Emergent blink rate in early childhood is associated with neural origins of executive function

Abstract

Executive function develops rapidly in early childhood and predicts later success in life, yet its neurobiological correlates during this period are not yet well characterised. In adults, executive function is supported by prefrontal–striatal circuits under neuromodulatory control, including dopamine. Here, we show that spontaneous eye-blink rate, a physiological output influenced by neuromodulatory systems, is associated with the developmental trajectory of executive function and prefrontal functional differentiation. We used fNIRS to monitor prefrontal cortex (PFC) activation in 113 children (35–80 months) during a cognitive flexibility task. Older children showed better task performance and greater right-hemisphere and dorsal bias in their PFC activation. Importantly, higher eye-blink rate was also associated with this right-dorsal PFC activation and with better task performance. The current findings demonstrate that emergent eye-blink rate is associated with the neural origins of executive function development in young children. This highlights the potential of eye-blink rate to serve as a unique and previously unrecognised indicator of this critical developmental neural mechanism.

Spontaneous eye-blink rate, prefrontal haemodynamic responses and cognitive flexibility are measured in 113 children aged 35–80 months. With age, prefrontal activation shifts towards a right-dorsal pattern and performance improves; a higher blink rate accompanies both, marking it as a candidate indicator of executive function development.

Introduction

Executive function—the higher-order cognitive processes enabling goal-directed behavior and self-control—is essential to quality of life and is an important factor predicting academic achievement, social adjustment, and health across the lifespan^1–3. Executive function starts to develop during infancy, showing rapid changes during preschool years, and continues to develop through adolescence and into adulthood^4,5. This trajectory is thought to reflect the functional maturation of the prefrontal cortex (PFC) and associated neural networks^6–8. According to a prevailing working hypothesis, task-related PFC activation shifts with age—from diffuse, widespread engagement in early childhood to more focal, specialized configurations that rely on efficient networks with other brain regions in later development^9–11. This transition signifies a functional differentiation of neural recruitment, marking the refinement of prefrontal organization. However, it remains unclear whether this functional reorganization already begins to emerge during the preschool years and what neural mechanisms might modulate it¹².

In adults, executive function is controlled by neuromodulatory systems, including dopaminergic tone. Accumulating evidence demonstrates that the dopaminergic transmission originating in the midbrain is linked with executive function via function in the dorsolateral PFC (DLPFC) and striatum, which is to say the frontostriatal networks^13,14. Dopamine projections to the PFC are thought to enhance the signal-to-noise ratio of neural representations¹⁵, stabilizing goal-relevant information, whereas dopaminergic input to the striatum acts as a gate that enables selective updating of task rules^16–18. Given its critical involvement in both cognitive switching and updating in adults, basic maturation in these circuits may also contribute to the rapid emergence of executive function and efficient cortical recruitment during early childhood. Adults exhibit stronger functional connectivity in the midbrain and lateral PFC (LPFC) during switching tasks compared to preadolescents¹⁹, supporting the importance of the mechanism underlying executive function development.

However, research on the neuromodulatory systems that may underlie rapid executive function improvement in early childhood remains scarce. PET and postmortem human brain studies indicate substantial developmental maturation of the human striatal dopaminergic system from infancy through adolescence, with some indices approaching stability by late adolescence or early adulthood^20,21. On the other hand, resting-state fMRI findings indicate that frontostriatal functional connectivity strengthens only after adolescence²², suggesting that it is still developing during early childhood. Interestingly, Moriguchi and Shinohara²³ reported that polymorphisms in the catechol-O-methyltransferase (COMT) gene—a key regulator of prefrontal dopaminergic tone—are associated with switching performance and right DLPFC (rDLPFC) activation in 3- to 6-year-old children. This finding suggests that the dopaminergic system may already play a functional role in the developing prefrontal function during early childhood. Nevertheless, the absence of non-invasive techniques for tracking the development of neuromodulatory systems means that this mechanism has remained largely unexplored.

To address this challenge, we turned to spontaneous eye-blink rate, a physiological output influenced by central neuromodulatory and arousal-related systems^24–29. Blinking is tightly coupled with brainstem arousal regulation and exhibits sensitivity to changes in internal state³⁰. Across adult, clinical, and developmental studies, individual differences in baseline eye-blink rate have been associated with frontostriatal-dependent functions, including executive function^{25,27,28,31,32,33}. Infant studies further suggest that eye-blink rate varies with learning contexts linked to dopaminergic function^34–38. Several animal, PET, and clinical studies have also reported relationships between eye-blink rate and striatal dopaminergic measures, including D2 receptor function^{24–26,28,39,40} and pharmacological modulation^{24–26,39,40}, although the neurochemical specificity of eye-blink rate remains debated^26,41–43. Interestingly, blink rates are exceptionally low in infancy but increase several-fold between approximately 1 and 5 years of age, and substantial inter-individual differences emerge over the same period^44–46. Together, these observations motivated us to treat eye-blink rate as a scalable trait-like physiological signature linked to neuromodulatory maturation, with dopamine as one contributor, and to test whether it covaries with executive function development and prefrontal functional differentiation in early childhood.

Here, we examined whether early changes in spontaneous eye blinking are linked to the rapid development of executive function and its neural underpinnings in early childhood (35–80 months), using fNIRS during a cognitive flexibility task widely used in young children^7,47 (Fig. 1). We addressed three questions. First, we examined whether eye-blink rate is higher at older ages and is associated with cognitive flexibility; we hypothesized that a higher eye-blink rate would relate to better switching performance. Second, we examined how task-evoked PFC activation differs across this age range; we hypothesized an age-related functional differentiation of prefrontal activation. Third, we examined whether prefrontal activation, eye-blink rate, and switching performance are interrelated; we hypothesized that this functional differentiation of the PFC would be associated with both higher eye-blink rate and better switching performance.

Fig. 1. Experimental procedures.

a Schematic illustration of the experimental settings. Participants performed the Dimensional Change Card Sort (DCCS) task while prefrontal hemodynamics and spontaneous eye-blink rate were recorded. b Spontaneous eye-blink rate. Eye blinks were identified from video recordings to calculate the spontaneous eye-blink rate (blinks/min). Raster plots from two example participants illustrating low vs. high sEBR during the 6-min are pictured. c Design of the DCCS task. Top panel: schematic of the task. The task consists of three phases: a pre-switch phase (sorting by the first dimension), a post-switch phase (sorting by the second dimension), and a mix-switch phase (alternating between the two dimensions). Middle panel: block design for fNIRS measurement. A 15-s rest period preceded each 25-s task phase. Bottom panel: the stimulus sets used in the task. d fNIRS probe placement. The 16 measurement channels are mapped onto a standard brain template. The measurement area covered the bilateral dorsolateral prefrontal cortex (DLPFC) and rostrolateral prefrontal cortex (RLPFC).

Methods

Participants and procedures

Participants were children aged 2–6 years old recruited from two preschools in Osaka and Kyoto, Japan. All children were reported to have no developmental abnormalities, and prior informed consent was obtained from their parents. The study was conducted in accordance with the Declaration of Helsinki and was approved by the Ethics Committee of the Unit for Advanced Studies of the Human Mind, Kyoto University (approval number: 3-P-27). Data from 113 children (55 female, 58 male; mean age: 59 months, SD = 11.8, range: 35–80 months) were included in the final analyses. Sex was based on information provided by the participating preschools. Data on socioeconomic status, communities of descent, and race/ethnicity were not collected. We were unable to obtain usable data from an additional 10 children due to experiment start refusal (n = 6), refusal to wear the fNIRS cap (n = 3), and task non-completion due to excessive movement (n = 1).

Sample size determination

Acknowledging that prior developmental neuroimaging studies have often faced limitations in statistical power due to small sample sizes¹², the present study was designed a priori to obtain a large and robust sample. An a priori power analysis conducted with G*Power (v. 3.1) indicated that a sample of n = 82 was required to detect an anticipated medium-sized correlation (r = 0.30) with 80% power at a significance level of α = 0.05 (two-tailed). Our final analyzed sample of n = 113 fulfils this requirement, providing robust statistical power for our hypotheses.

Behavioral task: Dimensional Change Card Sort (DCCS)

To measure executive function in young children, we used the DCCS task, a canonical measure of cognitive flexibility for this age^7,47. We implemented a multi-stage version of the task suitable for a wide age range from early 3-year-olds to 6-year-olds, following the procedure from previous studies, including our own^23,48–50. Stimulus cards consisted of two dimensions, “color” and “shape” (Fig. 1a). In each trial, a target card was presented with two test cards, each matching the target on only one of the two dimensions. The task comprised three sessions, each using a different set of cards. Each session consisted of a pre-switch phase (color rule), a post-switch phase (shape rule), and a mix phase (switching between both rules). Each phase served as a single fixed-duration block in the fNIRS analysis, yielding nine such blocks per participant (3 sessions × 3 phases). Each sorting phase had a fixed duration of 25 s, identical for every participant, during which the experimenter paced card delivery so that it was not too fast, with at most eight trials presented within the 25-s window. Across the three sessions (maximum 24 trials = 8 trials × 3 sessions), the mean number of completed trials per participant was 22.0 ± 2.4 for the pre-switch phase, 22.2 ± 2.6 for the post-switch phase, and 21.1 ± 2.8 for the mix phase. A sensitivity analysis adjusting for the number of completed trials is reported in Supplementary Note 1. A 15-s rest period was provided between task phases. The order of rule presentation was fixed for all participants: “color” for the pre-switch phase and “shape” for the post-switch phase. The mix phase followed a fixed sequence of rules: color, shape, shape, color, color, shape, shape, color. The primary outcome measure was the rule-switching success rate. This metric was defined as the accuracy on specific trials that required a rule switch, which is to say the first trial immediately following the transition from the pre-switch phase to the post-switch phase and up to 5 out of 8 trials within the mix phase for which the rule switched from that of the preceding trial, including the transition from the post-switch phase. The success rate was calculated as the rate of correctly sorted trials out of the total number of these specific switch trials that the participant reached across all three sets. Because switching accuracy was the primary outcome, participants completed all phases regardless of their accuracy, and all available data were analyzed. Excluding the one participant whose pre-switch accuracy was at or below chance (≤0.5) left the primary results essentially unchanged.

fNIRS recordings and analysis

We followed an fNIRS protocol refined through multiple studies by our group^{7,23,48,49,51}, which has confirmed methodological validity for DCCS in young children. Hemodynamic responses in the LPFC during the DCCS task were assessed using a multi-channel fNIRS optical topography system (OEG-16; Spectratech Inc., Tokyo, Japan) that emits near-infrared light at two wavelengths (770 nm and 840 nm). The configuration and placement of the fNIRS probe followed the procedures of our prior research^23,48,49. The probe consisted of six light-source optodes and six detector optodes, forming 16 channels with a source–detector distance of 3 cm. The probe was placed with the lower row centered at Fpz. We performed spatial registration using an independent sample to project fNIRS channel locations into MNI standard space^52–54. Probabilistic anatomical labeling⁵⁴ provided anatomical reference⁵⁵ for the a priori channel-based regions of interest (ROIs)^23,48,49,56. The upper-channel ROI primarily covered the DLPFC (Brodmann area 9/46⁵⁵), whereas the lower-channel ROI primarily covered the RLPFC (lateral Brodmann area 10, frontopolar area⁵⁵) (Supplementary Tables 1, 2 and Supplementary Note 2). The acquired optical data were converted into changes in hemoglobin concentration based on the modified Beer-Lambert law. We did not assume a specific differential pathlength factor and calculated the relative change in oxygenated hemoglobin (oxy-Hb) in units of mM·mm⁵⁷. Optical pathlength differences are reported to be minimal for symmetric hemispheric and adjacent regional comparisons⁵⁸.

fNIRS signal preprocessing, including moving-average smoothing, linear fitting, and hemodynamic modality separation, was performed using the proprietary OEG-16 analysis software (Spectratech Inc.), as described below. File concatenation and averaging of the preprocessed outputs were performed using Python. At acquisition, valid fNIRS values were obtained only from channels that passed the pre-measurement calibration implemented in the OEG-16 acquisition software. Trials with visible motion contamination were identified by synchronized video–fNIRS review and excluded (98.4% retained), using the following criteria: body movement, or clear probe displacement/ manipulation coinciding with a spike-like oxy-Hb deflection, with a magnitude on the order of 0.5 mM·mm from baseline used as a practical guide, consistent with prior block-design fNIRS studies reporting signals in mM·mm units^59,60. Reliability was confirmed by a second coder who independently re-rated a random 20% subset. The two coders concurred on 99.5% of decisions, and the oxy-Hb values showed high agreement (ICC(A,1) > 0.995 across the four ROIs). Data were sampled every 0.655 s. The data were smoothed using a 7-point moving average filter, following which linear fitting was performed to reduce slow drifts.

Next, based on the hemodynamic-modality principle that task-evoked cortical activation typically produces an anti-correlated pattern between oxy-Hb and deoxy-Hb, whereas systemic physiological fluctuations, superficial perfusion, and some motion artifacts tend to induce positively correlated changes, we applied hemodynamic modality separation⁶¹ to each channel to decompose the signals into a functional component (putatively neural) and a systemic component (physiological noise/artifact), potentially capturing residual non-neural fluctuations related to subtle facial/motion activity (e.g., blinks). This approach has been described as a promising option for mitigating systemic contamination in fNIRS data⁶². Consistent with earlier pediatric applications, this procedure was robust in preschoolers and supported reliable estimation of DCCS-evoked responses^23,48,49,56. Subsequent analyses used the functional component, which is expected to more closely reflect task-evoked cortical hemodynamics. We report oxy-Hb from the functional component, as oxy-Hb generally provides a higher signal-to-noise ratio than deoxy-Hb. To increase the signal-to-noise ratio, ROI signal values were calculated by averaging across channels (ch) for the rDLPFC (ch 2, 4, 5), lDLPFC (ch 11, 13, 14), rRLPFC (ch 3, 4, 6), and lRLPFC (ch 12, 13, 15), accounting for the similarity in optical properties of adjacent channels⁵⁸. Channels spanning two regions (ch 4, 13) were included in the calculation for each region with a weight of 0.5^23,48,49,63. Task-evoked activity was quantified as the mean oxy-Hb change in the task window (4.6 s post-onset to 20 s post-task) relative to the pre-onset baseline (−3.3 to 0 s), based on physiological plausibility and prior literature^32,49. We adopted this block-based quantification rather than an event-related approach because its higher signal-to-noise ratio provides more stable estimates of task-evoked activity, following prior DCCS–fNIRS studies in this age group that yielded robust task-evoked activation^23,48,49. Supplementary robustness analyses indicated that the main results were not driven by this specific window choice (Supplementary Note 3 and Supplementary Fig. 2). Finally, we applied an iterative Grubbs’ test (α = 0.0001) to the mean oxy-Hb values within each ROI to confirm that the results were not driven by rare artefactual extremes.

Spontaneous eye-blink rate

Spontaneous eye-blink rate was measured throughout the 6-min DCCS task (~40% rest, ~60% task). Each child’s face was recorded with a front-facing video camera (30 fps) positioned approximately 100–150 cm in front of the child. Two trained raters independently counted blinks for all videos (inter-rater correlation, r = 0.93; ICC(2,1) = 0.93, 95% CI [0.90, 0.95]), and their mean was used. A third rater independently scored a randomly selected 20% subset, showing high correlation with the average of the first two raters (r = 0.95; ICC(2,1) = 0.95, 95% CI [0.89, 0.98]). This video-based measurement has been used in prior studies; moreover, in adults, it correlates well with vertical electro-oculography (vEOG), underscoring its validity^31,32. Because attaching vEOG electrodes can impede task engagement and compliance in young children, the video-based approach provides the least intrusive and most feasible solution for preschooler assessments while maintaining measurement validity. For each participant, eye-blink rate (blinks/min) was computed by dividing the total number of blinks during the 6-min recording by six (Fig. 1b). To minimize reactivity and encourage natural blinking, children were not explicitly told that blinking was the focus of the study. We adopted this DCCS-concurrent approach to obtain trait-like eye-blink rate metrics; although dedicated resting-state blocks (e.g., cross-fixation) are used in adult studies^31,32, they are pragmatically difficult for young children, and our design provides a valid, standardized alternative that improves cross-study comparability. Supplementary analysis confirmed the cross-context stability of the current method (Supplementary Fig. 3). Data were unavailable for one child due to insufficient eye visibility in the video, leaving data from 112 participants for analyses.

Analysis plan

Statistical analyses were primarily conducted in R (version 4.3.2), with supplementary analyses and data visualization performed in GraphPad Prism (version 9). The significance level was set at α = 0.05; corrections for multiple tests, when applied, are reported alongside the corresponding analyses. Our analysis plan comprised two pre-specified tiers. The primary planned analyses, which constitute the confirmatory hypothesis tests of this study, were the bivariate correlations shown in Figs. 2 and 4 and correspond directly to our a priori power analysis; for these, because eye-blink rate is heteroscedastic, Spearman’s rank correlations (rho) and partial correlations (controlling for age) were used to assess the relationships between age, eye-blink rate, switching accuracy, and task-related LPFC activation. As a second, planned analytic framework for characterizing developmental changes in task-evoked PFC activation (Fig. 3), we analyzed oxy-Hb changes using linear mixed-effects models in R (using the “lme4”, “lmerTest”, and “emmeans” packages). We tested a comprehensive model including fixed factors of task phase (pre-switch, post-switch, mix), hemisphere (left, right), region (dorsolateral, rostrolateral), and age in months, along with all interaction terms, and participant as a random intercept. A likelihood ratio test confirmed this model’s superior fit against a base model without age terms. We used Type III analysis of variance (ANOVA; Satterthwaite’s approximation) and post-hoc comparisons to analyze main effects and interactions. Degrees of freedom were approximated using the Kenward–Roger method for post hoc comparisons and the Satterthwaite method for omnibus tests. Residual diagnostics indicated that the assumptions of the linear mixed-effects models were adequately met: residual variance was unrelated to the fitted values, differences in residual variance across design cells were modest (largest/smallest SD ratio = 1.57), and departures from normality were mild—deviations to which fixed-effect estimates in linear mixed-effects models are robust at this sample size. We used this single, pre-specified model, rather than many separate ROI- and phase-wise correlations, to guard against selective reporting by reporting all fixed effects together. To integrate brain and behavior, we correlated the cognitive flexibility-related activity (i.e., post-switch and mix phases; the former primarily indexing maintenance of the new rule and inhibition of the previously relevant dimension, the latter additionally requiring trial-by-trial rule switching) of the four ROIs with eye-blink rate and switching accuracy (Spearman’s rho) using a Bonferroni correction (α = 0.0125). We repeated the primary correlation analyses, adjusting for sex and site, and additionally assessed robustness to influential outliers; the conclusions were unchanged (Supplementary Notes 1 and 4). Although this study was not formally preregistered, the above analyses are confirmatory tests of our planned hypotheses, whereas exploratory analyses are reported in the Supplementary Information.

Fig. 2. Spontaneous eye-blink rate (sEBR) is associated with switching performance.

a Scatterplot of age in months vs. switching accuracy. The curve shows a nonlinear fit from an exponential plateau model to capture the ceiling effect (shown for visualization only). n = 113; rho = 0.68, 95% CI [0.56, 0.78], p < 0.001. b Scatterplot of age in months vs. sEBR. Higher sEBR was associated with older age. The line denotes the linear regression fit (shown for visualization only). n = 112; rho = 0.29, 95% CI [0.11, 0.46], p = 0.002. c Scatterplot of sEBR vs. switching accuracy. Higher sEBR is associated with greater switching accuracy; this association remains significant after statistically controlling for age. The curve shows a nonlinear fit from an exponential plateau model to capture the ceiling effect (shown for visualization only). n = 112; rho = 0.35, 95% CI [0.17, 0.51], p < 0.001, and partial rho = 0.22, 95% CI [0.03, 0.39], p = 0.022 after controlling for age. All associations were tested using Spearman’s rank correlation. ***p < 0.001, **p < 0.01. All p values are two-sided.

Fig. 4. Activation of the right dorsolateral prefrontal cortex (rDLPFC) is associated with both spontaneous eye-blink rate (sEBR) and switching performance.

a Relationships between cognitive flexibility-related OxyHb activation in four prefrontal subregions (rDLPFC, lDLPFC, rRLPFC, lRLPFC), and sEBR (coral) and switching accuracy (teal). Points indicate the correlation coefficient (rho) and horizontal bars the 95% confidence interval (CI); the vertical dashed line marks rho = 0. Sample sizes: n per ROI (sEBR/switching accuracy): rDLPFC 111/112, lDLPFC 106/107, rRLPFC 112/113, lRLPFC 111/112 participants. All associations were tested using Spearman’s rank correlation with Bonferroni correction (α = 0.0125). Exact statistics for all ROIs are reported in Supplementary Table 4. b Scatter plots of rDLPFC activation against the two measures. Top: OxyHb activation vs. sEBR. Bottom: OxyHb activation vs. switching accuracy. Each point represents an individual participant; the solid line shows the linear regression fit and the shaded band its 95% confidence interval　(shown for visualization only). Higher rDLPFC activation was associated with both higher sEBR and higher switching accuracy. rDLPFC associations were tested using Spearman’s rank correlation. Reported p values are uncorrected and were evaluated for significance against a Bonferroni-adjusted threshold (α = 0.0125 for the four ROIs). rDLPFC activation vs. sEBR (n = 111): rho = 0.27, 95% CI [0.09, 0.44], p = 0.004; rDLPFC activation vs. switching accuracy (n = 112): rho = 0.32, 95% CI [0.14, 0.48], p < 0.001. All p values are two-sided.

Fig. 3. Task-related prefrontal activation and its modulation by age.

a Grand-average waveforms (time courses) of oxy-Hb changes during each task phase—pre-switch (Pre; red), post-switch (Post; blue), and mix (Mix; green)—in the four regions of interest (ROIs: rDLPFC, lDLPFC, rRLPFC, and lRLPFC). Shaded areas represent the standard error (SE) of the grand-average mean. Vertical dotted lines indicate the onset (0 s) and offset (25 s) of the task block. b Distribution of mean oxy-Hb changes during each phase across all ROIs. Individual data points are shown. c, d Main effects on oxy-Hb signals (mM·mm) derived from the linear mixed-effects model (LMM). c Significant main effect of task phase, with greater activation in the post-switch (blue) and mix (green) phases than in the pre-switch phase (red), p < 0.001. d Significant main effects of hemisphere (right > left) and region (DLPFC > RLPFC). Hemisphere, p < 0.001; region, p = 0.008. Error bars represent the SE of the estimated marginal means. e, f Significant two-way interactions between age in months and the experimental factors. Plots show estimated marginal trends in oxy-Hb changes derived from the LMM. Higher age was associated with higher activation in the right hemisphere and dorsal region (DLPFC), but with lower activation in the left hemisphere and rostral region (RLPFC). Hemisphere × age, p < 0.001; region × age, p < 0.001. n = 113 children. Per-cell sample sizes for each phase × hemisphere × region combination were as follows: lRLPFC, 112; lDLPFC, 107; rRLPFC, 113; and rDLPFC, 112 in the pre-switch and mix phases, and 111, 106, 112, and 111, respectively, in the post-switch phase. Main effects and interactions were evaluated using Type III tests with Satterthwaite’s approximation for denominator degrees of freedom. Degrees of freedom for post-hoc contrasts were calculated using the Kenward–Roger method. Tukey adjustment was applied to pairwise comparisons involving the three-level phase factor, including pairwise comparisons of age-related slopes across phases. No multiplicity adjustment was applied to the single two-level contrast for either hemisphere or region. Shaded areas in e and f represent 95% confidence intervals around the estimated marginal trends. ***p < 0.001; **p < 0.01. All statistical tests were two-sided.

Results

Spontaneous eye-blink rate is associated with switching performance

First, we recorded the rate of correct responses during the pre-switch, post-switch, and mix phases of the Dimensional Change Card Sort (DCCS) task. Consistent with previous research²³, the rate of correct responses significantly decreased from the pre-switch phase to the post-switch and mix phases (Friedman test: χ²(2) = 68.82, p < 0.001, Kendall’s W = 0.31, 95% CI [0.23, 0.43]; post hoc Wilcoxon signed-rank tests, two-sided, Bonferroni-corrected: pre vs. post, V = 1321.5, rank-biserial r = 0.55, 95% CI [0.38, 0.68], p < 0.001; pre vs. mix, V = 3877, r = 0.89, 95% CI [0.84, 0.93], p < 0.001; Supplementary Fig. 1). Next, the percentage of successful shifting (= switching accuracy) was analyzed as primary task performance. As expected, higher switching accuracy was associated with older age in months (Spearman’s rho = 0.68, 95% CI [0.56, 0.78], p < 0.001, Fig. 2a). Higher spontaneous eye-blink rate was also associated with older age (rho = 0.29, 95% CI [0.11, 0.46], p = 0.002, Fig. 2b). Critically, supporting our main hypothesis, eye-blink rate was positively correlated with switching accuracy (rho = 0.35, 95% CI [0.17, 0.51], p < 0.001, Fig. 2c). This association remained significant after controlling for the effect of age (rho = 0.22, 95% CI [0.03, 0.39], p = 0.022, Fig. 2c). These results suggest that eye-blink rate is associated with cognitive flexibility independent of age. Exploratory and robustness analyses further support these findings and are reported in Supplementary Fig. 1b–d (phase-wise analyses), Supplementary Note 1 (sex- and site-adjusted analyses), and Supplementary Note 4 (outlier-adjusted analyses).

The cognitive flexibility task induced LPFC activity

To examine task-related prefrontal activation, we fitted a linear mixed-effects model (LMM) to oxy-Hb change with task phase (pre-switch, post-switch, mix), hemisphere (left, right), and region (dorsolateral, rostrolateral) as fixed factors and participant as a random intercept. This model tested the overall effects of the DCCS task on prefrontal activation. The ANOVA revealed significant main effects of phase (F(2, 1205.2) = 38.44, p < 0.001, partial η² = 0.06, 95% CI [0.036, 0.087]), hemisphere (F(1, 1207.2) = 20.07, p < 0.001, partial η² = 0.02, 95% CI [0.005, 0.033]), and region (F(1, 1207.2) = 7.14, p = 0.008, partial η² = 0.006, 95% CI [0.000, 0.017]). No significant interactions were observed in this model (phase × hemisphere: F(2, 1204.5) = 1.39, p = 0.249, partial η² = 0.002, 95% CI [0.000, 0.010]; phase × region: F(2, 1204.5) = 1.61, p = 0.201, partial η² = 0.003, 95% CI [0.000, 0.010]; hemisphere × region: F(1, 1207.2) = 0.00, p = 0.953, partial η² < 0.001, 95% CI [0.000, 0.000]; phase × hemisphere × region: F(2, 1204.5) = 0.06, p = 0.946, partial η² < 0.001, 95% CI [0.000, 0.001]). The model accounted for a moderate proportion of variance overall (conditional R² = 0.36), with the fixed effects explaining a smaller share (marginal R² = 0.05). Post hoc comparisons on estimated marginal means showed that oxy-Hb activity was significantly higher during the post-switch and mix phases (pre vs. post: t(1205.1) = 7.72, p < 0.001, Cohen’s d = 0.52, 95% CI [0.39, 0.65]; pre vs. mix: t(1204.0) = 7.46, p < 0.001, Cohen’s d = 0.50, 95% CI [0.37, 0.63]) compared with the pre-switch phase, while no difference was found between the mix and post-switch phases (t(1205.1) = 0.28, p = 0.957, Cohen’s d = 0.02, 95% CI [−0.11, 0.15]) (Fig. 3a–c). One-sample t-tests against baseline further clarified this pattern. Significant cortical activation was confirmed across the post-switch and mix phases (except for the left RLPFC (lRLPFC) in the post-switch phase), whereas no significant activation was observed in the pre-switch phase (Supplementary Table 3). Notably, activity was significantly greater in the right hemisphere compared to the left (t(1206.7) = 4.48, p < 0.001, Cohen’s d = 0.25, 95% CI [0.14, 0.36]) and in the DLPFC compared to the RLPFC (t(1206.7) = 2.67, p = 0.008, Cohen’s d = 0.15, 95% CI [0.04, 0.26]) (Fig. 3a, b, d). These results confirm that the cognitive flexibility-related task (post-switch and mix phases) robustly recruits the LPFC, with a particular emphasis on the right and dorsal regions.

Age-related differences in LPFC activity

To determine whether the prefrontal activation during the cognitive flexibility task was modulated by age in months, we compared the base model to a more comprehensive LMM that included age and all its interaction terms (Fig. 3e, f). A likelihood ratio test revealed that the model incorporating age provided a significantly better fit to the data (χ² (12) = 61.92, p < 0.001, change in marginal R² = 0.030, change in conditional R² = 0.031, ΔAIC = −37.9), indicating that age is a critical factor in explaining the observed prefrontal activation. We therefore focused our subsequent analysis on the LMM. The ANOVA for the age-interaction model not only confirmed significant main effects of phase, hemisphere, and region but also revealed significant two-way interactions between age and phase (F(2, 1194.8) = 10.12, p < 0.001, partial η² = 0.017, 95% CI [0.005, 0.033]), hemisphere (F(1, 1195.8) = 22.15, p < 0.001, partial η² = 0.018, 95% CI [0.006, 0.036]), and region (F(1, 1196.1) = 15.03, p < 0.001, partial η² = 0.012, 95% CI [0.003, 0.028]). No other interactions reached significance (phase × hemisphere: F(2, 1193.5) = 0.64, p = 0.529, partial η² = 0.001, 95% CI [0.000, 0.007]; phase × region: F(2, 1193.5) = 0.05, p = 0.954, partial η² < 0.001, 95% CI [0.000, 0.001]; hemisphere × region: F(1, 1195.3) = 1.62, p = 0.203, partial η² = 0.001, 95% CI [0.000, 0.009]; phase × hemisphere × region: F(2, 1193.5) = 0.27, p = 0.766, partial η² < 0.001, 95% CI [0.000, 0.004]; phase × hemisphere × age: F(2, 1193.5) = 0.78, p = 0.459, partial η² = 0.001, 95% CI [0.000, 0.007]; phase × region × age: F(2, 1193.5) = 0.01, p = 0.992, partial η² < 0.001, 95% CI [0.000, 0.000]; hemisphere × region × age: F(1, 1196.1) = 1.71, p = 0.191, partial η² = 0.001, 95% CI [0.000, 0.009]; phase × hemisphere × region × age: F(2, 1193.5) = 0.23, p = 0.797, partial η² < 0.001, 95% CI [0.000, 0.004]). To decompose these interactions, we performed post-hoc analyses on the estimated marginal trends. The hemisphere:age and region:age interactions converged to highlight a specific locus of age-related difference. The effect of age on oxy-Hb signal change differed significantly between the hemispheres (left–right = −0.0005, 95% CI [−0.0008, −0.0003], t(1195.4) = 4.71, p < 0.001) (Fig. 3e) and between regions (RLPFC–DLPFC = −0.0004, 95% CI [−0.0007, −0.0002], t(1195.6) = 3.88, p < 0.001) (Fig. 3f). These effects were driven by a pattern wherein older age was associated with lower activity in the left hemisphere and rostral subregion but higher activity in the right and dorsal subregions. This suggests a specific age-related functional differentiation in the PFC. In parallel, the phase:age interaction was driven by a significant difference in age-related trends between the mix and pre/post phases (pre vs. mix: −0.0005, 95% CI [−0.0008, −0.0001], t(1193.0) = 3.33, p = 0.003; post vs. mix: −0.0006, 95% CI [−0.0009, −0.0003], t(1195.0) = 4.28, p < 0.001). Taken together, these results demonstrate that the functional organization of this network during the cognitive flexibility task differs significantly across this age range, with greater specialization to the rDLPFC at older ages.

rDLPFC activity is associated with both spontaneous eye-blink rate and switching performance

Finally, we examined the relationship between PFC activation, eye-blink rate, and task performance (Fig. 4). We conducted correlation analyses to test for specific associations between the activity of each of the four ROIs and our key measures: eye-blink rate and switching accuracy. A Bonferroni correction was applied within each set of four comparisons (α = 0.05/4 = 0.0125).　Only the activity of the rDLPFC was significantly and positively correlated with both eye-blink rate (rho = 0.27, 95% CI [0.09, 0.44], p = 0.004) and switching accuracy (rho = 0.32, 95% CI [0.14, 0.48], p < 0.001); no other ROIs showed significant correlations with either measure at the Bonferroni-corrected threshold (rho < 0.20; uncorrected p > 0.039) (Fig. 4a, b; exact statistics for all ROIs are reported in Supplementary Table 4). Additionally, we confirmed that the positive correlations of rDLPFC activity with both eye-blink rate (partial correlation: rho = 0.21, 95% CI [0.03, 0.38], p = 0.027) and switching accuracy (partial correlation: rho = 0.21, 95% CI [0.02, 0.38], p = 0.029) were maintained after controlling for age. These correlation-based associations were also robust to adjustment for sex and site and to the exclusion of influential outliers (Supplementary Notes 1 and 4). As a supplementary, exploratory step, a mediation analysis was also consistent with this account (Supplementary Fig. 5), with rDLPFC activation accounting for part of the shared variance between eye-blink rate and switching accuracy.

Discussion

The neural mechanisms driving the emergence of human higher-order cognition from infancy remain one of the most significant unsolved mysteries. Progress has been fundamentally constrained by our general inability to apply restrictive or invasive measurements to children, leaving the neural substrates of this critical developmental period poorly understood. Here, we provide evidence that emergent spontaneous eye-blink rate, a physiological output influenced by neuromodulatory systems, is associated with the development of executive function. Furthermore, the results of our fNIRS analyses identify the involvement of task-related functional differentiation in the PFC as an underlying neural basis of the link between eye-blink rate and executive function development. Although there is a large amount of behavior-based literature on executive function development^4,5,64, this study demonstrates that the dramatic rise in eye-blink rate during early childhood could serve as a trait-like physiological readout of the brain development associated with complex cognition.

Our findings show that spontaneous eye-blink rate in early childhood is a neuro-behavioral marker associated with switching performance. This is particularly relevant given the known developmental trajectory of eye-blink rate, which is extremely low in newborns (approx. 0–5/min) and increases dramatically during development towards adult levels (approx. 20–30/min)^44–46. Our data mirrored this pattern: eye-blink rate was uniformly low with minimal individual variation in children under 4 years old, but individual differences were substantially larger at older ages, with some children already blinking over 20 times/min (Fig. 2b). Importantly, the age-related emergence of inter-individual variability in eye-blink rate allows eye-blink rate to be used as a tool to index cognitive flexibility beyond chronological age, rather than relegating it to a metric that merely tracks age in parallel: eye-blink rate was associated with switching performance even after controlling for age in months. This age-adjusted association is consistent with inter-individual differences in executive-control circuit maturation in a developmental period when prefrontal recruitment is thought to become more selective and switching performance begins to diverge across children. In adults, inter-individual differences in eye-blink rate have likewise been linked to executive function^32,65, raising the possibility that early-emerging variability in eye-blink rate reflects a trait-like signature that may persist into adulthood. This interpretation is consistent with evidence that individual differences in early executive function can forecast later outcomes in adults^1–3, although longitudinal testing is required in order to verify this. While the neural origin for this blink increase is unclear, a leading candidate is the maturation of the brain dopaminergic system⁶⁶. This hypothesis is supported by animal, clinical, and pharmacological evidence^24,27. For instance, adults with Parkinson’s disease, characterized by nigrostriatal dopaminergic pathway dysfunction, exhibit both a markedly reduced eye-blink rate—sometimes to levels comparable to young children^24,27,67—and impaired executive function, especially task-switching^68,69. Recent PET studies further confirm that eye-blink rate in these patients correlates closely with symptom severity and dopamine depletion²⁸. Therefore, the age-adjusted eye-blink rate-switching association we have revealed in children is highly consistent with the adult and clinical literature, suggesting that eye-blink rate in early childhood reflects the developing maturity of this dopaminergic system and frontostriatal network.

Despite the spatial constraints of fNIRS (restricted to the lateral prefrontal surface), we observed a clear age-related difference in the distribution of task-evoked PFC activity. Specifically, activation was more right-lateralized and dorsal-weighted at older ages across the preschool years, rather than showing uniform age-related differences across subregions (Fig. 3). This rightward and dorsal redistribution is consistent with the largely nonverbal DCCS, which often engages right-lateralized dorsal lateral frontoparietal control for task-set implementation and conflict resolution^{7,19,23,48,49,70}. In contrast, left-lateralized regions linked to verbal processing and the more rostral LPFC implicated in exploratory/relational control^71,72 may be less engaged in children with more efficient performance. This result supports and extends previous reports identifying rDLPFC activation during the DCCS task^7,23,48,49. Crucially, in a relatively large cohort spanning 35–80 months, we show systematic age-related differences in task-evoked LPFC recruitment across the preschool years. This reorganization is consistent with accounts of functional specialization theory (diffuse-to-focal transitions; interactive specialization)^9–12 and provides evidence that such specialization in the LPFC constitutes a neural basis for rapid early executive-function development.

Furthermore, rDLPFC activity was correlated with switching accuracy, confirming previous reports in a substantially larger sample⁷. Interestingly, eye-blink rate was also correlated with this same rDLPFC activity, but not with any of the other prefrontal subregions measured (Fig. 4a). Together, these results suggest that higher eye-blink rate, greater recruitment of the rDLPFC, and higher switching accuracy are correlated with one another. On this view, the link between eye-blink rate and executive function development is consistent with the functional maturation in LPFC, possibly related to the development of the brain dopaminergic system^27,29. This aligns with our prior work linking dopamine-related genes (COMT) to both DCCS performance and rDLPFC activity in young children^23,56. This pattern supports the hypothesis that a developmental improvement in the functional efficiency of the frontostriatal circuit promotes the rise in eye-blink rate and the functional differentiation of PFC recruitment, thereby enhancing executive function. As a supplementary, exploratory step, a mediation analysis was also consistent with this account (Supplementary Fig. 5); further mechanistic work is needed to clarify the underlying causal relationships.

More broadly, although the mechanisms of eye-blink rate development require further validation, our findings position eye-blink rate as a promising tool for assessing executive function in childhood. Executive function underpins a wide range of everyday activities, from following multistep instructions in the classroom to resolving social conflicts on the playground². These capacities scaffold performance on complex cognitive tasks and behavior regulation, providing a foundation for educational success and effective social interaction^1,73. Thus, supporting executive function is a key educational priority. Importantly, guidance should be informed not only by behavior but also by neural processes: measuring brain activity may inform the timing, targeting, and evaluation of interventions. In this context, eye-blink rate provides a non-invasive, low-cost, developmentally appropriate physiological marker associated with executive-function-related neurocognitive function that is otherwise difficult to access in early childhood. Because eye-blink rate can be acquired by non-experts from ordinary digital video—and paired with AI-based automated detection⁷⁴—it enables scalable at-home, preschool, and at-school data collection as well as “big-data” developmental science. Its minimal equipment requirements also make large-scale cross-cultural mapping of developmental trajectories feasible^56,75, opening access to vast, previously untapped observations from diverse daily-life settings. Because the present findings rest on the DCCS alone, extending this work to other executive-function measures and tasks will further clarify how broadly eye-blink rate indexes executive function in childhood.

Limitations

Several limitations should be noted, which also point to future directions for this line of exploration. First, the precise neuromodulatory mechanisms regulating eye-blink rate remain an important open question. Blink generation involves multiple subcortical structures and brainstem areas^30,76–78. It is likely a multifaceted process involving several neurotransmitter systems related to arousal, such as dopamine, acetylcholine, and noradrenaline^26,79. Even within the well-studied dopaminergic system, the specific receptor interactions governing eye-blink rate are not fully understood^26,41–43. This complexity underscores the importance of future research into the fundamental mechanisms of blinking, not only to understand the behavior itself, but also to fully comprehend its significance as a window into early brain development. Relatedly, the fine-grained timing of individual blinks is coupled to moment-to-moment fluctuations in arousal-related neural activity³⁰, and such within-trial blink dynamics are not captured by our trait-level eye-blink rate measure; resolving them would be a fruitful target for future work (Supplementary Fig. 4 for a tentative phase-wise rest/task analysis). Second, fNIRS signal interpretation remains debated because extracerebral/systemic physiology can contribute non-neural components⁸⁰. We mitigated these concerns by restricting recordings to the frontal scalp (minimal hair interference) in a Japanese nursery-school cohort and by applying quality control with hemodynamic modality separation to reduce residual non-neural/systemic contamination. Third, head circumference was not directly measured, so developmental differences in head size could, in principle, affect the cortical coverage of the fixed probe layout. We expect this influence to be limited. Based on Japanese national growth reference data, mean head circumference increases by about 2 cm (roughly 5%, from about 49.0 cm at 3 years to 51.1 cm at 6 years) across this age range⁸¹, in contrast to the rapid growth of infancy. A simple geometric estimate indicates that this corresponds to a small mediolateral cortical displacement: a channel sited at a fixed distance from the midline (Fpz) occupies a proportional circumferential position, so the ~4% increase in head circumference shifts the most central channels (sited 4.5 cm from the midline) by only about 2 mm. This shift is absorbed by our ROI-level analyses, because each ROI aggregates adjacent channels spanning more than 3 cm mediolaterally; this shift is no larger than the error inherent in probe placement and anatomical registration. Finally, as this study was not formally preregistered, longitudinal and independent-sample work could provide further verification, lending additional support to the developmental trajectories and potential causal relationships implied by our findings.

Conclusion

In conclusion, this study provides evidence in the 35-to-80-month age range that spontaneous eye-blink rate is associated with the development of executive function and the underlying prefrontal mechanisms. These findings indicate that blinking provides a unique and previously unrecognized insight into the neurodevelopmental processes of young children.

Author contributions

Ryuta Kuwamizu: conceptualization, methodology, performing the experiments, data analysis, visualization, writing—original draft preparation, project administration, and funding acquisition; Nozomi Yamamoto: performing the experiments, data analysis, and writing—review and editing; Kota Otani: performing the experiments, data analysis, and writing—review and editing; and Yusuke Moriguchi: conceptualization, writing—review and editing, supervision, and funding acquisition.

Peer review

Peer review information

Communications Psychology thanks Tanja Koenen, Şükrü Demiral, and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. Primary Handling Editors: Yana Fandakova and Troby Lui. A peer review file is available.

Funding

R.K. was supported by JSPS KAKENHI (Grant Numbers: 23H04830, 24K20598, and 23KJ1169). Y.M. and R.K. were supported by JSPS KAKENHI (Grant Number: 24K00486). K.O., N.Y., and R.K. also acknowledge the financial support from a crowdfunding project through the academic crowdfunding platform “academist” (Project No. 396). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

Data availability

The de-identified source data underlying the main figures and analyses reported in this study are openly available in the Open Science Framework (OSF) repository (https://doi.org/10.17605/OSF.IO/MRZ89)⁸².

Code availability

The R scripts used for the primary statistical analyses are openly available in the Open Science Framework (OSF) repository (https://doi.org/10.17605/OSF.IO/MRZ89)⁸². The repository includes a README file describing the software environment, package dependencies, and analysis parameters required to reproduce the reported analyses.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at https://doi.org/10.1038/s44271-026-00524-6.