Developmental trajectories of prefrontal activation underlying emotional inhibitory control from adolescence to young adulthood: a functional near-infrared spectroscopy study

Abstract

Background

Understanding the neural maturation of emotional inhibitory control (IC) from adolescence to young adulthood is critical but remains incompletely characterized. The present study examined age-related differences in prefrontal activation during emotional conflict processing using functional near-infrared spectroscopy (fNIRS).

Methods

Twenty-six adolescents (ages 13–16, 15 male) and thirty young adults (ages 18–22, 16 male) completed an emotional Flanker task while behavioral responses and prefrontal hemodynamic activity were recorded with fNIRS.

Results

Both groups exhibited a significant emotional Flanker effect, with slower responses on incongruent than congruent trials. fNIRS results suggested age-related differences in prefrontal recruitment patterns. Adolescents showed relatively stronger O₂Hb changes in the right dorsolateral prefrontal cortex (DLPFC), indicating a more right-lateralized response to emotional conflict. In contrast, young adults exhibited a more bilateral pattern of DLPFC activation.

Conclusions

These findings suggest a developmental shift from relatively focal, right-lateralized prefrontal engagement in adolescence to more distributed bilateral recruitment in young adulthood. Such age-related differences may reflect changes in cognitive strategies alongside ongoing maturation of prefrontal networks, with localized right-sided activation in adolescents supporting behavioral performance through compensatory engagement and bilateral recruitment in young adults reflecting greater network integration and neural efficiency.

Background

Adolescents increasingly encounter emotionally salient stimuli—such as peer conflict or distressing media—that challenge their attentional and emotional regulation. Effective adaptation in such contexts relies on inhibitory control (IC), a core executive function that enables selective attention to goal-relevant information while suppressing interference from distracting behavioral, cognitive, or emotional inputs [36]. IC supports higher-order cognitive processes such as reasoning and decision-making [36], and plays a crucial role in promoting psychological well-being by modulating impulsive and maladaptive behaviors [7, 35]. Impairments in IC have been consistently associated with emotion dysregulation and irritability during adolescence [12, 23], as well as with psychiatric disorders characterized by affective lability and behavioral impulsivity, such as obsessive-compulsive disorder (OCD) [35] and attention-deficit/hyperactivity disorder (ADHD) [63]. A precise characterization of IC’s normative developmental trajectory is therefore essential to elucidate its pathological deviations across psychopathology.

A range of well-established experimental paradigms—including the Flanker, Stroop, Go/No-Go, Stop-signal, and Antisaccade tasks—have been widely used to characterize the developmental trajectory of IC from childhood through adulthood [3, 36, 45]. These paradigms typically require individuals to respond rapidly to target stimuli while suppressing prepotent or conflicting responses, thereby providing robust behavioral indices of domain-general IC [37]. Converging evidence supports a protracted, predominantly linear developmental trajectory of IC beginning around age six, reaching maturity in mid-to-late adolescence, and continuing to refine into adulthood—particularly in tasks involving heightened cognitive interference [10, 17]. Importantly, this developmental pattern appears largely independent of individual difference variables such as gender [61] and IQ [65], underscoring the robustness and cross-population generalizability of IC maturation.

In contrast to traditional experimental paradigms that primarily assess IC under emotionally neutral conditions, everyday inhibitory demands are frequently embedded within socioaffective or motivationally salient contexts. For example, individuals may need to resist checking social media to meet a work deadline [17], and adolescents often must regulate romantic or peer-related preoccupations to sustain academic engagement [19]. These considerations have motivated a distinguish between cool IC, which supports regulation in affectively neutral contexts, and hot IC, which governs behavior under emotionally arousing or socially salient conditions [62, 64]. Cool IC is more strongly associated with academic achievement, whereas hot IC better predicts disruptive or maladaptive behaviors [62]. Although the developmental trajectory of cool IC has been extensively characterized, the maturation of hot IC remains theoretically unresolved.

Two primary models have been proposed to characterize the development of hot IC. The linear developmental model posits that hot IC improves with age but follows a more protracted and extended developmental trajectory than cool IC, continuing to change and specialize later in development [15, 98]. Using the Iowa Gambling Task (IGT) as an index of hot IC, both Hooper et al., [50] and Prencipe et al., [74] reported age-related improvements across childhood and adolescence; however, adult-like performance on the IGT was reached later than on cool IC measures, indicating delayed maturation of affectively driven control processes. Evidence from affective stop-signal paradigms further underscores this pattern: when children, adolescents, and young adults performed cool versus hot versions of the SST using neutral or emotional vocal cues, inhibitory performance increased linearly with age in neutral contexts, whereas development in emotionally charged contexts was more protracted, with children and adolescents showing comparable and less efficient inhibition relative to young adults [78]. Complementing these cross-sectional findings, a large-scale longitudinal study tracking 1,657 children across three years from middle childhood to early adolescence revealed substantial heterogeneity and greater instability in hot IC profiles, further suggesting that hot control processes mature more slowly and less uniformly than cool IC [15].

On the other hand, the quadratic developmental model proposes that hot IC follows a non-monotonic, U-shaped trajectory across development, characterized by improving inhibitory performance from late childhood, a mid-adolescent decline (particularly between 14 and 16 years), and subsequent recovery in adulthood. Consistent with this pattern, adolescents exhibit comparatively reduced inhibition in the presence of socioaffectively salient cues relative to children and adults [17, 18, 82]. Evidence from emotional interference paradigms supports this pattern. Studies using emotional Go/No-Go tasks show that adolescents (12–15 years) demonstrate lower accuracy when inhibiting responses to negative emotional cues than emerging adults and adults, suggesting greater difficulty integrating socioemotional information into regulatory control ([28], 2013). Similarly, emotional Stroop findings indicate that emotional IC is weakest during adolescence, whereas children and adults show relatively comparable performance [1, 55]. This developmental pattern may partly explain adolescents’ heightened vulnerability to risk-taking behaviors and affective disorders [19], as behavioral evidence from reward-based decision-making paradigms shows that adolescents engage in greater risk-taking under affectively arousing “hot” conditions [41] and exhibit peak sensitivity to rewards and incentives during adolescence before declining into adulthood [22, 85]. The nonlinear trajectory of hot IC is further consistent with the brain maturation imbalance model, which suggests that heightened incentive responsiveness during adolescence occurs alongside still-maturing top-down prefrontal regulatory systems, resulting in reduced modulation of socioemotional reactivity due to delayed maturation of regulatory circuitry [20].

Evidence supporting both linear and quadratic developmental models has accumulated, and the ongoing debate has prompted researchers to seek clarification by examining the neural systems underlying hot and cool IC. In the brain, IC critically depends on the prefrontal cortex (PFC) [42, 59], with partially dissociable subregions contributing to distinct control processes. Specifically, cool IC is primarily associated with the DLPFC and inferior frontal gyrus (IFG), which support abstract, goal-directed cognitive control, whereas hot IC more strongly engages orbitofrontal (OFC) and ventromedial prefrontal (vmPFC) regions involved in affective valuation and reward processing [4, 77, 97]. However, converging evidence indicates partial functional overlap between the neural systems supporting hot and cool IC. The DLPFC, situated at the apex of the frontal control hierarchy [5, 6], plays a central role in flexibly regulating goal-directed behavior according to contextual demands [48, 53], and thus may operate across both affectively neutral and emotionally salient contexts [32]. Causal evidence further supports this role: transcranial direct current stimulation (tDCS) targeting the DLPFC enhances performance on emotion-related inhibitory tasks (e.g., emotional Stroop, emotional Go/No-Go, and emotional 1-Back) in individuals with major depressive disorder [66].

Emerging evidence further suggests hemispheric specialization within the DLPFC, with the right DLPFC (rDLPFC) frequently implicated in IC processes. Developmental neuroimaging studies of cool IC indicate age-related changes in rDLPFC recruitment across adolescence. For example, longitudinal fMRI data from antisaccade tasks (302 visits, ages 9–26) show that rDLPFC activation decreases with age and reaches adult-like levels by adolescence, reflecting a decelerating trajectory of lateral prefrontal recruitment during development [68]. Complementary magnetoencephalographic evidence also implicates the rDLPFC in response inhibition, showing greater engagement during no-go relative to go trials [48]. Meta-analytic findings further indicate recruitment of rDLPFC and parietal circuits during tasks requiring higher-order stimulus–response associations [80]. Together, these findings underscore the important role of the DLPFC—particularly the rDLPFC—in cool IC, although developmental differences in activation magnitude may reflect changes in neural recruitment patterns rather than straightforward increases in efficiency.

With respect to hot IC, converging evidence also implicates the rDLPFC in the regulation of emotionally salient responses in healthy adults. High-frequency EEG studies (n = 51) indicate that greater rDLPFC tonic activity predicts higher self-reported behavioral inhibition [79]. Functional neuroimaging research further demonstrates rDLPFC involvement in voluntary emotion regulation: increased activation in the rDLPFC and right OFC has been observed during the suppression of sadness [56]; additionally, functional connectivity between the hippocampus and rDLPFC is reduced during emotion suppression [11]. Importantly, causal evidence from brain stimulation studies shows that repetitive transcranial magnetic stimulation (rTMS) targeting the rDLPFC impairs the inhibition of negative emotional stimuli [58] and reduces the efficiency of emotional face reappraisal in healthy adults [93, 94], underscoring the causal contribution of the rDLPFC to hot IC.

Despite accumulating evidence linking the rDLPFC to emotional IC, little is known about how its function changes across development. In particular, the transition from adolescence to early adulthood—a key period marked by profound neurocognitive reorganization—remains underexplored. Adolescence is characterized by substantial maturation of prefrontal–limbic circuitry and increasing integration between cognitive control and affective processing systems [13, 14, 67]. Consequently, the refinement of emotional IC during this period likely reflects ongoing developmental changes in these regulatory networks and may follow trajectories that differ from those observed in cool IC. However, the developmental dynamics of rDLPFC engagement during emotional IC remain largely unclear.

To address this gap, the present study examined the developmental trajectory of emotional IC from adolescence to early adulthood using a modified emotional Flanker task designed to elicit interference under affectively salient conditions. Prefrontal hemodynamic activity was recorded using fNIRS, a non-invasive and motion-tolerant neuroimaging technique well suited for developmental populations [2, 73]. By integrating behavioral measures with fNIRS indices of prefrontal activation, this study aimed to determine whether age-related improvements in emotional IC are accompanied by systematic changes in rDLPFC recruitment.

Methods

Participants

To determine whether the sample size was sufficient to detect between-group differences in the emotional Flanker effect, an a priori power analysis was conducted using G*Power (version 3.0) for both behavioral and fNIRS data. For behavioral outcomes, a two-way mixed-design repeated measures ANOVA was planned, with task congruency (congruent vs. incongruent) as the within-subject factor and group ((adolescents vs. young adults) as the between-subject factor. The parameters were set as follows: α = 0.05, power = 0.80, effect size f = 0.25 (medium; based on [27, 40, 54]), two groups, and two repeated measures, with the correlation among repeated measures set to 0.5. This analysis indicated that a minimum total sample size of N = 34 (17 per group) would be required. For fNIRS data, considering the hierarchical characteristics of neural signals, separate power analyses were conducted at the channel and ROI levels. Channel-level analysis based on a three-way mixed-design repeated-measures ANOVA (congruency × channel × group) suggested that 12 participants were sufficient, whereas ROI-level analysis (group × hemisphere × region) indicated that 24 participants were required.

A total of 26 adolescents (M = 14.15 years, SD = 0.92; 15 males) were recruited from Beijing Chenjinglun Secondary School, which includes both junior and senior high school levels. Additionally, 30 young adults (M = 19.33 years, SD = 1.47; 16 males) were recruited from several universities in Beijing via online postings and flyers. Demographic information—including age, gender, full-scale IQ, ethnicity, language, household income, and maternal education—was collected. The experiment was conducted in a quiet, private room on the school campus for adolescents, and in comparably arranged rooms on university campuses for young adults. Importantly, the actual sample size of the present study exceeded the minimum requirements estimated by the a priori power analysis, was comparable to that reported in similar fNIRS studies [24, 71, 95], and was further supported by methodological considerations, including averaging signals across 48 event-related segments per condition to obtain stable hemodynamic estimates and applying Bonferroni correction to control family-wise Type I error inflation.

All participants were native Chinese speakers with normal hearing and normal or corrected-to-normal vision. Cognitive functioning was assessed using the Wechsler Abbreviated Scale of Intelligence, Second Edition (WASI-II) [92]. Individuals with a full-scale IQ below 80 were excluded. Additional exclusion criteria included a history of seizures, psychiatric or neurological conditions, head injury involving loss of consciousness exceeding 30 min, brain tumors, or current use of psychoactive medications or illicit substances. Written informed consent was obtained from adult participants and the parents of adolescent participants, with adolescents providing written informed assent. Participants were compensated for their time.

Stimuli and apparatus

In this study, we employed an event-related emotional Flanker task that integrated emotional face stimuli with the classic Flanker paradigm, following the experimental protocol established in our previous research (Fig. 1) [24]. Twelve black-and-white photographs of four Asian models (two males and two females) were selected from the NimStim face stimulus set (http://www.macbrain.org/resources.htm) based on their high interrater agreement (76–100%) for expression valence. Each model displayed three universal facial expressions: happy, fearful and neutral. Each photograph was combined with distractor faces across eight configurations, based on the congruency between the centrally positioned target face and the flanking distractor faces, resulting in a total of 96 experimental trials.

Fig. 1.

Experimental design and stimuli. Examples of congruent and incongruent trials in the modified emotional Eriksen Flanker task are shown

The visual stimuli were displayed centrally on an 18.5-inch in-plane switching (IPS) monitor with a resolution of 1920 × 1080 pixels and a refresh rate of 100 Hz. As described earlier, each stimulus featured a target face at the center flanked by two distractor faces on either side. The size of each face image was 512 × 768 pixels. The combined stimuli photograph subtended a visual angle of approximately 17.06° horizontally × 7.63° vertically. Stimuli presentation and behavioral data collection were conducted using E-Prime software (Psychology Software Tools, Pittsburgh, PA).

Behavioral procedure

A total of 96 experimental trials were divided into two sessions, each consisting of 48 trials: 24 congruent and 24 incongruent trials. Each trial began with a white cross (‘+’) presented at the center of the screen, subtending 0.2° of visual angle, which remained for 500 ms. This was followed by the emotional face stimuli, which appeared for 2000 ms or until a response was made. Participants were instructed to judge whether the emotion displayed on the centrally located face was congruent with the emotion shown on the laterally positioned face. Responses were recorded by pressing the ‘F’ or ‘J’ keys with the forefinger, corresponding to ‘yes’ or ‘no’ answers.

Trials were separated by a 9–12 s inter-trial interval, during which a black screen was displayed to prevent participants from predicting the timing of subsequent trials. The left-right order of the response buttons was counterbalanced across participants. Both the accuracy rate and reaction time were recorded for each trial. The order of trials within each session was randomized.

All participants were tested individually in a quiet room to minimize distractions. At the beginning of the experiment, the instructions were presented and clarified for the. Participants then completed a practice block, including both congruent and incongruent trials, with feedback provided for incorrect responses. Participants had to achieve over 80% accuracy in the practice trials to proceed to the main experiment. The main trials were randomized, and participants were informed that they would receive compensation upon successful completion of the study.

fNIRS data acquisition

An eight-channel continuous-wave fNIRS system (Octamon, Artinis Medical Systems, The Netherlands) was used to measure changes in the optical density of oxygenated hemoglobin (O₂Hb) and deoxygenated hemoglobin (HHb) across the bilateral PFC. This system utilized dual-wavelength laser light (760 nm and 850 nm) emitted from each source optode at a sampling frequency of 10 Hz. Eight optical fiber bundles (transmitters) delivered light to the left and right PFC (four fibers for each hemisphere), while two separate optical fiber bundles (receivers) captured the emerging light from the same cortical regions. Each measurement channel consisted of a source-detector (Tx–Rx) pair, with a fixed inter-optode distance of 35 mm.

As shown in Fig. 2, the optodes were bilaterally arranged over the prefrontal cortex. Specifically, Channels 6 and 4 were positioned at the Fp1 and Fp2 locations, respectively, according to the International 10–20 EEG electrode system. Because individual MRI scans or 3D digitizing systems were not available, the fNIRS data were spatially registered to the Montreal Neurological Institute (MNI) standard brain space using a physical model of the head surface based on the ICBM152 template—the standard brain template widely used in neuroimaging studies [81].

Fig. 2.

Optode arrangements for fNIRS in 3D space. Eight transmitter optodes and two receiver optodes were configured to form a total of eight measurement channels. As highlighted by the red double-headed arrows, channels in the right hemisphere included Channel 1 (S1–R1), Channel 2 (S2–R1), Channel 3 (S3–R1), and Channel 4 (S4–R1); channels in the left hemisphere included Channel 5 (S5–R2), Channel 6 (S6–R2), Channel 7 (S7–R2), and Channel 8 (S8–R2). All optodes were positioned over the prefrontal cortex according to the International 10–20 EEG coordinate system

Singh et al., [81] employed a probabilistic registration approach for healthy adults (aged 24 to 39 years) to estimate the cortical projection (CP) points corresponding to all channel and optode locations in the prefrontal region, based on the 10–20 system. The standard deviations of estimated CP points within this region ranged from 4.7 to 7.0 mm, indicating that the spatial registration error falls within an acceptable tolerance for group-level analyses. Consequently, this method enables multi-subject fNIRS analyses and population-level inferences even without individual MRIs. The estimated MNI coordinates of all optodes are shown in Fig. 2.

To refine the anatomical localization of the measurement sites, the channels were projected onto the Brodmann area (BA) template using xjView software (http://www.alivelearn.net/xjview/). The corresponding BAs and cortical regions for the eight channels were as follows: right BA46 (Channels 1 and 3, right DLPFC), left BA46 (Channels 5 and 7, left DLPFC), right BA11 (Channels 2 and 4, right OFC), and left BA11 (Channels 6 and 8, left OFC). During data acquisition, changes in O₂Hb and HHb concentrations were displayed in real time, and signal quality was continuously monitored to minimize motion artifacts and ensure data integrity.

fNIRS data analysis

The fNIRS data analysis outlined here follows the procedures established in a previous study [24]. Raw optical density data were acquired using Oxysoft software (version 3.2.70, Artinis Medical Systems, 2016) and subsequently imported into MATLAB 2013a using the oxysoft2matlab function. The data were then processed using the NIRS-KIT toolbox [52]. First, the raw optical density data were converted into concentration changes for O₂Hb and HHb using the modified Beer–lambert law, with an age-dependent differential pathlength factor (DPF = 4.99 + 0.067*Age^0.814) [26]. Next, the fNIRS data underwent a series of preprocessing steps to minimize the influence of noise and artifacts from exogenous and endogenous sources. Specifically, a first-order polynomial regression model was employed to remove linear detrends from the raw time course. The temporal derivative distribution repair (TDDR) method was then applied to correct motion artifacts. In addition, a bandpass filtering (0.01 to 0.08 Hz) with a finite impulse response (FIR) filter was used to filter out machinery and physiological noise (e.g., respiration, heart rate, Mayer wave, and low-frequency oscillations).

After preprocessing, individual-level analysis was performed to detect task-evoked neural activation at the channel level using the mass univariate statistical method based on General Linear Model (GLM) [8]. This individual-level statistical analysis consisted of the following steps: (1) specification of the GLM, where the canonical hemodynamic response function (HRF), implemented in SPM, was used to construct the reference time series for task-releted variables; (2) estimation of GLM parameters on a channel-by-channel basis, yielding activation β-values (weight coefficients) for each experimental condition; and (3) computation of the condition-wise effects of interest using user-defined contrast vectors for group-level analysis. At the group level, paired t-tests were performed to obtain population-level inferences. The NIRS-KIT toolbox [52] was employed to interpolate and generate channel-wise 2D visualizations, while the EasyTopo toolbox [89] was used to create channel-wise 3D visualizations.

To further visualize the temporal characteristics of task-evoked hemodynamic responses, 48 segments of event-related fNIRS signals were extracted and averaged for each condition. In the present study, trials were separated by a 9–12 s inter-trial interval to minimize anticipation, which resulted in variable event durations ranging from 11 to 14 s. For consistency in plotting, the analysis window was standardized to the minimum duration of 11 s. For both congruent and incongruent conditions, time-course waveforms were derived for each of the eight channels by averaging the event-related signals across participants within each group (adolescent or young adult), providing a clear depiction of the temporal dynamics of condition-specific hemodynamic responses.

Statistical analysis

First, mean accuracy and reaction time (RT) were calculated across all trials and participants. RTs were calculated only for correct trials, with outliers exceeding ± 3 SD from each participant’s mean excluded. The normality of RT distributions was assessed using the Shapiro–Wilk test, and group homogeneity of variances was evaluated with Levene’s test to determine whether parametric tests were appropriate. If the behavioral data met assumptions of normality and homogeneity, a two-way mixed-design repeated measures ANOVA was conducted, with task congruency (congruent vs. incongruent) as the within-subject factor and group (adolescence vs. adulthood) as the between-subject factor. Significant main effects or interactions were followed by paired t-tests (within-group) or independent-samples t-tests (between-group), as appropriate. Bonferroni correction was applied to adjust for multiple comparisons unless otherwise specified. Effect sizes were reported as partial η² for ANOVAs and Cohen’s d for t-tests.

Next, PFC activation patterns were examined for each group under both congruent and incongruent conditions. For each participant, general linear model (GLM) parameters were estimated on a channel-by-channel basis, yielding β-values corresponding to each condition. The β-values for the eight channels were extracted from the NIRS-KIT–processed .mat files and imported into SPSS 20.0 (SPSS Inc., Chicago, IL) for statistical analysis. When the assumptions of normality and homogeneity were met, a three-way mixed-design repeated-measures ANOVA was performed, with task congruency (congruent vs. incongruent) and channel (CH1 ~ CH8) as within-subject factors, and group (adolescents vs. young adults) as the between-subject factor. Post hoc paired t-tests were conducted to compare conditions within each group, and Bonferroni correction was applied across the eight channels to control for Type I error inflation.

Finally, Pearson correlation coefficients were calculated to assess relationships between behavioral performance and task-related hemoglobin variables within each group. When normality or linearity assumptions were violated, Spearman’s rank correlation was used instead. All statistical tests were two-tailed, with statistical significance set at α = 0.05. Data are reported as mean ± SD. All group-level analyses were performed in SPSS 20.0.

Results

Demographic characteristics

Demographics of all participants are summarized in Table 1. Initially, the Shapiro–Wilk test was conducted to assess the normality of age and IQ. Results showed that the IQ scores were normally distributed (p > .05), permitting the use of independent samples t-tests. In contrast, age did not follow a normal distribution (p < .001), so the Mann–Whitney U test was applied instead. Independent samples t-tests (assuming equal variances based on Levene’s test; p > .05), revealed no significant differences in IQ between the two groups. Meanwhile, the Mann–Whitney U test indicated a significant difference in age between the groups, which aligns with the study design, as the participants were intentionally recruited from two distinct age stages.

Table 1.

Demographic characteristics of the adolescent and young adult group

Variables	Adolescent group (n = 26)	Young adult group (n = 30)	Test statistic	p
Age (years)	14.15 ± 0.92	19.33 ± 1.47	U = 0.000; Z = -6.470	< 0.001
Gender (male/female)	15/11	16/14	χ² (1, N = 56) = 0.107	0.743
Full Scale IQ	108.27 ± 5.20	110.30 ± 5.47	t (54) = -1.419	0.162
Ethnicity (%)	Asian (100%)	Asian (100%)
Language (%)	Mandarin Chinese (100%)	Mandarin Chinese (100%)
Site of Testing (%)	Beijing, urban middle school campus (100%)	Beijing, urban university campus (100%)
Household Income (¥/month)			χ² (2, N = 56) = 5.161	0.076
Low (≤ 8,000)	1 (3.8%)	6 (20.0%)
Middle (8,001–20,000)	13 (50.0%)	17 (56.7%)
High (> 20,000)	12 (46.2%)	7 (23.3%)
Maternal Education			χ² (2, N = 56) = 4.738	0.094
Below bachelor’s degree	4 (15.4%)	12 (40.0%)
Bachelor’s degree	12 (46.2%)	12 (40.0%)
Postgraduate degree	10 (38.5%)	6 (20.0%)

IQ  Intelligence Quotient. Continuous variables are presented as M ± SD; categorical variables as n (%). Mann-Whitney U test was used for non-normally distributed data (Age); chi-square (χ²) tests were used for categorical variables (Gender, Income and Education); independent t-tests was used for normally distributed continuous variables (IQ). p < .05

Subsequently, Pearson χ² tests were used to examine differences in gender, household income, and maternal education. Household income was categorized into three levels—Low (≤ ¥8,000), Middle (¥8,001–20,000), and High (> ¥20,000)—based on the economic classifications commonly used in mainland China, reflecting conditions across first-, second-, and third-tier cities. In the Chinese educational context, mothers are typically the primary caregivers and are more actively involved in children’s daily education and development. Therefore, maternal education was used as the indicator of socioeconomic status (SES) in this study. Results from the χ² tests indicated no significant group differences in gender, household income, or maternal education between the adolescent and young adult groups.

Behavioral performance

First, the Shapiro–Wilk test was conducted to evaluate the normality of the distribution of accuracy and mean RT under two conditions (congruent vs. incongruent). The results indicated that mean RT in both conditions followed a normal distribution (Shapiro–Wilk test; ps > 0.05) (Fig. 3A), leading to the application of Levene’s test to assess homogeneity of variances, followed by a 2 (group: adolescence vs. adulthood) × 2 (congruency: congruent vs. incongruent) repeated measures analysis (ANOVA). In contrast, the accuracy data did not meet the assumption of normality, as the Shapiro–Wilk test rejected the null hypothesis (ps < 0.05) (Fig. 3B). Consequently, the Wilcoxon signed-rank test was used to compare the two task conditions within each group.

Fig. 3.

Behavioral performance on the Emotional Flanker task. Violin plots showing the distribution of mean RT (A) and accuracy (B) under congruent and incongruent conditions. Comparisons of mean RT (C) and accuracy (D) between congruent and incongruent conditions in the adolescent and young adult groups. * p < .05, ** p < .01

For mean RT, Levene’s test indicated that the two groups exhibited homogeneity of variances. ANOVA results showed a significant main effect of congruency, F (1, 54) = 53.756, p < .001, η_p² = 0.499, with faster mean RT for congruent trials (M = 898 ms) compared to incongruent trials (M = 966). The main effect of group and the interaction of group and congruency were not significant, F (1, 54) = 2.851, p = .097, η_p² = 0.050, and F (1, 54) = 3.166, p = .081, η_p² = 0.055, respectively. To explore whether the magnitude of the emotional Flanker effect differed across groups, we compared the RTs between congruent and incongruent conditions for adolescents and young adults. Paired-samples t tests revealed a significant difference between congruent (M = 865 ms, SD = 22.24) and incongruent conditions (M = 949, SD = 24.01) in the adolescent group (t ₍₂₅₎ = -6.449, p < .001, Cohen’s d = 0.69). Similarly, a significant difference between congruent (M = 932 ms, SD = 20.70) and incongruent conditions (M = 984, SD = 22.35) was also found in the young adult group (t ₍₂₉₎ = -3.959, p < .001, Cohen’s d = 0.42). These findings indicate that an emotional Flanker effect was observed in both groups (Fig. 3C).

For accuracy, the Wilcoxon signed-rank tests were conducted to compare performance between congruent and incongruent conditions within each group. In the adolescent group, accuracy was significantly higher for congruent (M = 0.91, SD = 0.08) than for incongruent trials (M = 0.85, SD = 0.08) (Z = -2.748, p = .006, r = .54), indicating a large effect size. Similarly, in the young adult group, accuracy was higher for congruent (M = 0.94, SD = 0.07) and for incongruent trials (M = 0.91, SD = 0.06) (Z = -2.411, p = .016, r = .44), indicating a moderate to large effect size. These results demonstrate that both groups exhibited reduced accuracy in the incongruent condition, reflecting a robust emotional Flanker effect across age groups (Fig. 3D).

fNIRs results

Consistent with previous findings, O₂Hb signals demonstrated greater sensitivity to changes in cerebral blood flow [87], better signal-to-noise ratio [87, 96], and higher intraclass correlation coefficients compared with HHb signals [60]. A similar response pattern was observed in the present study (Fig. 4). To visualize the temporal dynamics of task-evoked hemodynamic responses, 48 event-related fNIRS segments were extracted and averaged for each condition. For consistency across participants, the analysis window was standardized to the minimum event duration of 11 s.

Fig. 4.

Averaged O₂Hb and HHb time courses across eight channels, presented for the congruent condition as a representative example, showing the canonical increase in O₂Hb and decrease in HHb in both adolescent and young adult groups

As illustrated in Fig. 4, the congruent condition elicited a canonical hemodynamic response pattern, characterized by a marked decrease in HHb and an approximately threefold increase in O₂Hb across all eight channels in response to the emotional Flanker task. Given the higher stability and reliability of O₂Hb signals, subsequent statistical analyses were therefore focused on O₂Hb data.

To examine the neural substrates underlying cognitive processing, β-values (regression coefficients) each of the eight channels under the congruent and incongruent conditions were extracted from NIRS-KIT–processed .mat files. The Shapiro–Wilk test confirmed normal distributions of β-values across all channels (ps > 0.05) (Fig. 5A), and Levene’s test indicated homogeneity of variance between groups (ps > 0.05). Therefore, a 2 (group: adolescence vs. adulthood) × 2 (congruency: congruent vs. incongruent) × 8 (channel: CH1 ~ CH8) mixed-design repeated-measures ANOVA was performed. The analysis revealed a significant main effect of congruency, F (1, 54) = 8.47, p < .01, η_p² = 0.14, a significant interaction of congruency × channel, F (7, 378) = 4.54, p < .005, η_p² = 0.08, and a significant interaction of group × congruency × channel, F (7, 378) = 2.54, p < .05, η_p² = 0.05. No other main effects or interactions were significant.

Fig. 5.

fNIRS results for the Emotional Flanker task. A Violin plots showing the distribution of β-values of O₂Hb under congruent and incongruent conditions across the eight channels. B β-values of O₂Hb for the incongruent versus congruent condition across the eight channels in the adolescent group. Channels 1 and 3 in the right PFC showed greater activation for incongruent trials compared to congruent trials. C β-values of O₂Hb for the incongruent versus congruent condition across the eight channels in the young adult group. Channels 1 and 3 in the right PFC, as well as Channels 5 and 7 in the left PFC, showed greater activation for incongruent than congruent trials. * p < .05 ** p < .01

Because the three-way interaction between congruency, group, and channel was significant, we performed post-hoc paired contrasts [incongruent–congruent] for both the adolescent and young adult groups across the eight channels to further examine the effects of congruency at each channel. For the adolescent group (Fig. 5B), significant differences between incongruent and congruent conditions were found for two measurement points in the right hemisphere (CH1: t ₍₂₅₎ = -4.404, p = .000; CH2: t ₍₂₅₎ = -1.319, p = .199; CH3: t ₍₂₅₎ = -3.219, p = .004; CH4: t ₍₂₅₎ = -1.406, p = .172), but no significant differences were observed in the left hemisphere (CH5: t ₍₂₅₎ = -1.258, p = .220; CH6: t ₍₂₅₎ = -0.091, p = .928; CH7: t ₍₂₅₎ = 0.447, p = .658; CH8: t ₍₂₅₎ = -0.253, p = .802). For the young adult group (Fig. 5C), significant differences were observed at two measurement points in both the right hemisphere (CH1: t ₍₂₉₎ = -2.015, p = .050; CH3: t ₍₂₉₎ = -2.312, p = .028) and the left hemisphere (CH5: t ₍₂₉₎ = -3.773, p = .001; CH7: t ₍₂₉₎ = -3.174, p = .004), but no significant differences were found for other points (CH2: t ₍₂₉₎ = 0.501, p = .620; CH4: t ₍₂₉₎ = 0.209, p = .836; CH6: t ₍₂₉₎ = 0.400, p = .692; CH8: t ₍₂₉₎ = 0.718, p = .478).

Furthermore, based on the cortical regions identified, the following regions of interest (ROIs) were defined: right BA46 (Channels 1 and 3, right DLPFC), left BA46 (Channels 5 and 7, left DLPFC), right BA11 (Channels 2 and 4, right OFC), and left BA11 (Channels 6 and 8, left OFC). The β contrasts [incongruent – congruent] were computed separately for the adolescent and young adult groups across the four ROIs to further examine potential hemispheric lateralization effects. Therefore, a 2 (group: adolescence vs. adulthood) × 2 (hemisphere: right vs. left) × 2 (region: BA46 vs. BA11) mixed-design repeated-measures ANOVA was conducted. The analysis revealed a significant main effect of region, F (1, 54) = 26.46, p < .001, η_p² = 0.33, as well as significant interactions of group × region, F (1, 54) = 6.24, p < .05, η_p² = 0.10, and hemisphere × region, F (1, 54) = 4.94, p < .05, η_p² = 0.07. Importantly, a three-way interaction of group × hemisphere × region was significant, F (1, 54) = 2.56, p < .05, η_p² = 0.04. No other main effects or interactions reached significance.

Post-hoc t-tests revealed a significant difference between the right and left BA46 in the adolescent group, t ₍₂₅₎ = 2.880, p = .008, whereas no significant difference was observed between the right and left BA11, t ₍₂₅₎ = 1.158, p = .258. In contrast, for the young adult group, neither the right–left BA46 nor the right–left BA11 comparisons reached significance (ps > 0.05). These findings suggest that adolescents exhibited greater right-lateralized activation within DLPFC (Fig. 6), whereas young adults showed a more balanced, bilateral activation pattern across both DLPFC (Fig. 7).

Fig. 6.

Cortical activation during the Emotional Flanker task in the adolescent group. Upper: Channel-wise 3D visualization. Lower: Channel-wise 2D visualization. T-maps of β-values for O₂Hb showing the emotional Flanker effect ([incongruent − congruent] contrast), based on GLM β-value contrasts

Fig. 7.

Cortical activation during the Emotional Flanker task in the young adult group. Upper: Channel-wise 3D visualization. Lower: Channel-wise 2D visualization. T-maps of β-values for O₂Hb showing the emotional Flanker effect ([incongruent − congruent] contrast), based on GLM β-value contrasts

Correlations between activation and behavioral performance

Two-tailed Pearson’s correlation analyses were conducted to examine the relationship between [incongruent-congruent] O₂Hb β-value contrasts at channels 1, 3, 5, and 7 and the corresponding [incongruent-congruent] RT differences in both the adolescent and young adult groups.

In the adolescent group, significant positive correlations were observed at Channels 1 (r (26) = 0.445, p = .023) and 3 (r (26) = 0.536, p = .005), whereas Channels 5 (r (26) = -0.193, p = .344) and 7 (r (26) = 0.301, p = .135) showed no significant correlations (Fig. 8A). In the young adult group, significant positive correlations were found at Channels 1 (r (30) = 0.492, p = .006), 3 (r (30) = 0.497, p = .005) and 5 (r (30) = 0.594, p = .001), while Channel 7 showed a marginally significant correlation (r (30) = 0.351, p = .058) (Fig. 8B). These results indicate a significant association between subjects’ behavioral performance and asymmetric PFC activations across both groups.

Fig. 8.

Scatterplots showing the relationship between [incongruent-congruent] ΔRT and ΔO₂Hb β-values. A In the adolescent group, ΔRT showed a significant positive correlation with ΔO₂Hb β-values at Channels 1 and 3 in the right DLPFC. B In the young adult group, ΔRT showed a significant positive correlation with ΔO₂Hb β-values at Channels 1 and 3 in the right DLPFC, and Channels 5 and 7 in the left DLPFC

Discussion

The present study investigated the developmental trajectory of emotional IC from adolescence to young adulthood using an emotional Flanker task. Behaviorally, both groups exhibited a robust emotional Flanker effect, with slower responses and lower accuracy on incongruent trials relative to congruent trials. At the neural level, fNIRS data indicate possible age-related differences in prefrontal recruitment, with adolescents exhibiting relatively stronger right-lateralized DLPFC activation and young adults showing comparatively more bilateral engagement. These findings appear more consistent with a linear developmental model of hot inhibitory control, in which the core behavioral abilities are established by adolescence but the underlying prefrontal networks continue to mature and become more integrated and efficient into young adulthood. This pattern highlights developmental differences in the recruitment and organization of prefrontal control systems during emotional conflict processing.

Comparable emotional Flanker effects across age groups

Both adolescents and young adults exhibited a robust emotional Flanker effect, consistent with previous findings from studies using emotional Flanker, [28, 29, 51] and Stroop tasks [1]. Incongruent trials, in which the emotional valence of flanker stimuli conflicted with that of the target, elicited greater cognitive interference. This interference was likely amplified by the automatic processing of emotional facial expressions [88], resulting in slower responses and reduced accuracy relative to congruent trials [51]. Such performance costs are consistent with theories of emotion–cognition interaction, which propose that emotionally salient stimuli receive prioritized processing and therefore compete with task-relevant information for attentional resources. Importantly, the presence of a robust emotional Flanker effect in both age groups suggests that the basic mechanisms supporting emotional conflict detection and resolution are already present by adolescence, although these processes may continue to undergo refinement into early adulthood.

Neural mechanisms underlying developmental differences

Adolescence is a transitional developmental period marked by substantial changes in social behavior and the gradual emergence of mature cognitive abilities. These advances are closely linked to the ongoing maturation of neural systems supporting motivation, emotion, and cognitive control, with the PFC exhibiting particularly protracted development [13, 14, 16, 67]. Longitudinal MRI studies indicate that gray matter volume in the frontal cortex peaks around early adolescence (approximately age 12) and subsequently undergoes substantial structural remodeling through adulthood. This remodeling reflects intra-frontal reorganization and continued fine-tuning processes, including synaptic pruning and myelination, which enhance the coherence and efficiency of prefrontal circuitry and support the progressive emergence of effective cognitive and emotional regulation [44, 47, 83]. During this period, heightened emotional reactivity further increases demands on top-down control systems, while the progressive strengthening of intra-frontal connectivity facilitates the stabilization and coordination of higher-order executive functions [75].

Within this developmental framework, the right PFC has been proposed to play an important role in integrating cognitive and emotional control. Structural MRI findings indicate that the right PFC exhibits the greatest predictive power for brain maturity, underscoring its importance in higher-order reasoning and regulatory processes [38]. Converging clinical evidence supports this specialization: greater cortical thickness in the right PFC predicts better response to cognitive-behavioral therapy in children with OCD [9, 84], whereas adolescents with emotional dysregulation exhibit both reduced gray matter volume and hyperactivation in the right orbitofrontal cortex during inhibitory control tasks [86]. Causal evidence from neuromodulation further underscores the rDLPFC’s key role in emotional regulation—rTMS over this region disrupts inhibition of negative emotions [58] and impairs emotional reappraisal efficiency [93, 94], highlighting its essential function in top-down control under high regulatory demands [32, 33]. Moreover, developmental fMRI studies reveal that prefrontal activation becomes increasingly bilateral with age during tasks involving verbal working memory and long-term memory, reflecting a shift toward more distributed and compensatory processing as the PFC functionally matures [69].

In the present study, the emotional Flanker task simultaneously engages executive control and emotional regulation processes [70], thereby imposing higher demands on the PFC. Despite these task characteristics, the overall activation pattern was broadly consistent with previous developmental findings [69]. Specifically, adolescents showed significant O₂Hb changes predominantly in the right DLPFC (CH1 and CH3), whereas young adults exhibited significant bilateral activation across the DLPFC, including channels in the left hemisphere (CH5 and CH7). While this overall pattern aligns with prior developmental work, a nuanced interpretation is warranted. In particular, multiple levels of explanation should be considered, ranging from physiological constraints inherent to fNIRS signal interpretation to the neurobiological mechanisms underlying cognitive maturation.

Developmental differences in vascular responsivity represent an important physiological consideration when interpreting fNIRS findings. The fNIRS signal depends on neurovascular coupling, yet evidence suggests that both peripheral and cerebral vascular responses differ between adolescents and adults. When exposed to emotionally salient contexts, adolescents have been shown to exhibit more pronounced vasoconstrictive responses and longer recovery times before returning to baseline [25]. Such a pattern—characterized by an initial constrictive response followed by a delayed or attenuated dilatory phase—may have implications for hemodynamic-based neuroimaging signals. If a similar vascular profile occurs in the PFC during cognitively or emotionally demanding tasks, the expected task-related increase in oxyhemoglobin could be partially attenuated. Consequently, the absence of significant left DLPFC activation in adolescents may partly reflect developmental differences in regional vasodilatory responses rather than purely neural differences.

Beyond these physiological considerations, the observed shift in activation patterns may also reflect ongoing neurodevelopmental maturation of the PFC. Developmental neuroimaging studies have consistently documented protracted PFC maturation extending from adolescence into young adulthood [14]. During this period, the PFC undergoes substantial synaptic pruning, which refines neural circuitry and stabilizes synaptic connectivity [14, 72]. Such structural reorganization may provide the anatomical basis for developmental differences in DLPFC recruitment during emotional interference processing. In particular, adolescents may rely more strongly on right-lateralized control processes, whereas young adults appear to recruit a broader and more bilateral prefrontal network. Consistent with this view, accumulating evidence suggests that functional PFC networks continue to strengthen and reorganize throughout development into young adulthood, and the coordinated activation patterns observed in the emotional Flanker task may reflect this ongoing developmental refinement [21].

In addition to these structural changes, maturational alterations in the prefrontal dopamine system may further shape the functional organization of prefrontal control networks. During late adolescence, D2 receptor signaling becomes increasingly involved in modulating inhibitory activity within prefrontal circuits. Whereas dopamine regulation in the immature PFC is primarily mediated by D1 receptors, the mature system engages both D1 and D2 receptors to coordinate PFC activity. This dual-receptor mechanism is thought to enhance the precision and efficiency of information processing, thereby supporting flexible adaptation to changing task demands [16, 57]. The bilateral recruitment of the DLPFC observed in young adults during emotional conflict processing may therefore reflect the increased computational capacity of a more mature prefrontal dopamine system, whereas adolescents—whose dopaminergic regulation of prefrontal circuits is still undergoing refinement—may rely on a more limited, right-lateralized network to achieve comparable behavioral performance.

Beyond these neurobiological mechanisms, age-related differences in neural recruitment may also reflect differences in the cognitive strategies used to regulate emotional interference. Developmental research suggests that adolescents and adults may extract different types of information from emotionally salient stimuli [14]. Adolescents tend to rely more heavily on featural processing when responding to emotional cues and are generally more prone to risky decision-making in “hot” contexts. Such processing styles may be preferentially supported by right-lateralized prefrontal engagement. In contrast, young adults appear to employ more strategic, task-dependent processing during emotional conflict tasks [30, 31, 46]. Bilateral engagement of the DLPFC may therefore facilitate the integration of emotional and executive information required for flexible cognitive control.

Developmental accounts of behavioral–neural dissociation

Building on the multi-level mechanisms discussed above, the similar behavioral outcomes observed across age groups may arise from distinct neural recruitment patterns. This raises important questions about how behavioral performance and neural activation become dissociated across development. Behaviorally, both groups exhibited robust emotional Flanker effects, indicating that emotional conflict processing is operative across adolescence and early adulthood [14]. However, this behavioral continuity appeared to be supported by different patterns of neural recruitment, suggesting ongoing developmental reorganization of prefrontal control systems.

Adolescents in the present study exhibited relatively stronger activation in the right DLPFC during emotional conflict processing, a region consistently implicated in inhibitory control in both human and non-human primate research [43, 49]. Contemporary models of prefrontal organization propose that frontal control networks operate within a hierarchical architecture, in which more rostral regions are recruited as task demands increase [6]. Consistent with this view, meta-analytic evidence indicates that the DLPFC is particularly engaged in tasks requiring flexible updating of stimulus–response associations and complex inhibitory control [6, 90, 91]. Emotional Flanker tasks impose greater regulatory demands than standard arrow-based Flanker paradigms because emotional faces automatically capture attention and introduce affective interference. Given adolescents’ heightened sensitivity to emotionally salient contexts, resolving such interference may require greater reliance on prefrontal control mechanisms. Within this developmental context, the right-lateralized recruitment observed in adolescents may therefore reflect a form of neural compensation, whereby localized engagement of the right DLPFC helps maintain behavioral performance despite the ongoing immaturity of distributed prefrontal control networks.

In contrast, the more bilateral DLPFC engagement observed in young adults may reflect increasing neural efficiency as prefrontal networks mature. Neural efficiency models propose that comparable behavioral performance can be achieved with fewer localized neural resources when processing becomes more efficient. Importantly, such efficiency does not necessarily imply uniformly reduced activation, but may instead reflect more coordinated interactions across distributed brain regions [34]. Consistent with this view, neuroimaging studies have shown that individuals with faster processing speed may exhibit reduced activation in some prefrontal regions but increased activity in others, reflecting more direct and efficient processing links between neural nodes that support faster and more accurate performance [76]. Moreover, when task difficulty is adjusted to match person-specific capacity, differences in brain activation between individuals are greatly reduced, suggesting that neural efficiency reflects an adaptive allocation of neural resources to task demands [39]. Within this framework, the bilateral DLPFC recruitment observed in young adults may reflect a more integrated prefrontal control system that enables efficient coordination between emotional and executive processes during emotional conflict resolution.

Limitations

Several methodological considerations should be acknowledged when interpreting the present findings. First, the spatial localization of fNIRS channels was estimated using a probabilistic registration approach based on the international 10–20 system rather than individual MRI scans or digitized optode positions. Consequently, the mapping between channels and specific cortical regions should be interpreted with caution, as this approach introduces a degree of spatial uncertainty. Although probabilistic registration methods have been widely adopted in multi-subject fNIRS research and provide acceptable spatial accuracy for group-level analyses, the anatomical attribution of channels to specific Brodmann areas (e.g., BA46 and BA11) should be considered approximate rather than precise. Future studies integrating individual structural MRI or 3D digitization of optode locations would allow more accurate cortical localization and strengthen anatomical interpretations. Second, fNIRS recordings were limited to the bilateral prefrontal cortex using an eight-channel configuration, which constrained the ability to capture neural activity and functional connectivity with other emotion-related regions, such as the limbic system. Future research could address these issues by employing wider cortical and subcortical coverage and channel-specific power estimations to yield a more comprehensive understanding of the neural mechanisms underlying emotional inhibitory control. Third, the present study focused on adolescents and young adults, which limits the ability to characterize the broader developmental trajectory of emotional inhibitory control. Although this design allowed us to examine neural and behavioral differences across a key transitional period, the absence of a child group prevents a more comprehensive assessment of how emotional inhibitory control and its neural substrates emerge earlier in development. Including children in future studies would allow researchers to test whether the observed patterns of prefrontal recruitment reflect a gradual developmental progression or a nonlinear shift across developmental stages. Such designs would also help clarify how behavioral performance and neural recruitment become increasingly coordinated across development.

Conclusion

In sum, the present study provides evidence for a developmental shift in prefrontal activation patterns during emotional inhibitory control, from relatively right-lateralized recruitment in adolescence to more bilateral engagement in young adulthood. This pattern likely reflects the combined influence of multiple developmental mechanisms. Differences in vascular responsivity, together with ongoing structural maturation of the prefrontal cortex and refinement of dopaminergic modulation, may collectively contribute to increasing efficiency and integration within prefrontal control networks. At the cognitive level, these neural changes may be accompanied by a shift from relatively featural processing of emotional stimuli toward more integrated control strategies that coordinate emotional and executive demands. Future research employing multimodal imaging and longitudinal designs will be important for clarifying the dynamic relationships among neurovascular coupling, neurochemical development, and cognitive control, and for establishing normative developmental trajectories that may help identify early neurobiological markers of psychiatric risk.

Author contributions

Tingting Chen: Formal analysis, Data curation, Methodology, Investigation, Writing–original draft, Writing – review & editing. Lei Jiang: Formal analysis, Visualization. Rou Wen: Methodology, Supervision. Changhao Jiang: Methodology, Supervision, Writing – review & editing, Funding acquisition.

Funding

This research was supported by the National Natural Science Foundation of China (No. 32371132) to Changhao Jiang.

Data availability

The datasets generated and analyzed during the current study are available upon reasonable request from the corresponding author. The data are not publicly archived but can be shared for research purposes with appropriate acknowledgment and with the approval of the corresponding author.

Declarations

Ethics approval and consent to participate

All procedures involving human participants were approved by the Ethics Committee of the Scientific Research Department at Beijing Dance Academy (Approval No.: BDA-SRD-IRB-2024-006). The study was conducted in accordance with the ethical standards of the Declaration of Helsinki. Written informed consent was obtained from all participants, and from parents or legal guardians for participants under 18 years of age.

Competing interests

The authors declare no competing interests.