms interval in younger and older adults, and topographical difference map obtained from cluster‐based permutation analyses.

4. Discussion

This study investigated how auditory change detection mechanisms evolve across the adult lifespan in autistic individuals with high levels of cognitive and adaptive functioning, compared to nonautistic individuals, using a cross‐sectional sample ranging from 20 to 69 years of age. Age‐related modulations in ERP components associated with automatic change detection (MMN) and preattentional processes (P3a) were examined using a passive auditory oddball paradigm with frequency deviants. Early auditory processing (N1 and P2) was also studied in relation to standard tones.

4.1. Early Obligatory Auditory Components (N1 and P2)

This study addresses a gap in the literature by examining early auditory processing across adulthood and aging in autism, enabling us to verify whether basic sensory encoding is preserved and to rule out low‐level auditory processing differences as a potential confound. We found no significant group differences in N1 or P2 amplitude or latency, aligning with prior work in autistic adults (e.g., Borgolte et al. ²⁰²¹; Haigh et al. ²⁰²², Courchesne et al. 1985), suggesting that early auditory processing remains largely preserved in our autistic adults group. Similarly, no main effects of age or age‐by‐group interactions were observed on N1 measures (see Bertoli et al. ²⁰⁰²; Czigler et al. ¹⁹⁹²; Gaeta et al. ¹⁹⁹⁸; Schiff et al. ²⁰⁰⁸ for similar results). For the P2 component, amplitude remained stable with age (see Bertoli et al. ²⁰⁰²; Cooper et al. ²⁰⁰⁶; Czigler et al. ¹⁹⁹² for different results), though latency increased with age in both groups, consistent with previous findings (e.g., Bertoli et al. ²⁰⁰²; Cooper et al. ²⁰⁰⁶). Taken together, the results suggest that aging does not significantly impact early auditory detection per se, as reflected by the N1, but rather leads to a slowing of later stages of perceptual processing, indexed by the P2 component, which is believed to reflect automatic attentional modulation or early stimulus classification (see for review: Crowley and Colrain ²⁰⁰⁴). This age‐related delay appears similar in autistic and nonautistic individuals, supporting the idea of parallel neural aging. These results underscore the limited sensitivity of N1 and P2 to age‐related or autism‐related changes and highlight the MMN as a more promising component for investigating perceptual and cognitive decline across aging in autism.

4.2. Automatic Change Detection and Preattentional Processes (MMN and P3a)

Our results revealed no significant differences between groups in either amplitude or latency for MMN or P3a. Although research into auditory processing in autistic adults remains limited, our findings are consistent with prior studies suggesting that MMN and P3a responses tend to get closer to those of nonautistic individuals in adulthood (MMN: Chen et al. ²⁰²⁰; Chien et al. ²⁰¹⁸; Kujala et al. ²⁰⁰⁷; Lepistö et al. ²⁰⁰⁷; Schwartz et al. ²⁰¹⁸), P3a: Chien et al. ²⁰¹⁸; Cui et al. ²⁰¹⁷; Kujala et al. ²⁰⁰⁷; Lepistö et al. ²⁰⁰⁷). In adulthood, autistic individuals appear to detect auditory changes and reorient attention as efficiently and rapidly as nonautistic adults. From a predictive coding perspective, this suggests that the internal models used to generate predictions in response to sensory input may become more aligned with those of nonautistic individuals in adulthood. In other words, the prediction errors may reduce over time, leading to more typical sensory responses in adulthood.

Regarding age effects, a significant age‐related decline in MMN amplitude over frontal site was found, consistent with previous studies (Cheng et al. 2013; Näätänen et al. ²⁰¹²). Given the absence of age effects on the N1 component, the decline in MMN amplitude does not appear to reflect changes in early sensory processing but rather suggests a specific decline in change processing. Unlike studies using categorical age groups, the use of a continuous age variable provides evidence of a gradual decline starting as early as the 20s. Consistent with this finding, the complementary cluster‐based permutation analysis showed only a trend toward an age‐related difference when participants were divided into younger and older groups. This may reflect the lower sensitivity of using age categories rather than age as a continuous variable, particularly because several participants were close to the 40‐year cutoff. MMN latency, however, remained stable across age (see Aghamolaei et al. ²⁰¹⁸; Alain and Woods ¹⁹⁹⁹; Rimmele et al. ²⁰¹²; Schiff et al. ²⁰⁰⁸; Yang et al. ²⁰²³ for similar results), suggesting that while the ability to discriminate auditory change decreases with age, the speed of this processing remains preserved. Regarding the P3a component, no significant age‐related effects were observed in either the amplitude or latency, contrasting with previous studies (Aghamolaei et al. 2018; Gaeta et al. ¹⁹⁹⁸; Knight ¹⁹⁸⁷; Schiff et al. ²⁰⁰⁸; Tsolaki et al. ²⁰¹⁵). This discrepancy may be due to our relatively limited upper age range as well as the high levels of cognitive functioning in our sample (mean verbal IQ = 120.48; mean nonverbal IQ = 106.40) which may have contributed to the preservation of preattentional processes in older adults. Additionally, using age as a continuous variable may be less sensitive to detect more pronounced age‐related differences than categorical group comparisons that can accentuate differences. Taken together, these results suggest that while automatic change detection declines with age, involuntary reorientation of attention remains relatively preserved. These findings support the idea that auditory change detection, as indexed by the MMN, may precede changes in preattentional processes indexed by P3a component. Our findings complement those of Yang et al. (2023), who reported that reduced MMN amplitude in older nonautistic adults predicted cognitive decline 1 year later. Although their work focused on later life and cognitive outcomes, our results suggest that automatic change detection decline occurs earlier and in the absence of changes in preattentional mechanisms. Together, these findings support the hypothesis that auditory change detection deficits emerge early and may serve as a precursor to cognitive decline.

Importantly, while we observed a significant age‐related decline in MMN amplitude over frontal site, this decline was similar across groups. Additionally, the absence of age‐related effects on MMN latency and P3a measures was consistent across both groups. These findings suggest that both automatic change detection and preattentional processes remain largely preserved in autistic individuals across adulthood, and that age effect is comparable to that of nonautistic individuals. These results are particularly relevant in regard to the different hypotheses proposed to describe neurocognitive aging in autism, namely the accelerated aging hypothesis, the parallel aging hypothesis and the safeguard hypothesis (Geurts and Vissers 2012; Wang et al. ²⁰²⁴). Our findings support the parallel aging hypothesis, which posits that the aging process in autistic individuals is similar to that of nonautistic individuals. More specifically, at adulthood, they align with the first parallel hypothesis variant, which proposes an absence of group differences across the lifespan, as opposed to the second parallel hypothesis, which suggests initial differences that persist but decline at a similar rate (see Wang et al. ²⁰²⁴ for a full description of these hypotheses). This distinction has important clinical implications. While the second hypothesis pattern might imply earlier vulnerability in aging due to earlier developmental differences, our findings support the view that sensory‐level auditory processes are similarly impacted in autistic and nonautistic individuals to older adulthood. In contrast to the commonly discussed accelerated aging hypothesis, which is often emphasized because of autism's association with increased vulnerability to neuroevolutive conditions (Starkstein et al. 2015; Vivanti et al. ^{2021, 2025}), our findings did not support this view. However, these conclusions should be interpreted cautiously. Our findings apply specifically to autistic adults with relatively high cognitive functioning and do not necessarily generalize to the broader autistic population, particularly individuals with intellectual disability or different adaptive profiles. Further studies including autistic adults with associated IDD are required in order to examine whether these findings may generalize throughout the broader autistic population. While the MMN and P3a components examined are known to decline with age, they may index lower‐level sensory and preattentional processes that are less susceptible to be affected by the neurobiological mechanisms underlying neuroevolutive conditions.

Additionally, as shown by Yang et al. (2023), reduced MMN amplitude in older nonautistic adults predicted cognitive decline 1 year later. The similar MMN decline observed in autistic individuals suggests that this measure could also serve as an early marker of cognitive vulnerability in autism. This possibility highlights the need for further research to determine if MMN can also predict cognitive decline among autistic adults. Additionally, while we observed similar patterns in sensory processing between autistic and nonautistic individuals during EEG, it would be valuable to investigate whether higher‐order cognitive processes exhibit the same pattern.

Finally, these results could offer new insights into resistance to change, a core clinical feature of autism often linked to atypical sensory processing, particularly in the automatic detection of environmental changes (Gomot et al. 2002, 2006, 2011). Our finding of intact and stable MMN responses in autistic adults suggests that, at the sensory level, neural mechanisms underlying change detection do not appear to show significant modulation. This raises the possibility that behavioral resistance to change may also improve in adulthood. However, even in the absence of neural level changes in autistic individuals, behavioral reactivity to change may be more or less impacted over time. Resistance to change is often associated with stereotyped and repetitive behaviors (see Rapaport and Sowman ²⁰²⁴), which are core features of autism, as well as with cognitive and behavioral flexibility (Chung et al. 2012; Green et al. ²⁰⁰⁷). Therefore, exploring how these patterns relate to these clinical and cognitive features could offer important insights into the underlying mechanisms.

4.3. Limitations and Perspectives

Several limitations should be acknowledged in this study. First, the participants were primarily high‐functioning autistic adults, many of whom (77%) were diagnosed with what was previously termed “Asperger's syndrome.” They exhibited high verbal and nonverbal IQ scores, and a majority of them (91%) received their diagnosis in adulthood. Together, these factors limit the generalizability of our findings to the entire autism spectrum, particularly to individuals with co‐occurring intellectual developmental disorder or those requiring higher levels of support and may have contributed to the absence of group differences observed in the present study. The cognitive and procedural demands of the experimental paradigm required participants capable of completing complex tasks and tolerating EEG procedures, meaning the findings remain informative for this studied subgroup, even if they do not generalize to all autistic adults. Second, although we aimed to examine age‐related effects in MMN and P3a components, the age range of our participants was relatively limited. Therefore, our conclusions regarding aging effects should be interpreted with caution. Future studies incorporating older participants would provide a more comprehensive understanding of how automatic change detection and preattentional processes evolve during late adulthood. Another limitation of the present study is the lack of detailed information on participants' sensory profiles. Perceptual differences (hypo vs. hypersensitivity) are commonly reported in autism and could influence MMN responses. Future studies should investigate sensory profiles to examine how individual perceptual differences affect neural change‐detection during aging in autism. Additionally, this study should also be replicated using a more rigorously controlled paradigm in which the standard and deviant stimuli are reversed halfway through the sequence, thus preventing from acoustical effects. An optimal design incorporating other types of deviance—such as duration, which has also been found to be particularly affected in older adults (e.g., see for meta‐analysis: Cheng et al. ²⁰¹³)—could further specify automatic deviance detection in autism and its trajectory across aging. Furthermore, although this study shed light on the neural mechanisms underlying automatic change detection, it does not assess these processes at the behavioral level. Understanding how resistance to change, often linked to particularities in the automatic change detection processes, evolves with age would offer complementary insights. Incorporating behavioral measures, such as the RRB scale (Bourreau et al. 2009) for example, in future research could help fill the gap between neurophysiological findings and everyday behavioral manifestations. Finally, an important limitation of our study is its cross‐sectional design, which prevents conclusions about individual aging trajectories. Longitudinal studies are better suited to capture true aging effects and to account for cohort differences, such as variations in access to care or interventions across generations of autistic adults.

5. Conclusion

The present study contributes to our understanding of neurocognitive aging in autism by showing that automatic change detection and preattentional processes, as reflected by MMN and P3a components, are largely preserved in autistic individuals across adulthood. Our findings reveal no significant differences between autistic and nonautistic individuals in the amplitude or latency of these ERP components across the overall time course of aging. These results support the parallel aging hypothesis, suggesting that neural indexes of auditory processes in autism follow a similar course of aging as nonautistic individuals. By extending research into later stages of life, this study adds to the growing body of literature on aging in autism.

Author Contributions

Marine Bessé: conceptualization, formal analysis, investigation, methodology, writing – original draft, visualization. Shasha Morel‐Kohlmeyer: conceptualization, supervision, writing – review and editing, validation. Emmanuelle Houy‐Durand: conceptualization. Philippe Prévost: conceptualization. Laurie Tuller: conceptualization. Badiâa Bouazzaoui: conceptualization. Laurence Taconnat: conceptualization. Julia Capdeville: conceptualization, formal analysis, writing – review and editing. Lucie Angel: conceptualization, supervision, validation, writing – review and editing. Marie Gomot: conceptualization, supervision, validation, writing – review and editing.

Funding

This work was supported by Région Centre‐Val de Loire (IA‐2021).

Ethics Statement

The study was approved by the Ethics Committee (PROSCEA; IDRCB: 2017‐A00756‐4), and sponsored by the Tours University Hospital (CHU).

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Table S1: Descriptive statistics of demographic, cognitive and clinical variables across young and older autistic and nonautistic groups.

Table S2: Means and standard deviations (SD) of N1 and P2 amplitude (μV) and latency (ms) in younger and older autistic and nonautistic individuals.

Table S3: Means and standard deviations (SD) of MMN and P3a amplitude (μV) and latency (ms) in younger and older autistic and nonautistic individuals.

Bessé, M. , Morel‐Kohlmeyer S., Houy‐Durand E., et al. 2026. “Aging in Autism: An Event‐Related Potentials Study of Auditory Change Detection.” European Journal of Neuroscience 64, no. 8: e70719. 10.1111/ejn.70719.

Data Availability Statement

Due to the terms of our ethics approval, the data supporting this study cannot be made publicly available. Researchers interested in accessing the data should contact the lead author. Access will be provided unconditionally to named individuals, in line with our ethics approval and standard ethical guidelines for data reuse.