The representation of facial emotion expands from sensory to prefrontal cortex with development

Facial expression recogni>on develops rapidly during infancy and improves from childhood to adulthood. As a cri>cal component of social communica>on, this skill enables individuals to interpret others’ emo>ons and inten>ons. However, the brain mechanisms driving the development of this skill remain largely unclear due to the difficulty of obtaining data with both high spa>al and temporal resolu>on from young children. By analyzing intracranial EEG data collected from childhood (5-10 years old) and post- childhood groups (13-55 years old), we find differen>al involvement of high-level brain areas in processing facial expression informa>on. For the post-childhood group, both the posterior superior temporal cortex (pSTC) and the dorsolateral prefrontal cortex (DLPFC) encode facial emo>on features from a high-dimensional, con>nuous space. However, in children, the facial expression informa>on is only significantly represented in the pSTC, not in the DLPFC. Further, the encoding of complex emo>ons in pSTC is shown to increase with age. Taken together, these data suggest that young children rely more on low-level sensory areas than on the prefrontal cortex for facial emo>on processing, leading us to hypothesize that top-down modula>on from prefrontal cortex to pSTC gradually matures during development to enable a full understanding of facial emo>ons, especially complex emo>ons which need social and life experience to comprehend. Introduc.on Understanding others' emo>onal states through their facial expressions is an important aspect of effec>ve social interac>ons throughout the lifespan. Behavioral data suggest that facial emo>on processing emerges very early in life1,2, as infants just months old can dis>nguish happy and sad faces from surprised faces3–5. However, children’s emo>on recogni>on is substan>ally less accurate than adults, and this ability prominently improves across childhood and adolescence6–11. Although extensive research in cogni>ve and affec>ve neuroscience has assessed developmental changes using behavioral and non-invasive neuroimaging approaches , our understanding of brain development related to facial expression percep>on remains limited. One influen>al perspec>ve on the development of face recogni>on is that it depends on the matura>on of face-selec>ve brain regions, including the fusiform face area (FFA), occipital face area (OFA), and posterior superior temporal sulcus (pSTS)12. Suppor>ng this view, Gomez, et al. found evidence for .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted May 23, 2025. ; https://doi.org/10.1101/2025.05.23.655726doi: bioRxiv preprint microstructural prolifera>on in the fusiform gyrus during childhood, sugges>ng that improvements in face recogni>on are a product of an interplay between structural and func>onal changes in the cortex13. Addi>onally, monkeys raised without exposure to faces fail to develop normal face-selec>ve patches, sugges>ng that face experience is necessary for the development of the face-processing network14. It is likely that the gradual matura>on of pSTS and FFA, two early sensory areas involved in the processing of facial expressions15,16, contributes to the improved facial expression recogni>on over development. Yet, few studies have inves>gated the development of neural representa>on of emo>onal facial expressions in FFA and pSTS from early childhood to adulthood in human. Besides the visual processing of facial configura>ons, understanding the emo>onal meaning of faces requires the awareness and interpreta>on of the emo>onal state of the other person, which is significantly shaped by life experience17,18. Thus, some researchers have proposed that the matura>on of emo>onal informa>on processing is related to the progressive increase in func>onal ac>vity in the prefrontal cortex 6,19,20. With development, greater engagement of the prefrontal cortex may facilitate top-down modula>on of ac>vity in more primi>ve subcor>cal and limbic regions, such as the amygdala21–23. Despite these theore>cal advances, the func>onal changes in the prefrontal cortex during the perceptual processing of emo>onal facial expressions over development remains largely unknown. Here, we analyze intracranial EEG (iEEG) data collected from childhood (5-10 years old) and post- childhood groups (13-55 years-old) while par>cipants were watching a short audiovisual film. In our results, children’s dorsolateral prefrontal cortex (DLPFC) shows minimal involvement in processing facial expression, unlike the post-childhood group. In contrast, for both children and post-childhood individuals, facial expression informa>on is encoded in the pSTC, a brain region that contributes to the perceptual processing of facial expressions. Furthermore, the encoding of complex emo>ons in the pSTC increases with age. These neuroimaging data imply that social and emo>onal experiences shape the prefrontal cortex’s involvement in processing the emo>onal meaning of faces throughout development, probably through top- down modula>on of early sensory areas.

Using AI and encoding models to study the neural representa.on of facial expression In this study, we analyzed intracranial EEG (iEEG) data collected from a large group of human neurosurgical pa>ents while they watched a short audiovisual film at the University Medical Center Utrecht24. The movie consisted of 13 interleaved blocks of videos accompanied by speech or music, Figure 1. Task design and analysis methods . (A) Movie structure. A 6.5-minute short film was created by ediPng fragments from Pippi on the Run into a coherent narraPve. The movie consisted of 13 interleaved blocks of videos accompanied by speech or music. (B) Data analysis schemaPc. Standard analysis pipeline for extracPng emoPon features from the movie and construcPng encoding model to predict iEEG responses while parPcipants watching the short film. .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted May 23, 2025. ; https://doi.org/10.1101/2025.05.23.655726doi: bioRxiv preprint 30 seconds each (Figure 1A). To characterize the neural representa+on of facial expression in the prefrontal cortex and low-level sensory areas across development, we analyzed iEEG data from 9 children (5-10 years old) and 31 post-childhood individuals (13-55 years old) who have electrode coverage in DLPFC, pSTC or both. First, Hume AI facial expression models were used to con>nuously extract facial emo>on features from the movie (Figure 1B). Then, we tested how well encoding models constructed from the 48 facial emo>on features (e.g., fear, joy) predict cor>cal high-frequency band (HFB) ac>vity (110-140 Hz) induced by the presented movie (Figure 1B). The model performance was quan>fied as the correla>on between the predicted and actual HFB ac>vi>es, which is also called predic>on accuracy. Differen.al representa.on of facial expression in children’s DLPFC Using the analysis approach described above, we examined how facial emo>on informa>on is represented by DLPFC (Figure 2A) while watching videos accompanied by speech (i.e. speech condi>on) in childhood and post-childhood groups. The predic>on accuracy of the encoding model was significantly greater than zero in the post-childhood group (Figure 2B, P=0.0096, two- tailed permuta>on test), sugges>ng that the neural responses in DLPFC were dynamically modulated by the facial emo>on features from the movie. However, facial emo>on features were not encoded in children’s DLPFC (Figure 2B, P=0.825, two-tailed permuta>on test). Moreover, the predic>on accuracy in children’s DLPFC was significantly lower than in the post- childhood group (P=0.0114, two-tailed permuta>on test). These findings show that the DLPFC dynamically encodes facial expression informa>on in post-childhood individuals but not in young children. To further understand the func>onal development of children’s DLPFC, we compared the effect of human voice on the representa>on of facial expression in DLPFC between the two groups. The effect of human voice was quan>fied as difference in predic>on accuracy between the speech and music condi>ons. Our results showed that human voice influences facial expression representa>on in the DLPFC differently across development (Figure 2C, P=0.0034, two-tailed permuta>on test). The presence Figure 2. Predic:on performance of encoding models in DLPFC. (A) SpaPal distribuPon of electrodes in DLPFC. Electrodes in all parPcipants from each group are projected onto MNI space and shown on the average brain. Red shaded areas indicate middle frontal cortex provided by the FreeSurfer Desikan-Killiany atlas25. Electrodes outside DLPFC are not shown. (B) The average predicPon accuracy across parPcipants for speech condiPon. The performance of encoding model is measured as Pearson correlaPon coefficient (r) between measured and predicted brain acPviPes. (C) PredicPon accuracy difference between speech condiPon and music condiPon for each group. Error bars are standard error of the mean. *P<0.05; **P<0.01. .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted May 23, 2025. ; https://doi.org/10.1101/2025.05.23.655726doi: bioRxiv preprint of human voice enhances facial expression representa>on in the DLPFC of post-childhood individuals but impairs it in children. Taken together, there are significant developmental changes in DLPFC‘s involvement in facial expression percep>on. The neural representa5on of facial expression in young children’s pSTC Ager iden>fying developmental differences in the involvement of high-level brain areas in processing facial expression, we next examined the neural representa>on of facial expression in children’s early sensory areas. As an area in the core face network, posterior superior temporal sulcus (pSTS) has been associated with early stages of facial expression processing stream12,15,26,27. Although previous studies suggested that the development of facial recogni>on depends on the matura>on of face- selec>ve brain regions13,14, it is s>ll unclear how facial expression informa>on is encoded in children’s pSTS. Here, we examined the performance of the facial expression encoding model in a rare sample of two children (S19:8-year-old and S39: 5-year-old) with electrode coverage in pSTC (Figure 3A). In both cases, the encoding model significantly predicts the HFB neural signals in the pSTC under the speech condi>on (Figure 3B, S19speech: P=0.0014, r=0.1951; S39speech: P=0.0183, r=0.15). The predic>on accuracy is reduced when human voice is absent from the video (S19music: P=0.0313, r=0.1674; S39music: P=0.3688, r=0.0574). Similarly, group-level results showed that the model performance is significantly greater than zero in the pSTC of post-childhood individuals (N=25, Figure 3C and 3D, P=0.003, two-tailed permuta>on test) and this neural representa>on of facial expression informa>on is significantly reduced when human voice is absent (paired-t-test, t24=2.897,P=0.0079). These results provide evidence that children's sensory areas encode facial emo>on features from a high- dimensional, con>nuous space in a manner similar to that of post-childhood individuals. The complexity of facial expression encoding in the pSTC increases across development To understand how facial expression representa>on in pSTC changes across development, we examined the feature weights of the facial expression encoding models in all par>cipants with significant predic>on Figure 3. Predic:on performance of encoding models in pSTC. (A) The electrode distribuPon of two children (s19 and s39). Electrodes in pSTC are green. (B) PredicPon accuracy of encoding models in the two children. (C) SpaPal distribuPon of recording contacts in post-childhood parPcipants’ pSTC. The pSTC electrodes idenPfied in individual space are projected onto MNI space and shown on the average brain. Contacts other than pSTC are not shown. Blue shaded areas indicate superior temporal cortex provided by the FreeSurfer Desikan-Killiany atlas25. (D) Average predicPon accuracy across post-childhood parPcipants. Error bars are standard error of the mean. **P<0.01. .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted May 23, 2025. ; https://doi.org/10.1101/2025.05.23.655726doi: bioRxiv preprint accuracy (10 post-childhood individuals and 2 children). The weight for each feature represents its rela>ve contribu>on to predic>ng the neural response. First, we calculated the encoding weights for complex emo>ons (averaging guilt, embarrassment, pride, and envy, which were selected as the most representa>ve complex emo>ons based on previous studies28–30) and basic emo>ons (averaging joy, sadness, fear, anger, disgust, and surprise). Then, we calculated their correla>ons with age separately. Our results showed that the encoding weight of complex emo>on was significantly posi>vely correlated with age (r12=0.8512，P=0.004 , Figure 4A leg). No significant correla>on between encoding weight of basic emo>on and age was observed (r12=0.3913，P=0.2085, Figure 4A right). In addi>on, we computed Pearson correla>ons between each individual feature weight and age, ranking the r values from largest to smallest (Figure 4B). The highest correla>ons were found for embarrassment, guilt, pride, interest, and envy—emo>ons that are all considered complex emo>on. Among them, the weights for embarrassment, guilt, pride, and interest showed significant posi>ve correla>ons with age (Figure 4C, embarrassment: r=0.7666, P=0.0036; pride: r=0.6773, P=0.0155; guilt: r=0.6421, P=0.0244, interest: Figure 4. Correla:on between encoding weights and age. (A) Lec: CorrelaPon between averaged encoding weights of five complex emoPons and age. Right: CorrelaPon between averaged encoding weights of six basic emoPons and age. (B) Pearson correlaPon coefficient between encoding weights of 48 facial expression features and age. The results are ranked from largest to smallest. Significant correlaPons noted with *(P<0.05, uncorrected) or **(P<0.01, uncorrected). (C) CorrelaPon between encoding weights of embarrassment, pride, guilt, interest and age (N=12). .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted May 23, 2025. ; https://doi.org/10.1101/2025.05.23.655726doi: bioRxiv preprint r=0.6377, P=0.0257, uncorrected for mul>ple comparisons), sugges>ng that the encoding of these complex emo>ons in pSTC increases with age. Thus, our results suggest that as development progresses, the pSTC becomes increasingly engaged in encoding complex emo>ons which requires represen>ng others' mental states and emerges later in development 31–33.

In this study, iEEG data from an open mul>modal iEEG-fMRI dataset were analyzed24. Par$cipants and electrode distribu$on Due to the research purposes of the current study, only par>cipants who had at least four electrode contacts in either DLPFC or pSTC were included in the data analysis (Table 1 and Table 2). Nine children (5-10 years old, 5 females) and thirty-one post-childhood individuals (13-55 years old, 18 females) are included in the present study. In the childhood group, eight par>cipants had enough electrodes implanted in the DLPFC, and two had enough electrodes implanted in the pSTC. In the post-childhood group, thirteen par>cipants had enough electrodes implanted in the DLPFC, and twenty-five had enough electrodes implanted in the pSTC. Experimental procedures A 6.5-minute short film was craged by edi>ng fragments from Pippi on the Run into a coherent narra>ve. The film is structured into 13 interleaved 30-second blocks of either speech or music, with seven blocks featuring background music only and six blocks retaining the original dialogue and voice from the video. Pa>ents were asked to watch the movie while the intracranial EEG signals were recorded. No fixa>on cross was displayed in the middle of the screen or elsewhere. The movie was presented using the Presenta>on sogware (Neurobehavioral Systems, Berkeley, CA) and the sound was synchronized with the neural recordings. More data acquisi>on details can be found in Berezutskaya et al.’s ar>cle24. iEEG data processing Electrode contacts and epochs contaminated with excessive ar>facts and epilep>form ac>vity were removed from data analysis by visual inspec>on. Raw data were filtered with a 50-Hz notch filter and re- referenced to the common average reference. For each electrode contact in each pa>ent, the preprocessed data were band-pass filtered (110– 140 Hz, 4th-order Bunerworth). The Hilbert transform was then applied to extract the analy>c amplitude. Each event (block) was extracted in the 0 to 30 s >me window around its onset. The figh music block was excluded, as there were no faces presented on screen. Subsequently, the data were down-sampled to 400 Hz and square-root transformed. Finally, the data were normalized by z-scoring with respect to baseline periods (−0.2 to 0 s before s>mulus onset). Contact Loca$on and Regions of Interest We iden>fied electrode contacts in STC in individual brains using individual anatomical landmarks (i.e., gyri and sulci). Superior temporal sulci and lateral sulci were used as boundaries. A coronal plane including the posterior >p of the hippocampus served as an anterior/posterior boundary. To iden>fy electrode contacts in DLPFC, we projected the electrode contact posi>ons provided by the open dataset .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted May 23, 2025. ; https://doi.org/10.1101/2025.05.23.655726doi: bioRxiv preprint onto Montreal Neurological Ins>tute-152 template brain (MNI) space, using FreeSurfer. DLPFC was defined based on the following sets of HCP-MMP134 labels on both leg and right hemispheres: 9-46d, 46, a9-46v, and p9-46v. Emo$on feature extrac$on Hume AI (hnps://www.hume.ai) was used to extract the facial emo>on features from the video. When mul>ple faces appeared in the movie, the maximum score of the facial expression features across all faces was used for each emo>on category. All the >me courses of facial emo>on features were resampled to 2Hz. No facial emo>on features were extracted for the figh music block due to the absence of faces. The full list of the 48 facial emo>on features is shown in Figure 4B. Encoding model fi>ng To model iEEG responses to emo>on, we used a linear regression approach with 48 facial emo>on features extracted by Hume AI. Time-lagged versions of each feature (with 0, 0.5 and 1-second delays) were used in the model fiqng. For each par>cipant, high-frequency broadband (HFB) responses from all electrode contacts within each area were concatenated. To match the temporal resolu>on of the emo>on feature >me course, the HFB responses were binned into 500 ms windows. We then modeled the processed HFB response for each par>cipant, each brain area, and each condi>on (speech vs music) using ridge regression. The op>mal regulariza>on parameter was assessed using 5-fold cross-valida>on, with the 20 different regulariza>on parameters (log spaced between 10 and 10000). To keep the scale of the weights consistent, a single best overall value of the regulariza>on coefficient was used for all areas in both the speech and music condi>ons in all pa>ents. We used cross-valida>on iterator to fit the model and test it on held-out data. The model performance was evaluated by calcula>ng Pearson correla>on coefficients between measured and predicted HFB response of individual brain areas. The mean predic>on accuracy (r value) of the encoding model with 5-fold cross-valida>on was then calculated. Non-parametric permuta>on tests were used to test whether the encoding model performance was significantly > 0 and whether there was a significant difference between groups. Specifically, we shuffled facial emo>on feature data in >me, and then we conducted the standard data analysis steps (described above) using the shuffled facial emo>on features. This shuffle procedure was repeated 5000 >mes to generate a null distribu>on, and p-values were calculated as the propor>on of results from shuffled data more extreme than the observed real value. A two-sided paired t-test was used to examine differences in encoding accuracy between speech and music condi>ons in post-childhood group (Figure 3D). Weight analysis To examine the correla>on between encoding model weights and age, we obtained 48 encoding model weights from all folds of cross-valida>on for all par>cipants whose pSTC significantly encoded facial expression (i.e. the p-value of predic>on accuracy is less than 0.05). Thus, 10 post-childhood individuals and 2 children were involved in the weight analysis. The weight for each feature represents its rela>ve contribu>on to predic>ng the neural response. A higher weight indicates that the corresponding feature has a stronger influence on neural ac>vity, meaning that varia>ons in this feature more significantly impact the predicted response. We used the absolute value of weights and therefore did not .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted May 23, 2025. ; https://doi.org/10.1101/2025.05.23.655726doi: bioRxiv preprint discriminate whether facial emo>on features were mapped to an increase or decrease in the HFB response.

The current study examines func>onal changes in both low-level and high-level brain areas across development to provide valuable insights into the neural mechanisms underlying the matura>on of facial expression percep>on. Based on our findings, we propose that young children rely primarily on early sensory areas rather than the prefrontal cortex for facial emo>on processing. As development progresses, the prefrontal cortex becomes increasingly involved, perhaps serving to modulate responses in early sensory areas based on emo>onal context and enabling them to process complex emo>ons. This developmental progression ul>mately enables the full comprehension of facial emo>ons in adulthood. Behavioral results suggest that infants as young as only 7–8 months can categorize some emo>ons 3–5. However, sensi>vity to facial expressions in young children does not mean that they can understand the meaning of that affec>ve state. For example, Kaneshige and Haryu (2014)35 found that although 4- month-old infants could discriminate facial configura>ons of anger and happiness, they responded posi>vely to both, sugges>ng that at this stage, they may lack knowledge of the affec>ve meaning behind these expressions. This underscores the idea that addi>onal processes need to be developed for children to fully grasp the emo>onal content conveyed by facial expressions. Although the neural mechanism behind this development is s>ll unclear, a reasonable perspec>ve is that it requires both visual processing of facial features and emo>on-related processing for the awareness of the emo>onal state of the other person17,36,37. Indeed, growing evidence suggests that the prefrontal cortex plays an important role in integra>ng prior knowledge with incoming sensory informa>on, allowing interpreta>on of the current situa>on in light of past emo>onal experience19,38. In the current study, we observed differen>al representa>on of facial expressions in the DLPFC between children and post-childhood individuals. First, in post-childhood individuals, neural ac>vity in the DLPFC encodes high-dimensional facial expression informa>on, whereas this encoding is absent in children. Second, while human voice enhances the representa>on of facial expressions in the DLPFC of post- childhood individuals, it instead reduces this representa>on in children. These results suggest that the DLPFC undergoes developmental changes in how it processes facial expressions. The absence of high- dimensional facial expression encoding in children implies that the DLPFC may not yet be fully engaged in emo>onal interpreta>on at an early age. Addi>onally, the opposite effects of human voice on facial expression representa>on indicate that mul>modal integra>on of social cues develops over >me. In post-childhood individuals, voices may enhance emo>onal processing by providing congruent informa>on39–41, whereas in children, the presence of voice might interfere with or redirect anen>onal resources away from facial expression processing 39,42,43. There have been few neuroimaging studies directly examining the func>onal role of young children’s DLPFC in facial emo>on percep>on. Some evidence suggest that the prefrontal cortex con>nues to develop un>l adulthood to achieve its mature func>on in emo>on percep>on9,20,44, and for some emo>on categories, this development may extend across the lifespan. For example, prefrontal cortex .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted May 23, 2025. ; https://doi.org/10.1101/2025.05.23.655726doi: bioRxiv preprint ac>va>on during viewing fearful faces increases with age 18,19,45. As there were not enough par>cipants for us to calculate correla>on between encoding model performance in DLPFC and age, it is s>ll unclear whether the representa>on of facial expression in DLPFC increase linearly with age. One possibility is that the representa>on of facial expressions in the DLPFC gradually increases with age un>l it reaches an adult-like level. This would suggest a con>nuous developmental trajectory, where incremental improvements in neural processing accumulate over >me. Another possibility is that development follows a more nonlinear panern, showing improvement with prominent changes at specific ages. Interes>ngly, research has shown that performance on matching emo>onal expressions improves steadily over development, with notable gains in accuracy occurring between 9 and 10 years and again between 13 and 14 years, ager which performance reaches adult-like levels 6. Although there are only two children in our sample with enough electrodes in pSTC, our results clearly showed that facial expression is encoded in each child’s pSTC. Moreover, the predic>on accuracy of the encoding model in the two children was comparable to or higher than the average level in the post- childhood group. In the 5-year-old child (S19) who had electrode coverage in both DLPFC and pSTC, facial expressions were represented in the pSTC but not in the DLPFC. This rare and fortunate sampling allows us to rule out the possibility that the low predic>on accuracy of the facial expression encoding model in the DLPFC is due to the reduced engagement in the movie-watching task for children. Consistent with our findings, previous studies have shown that the fusiform and superior temporal gyri are involved in emo>on-specific processing in 10-year-old children46. Meanwhile, some other researchers found that responses to facial expression in the amygdala and posterior fusiform gyri decreased as people got older20, but the use of frontal regions increased with age44. Therefore, we propose that early sensory areas like the fusiform and superior temporal gyri play a key role in facial expression processing in children, but their contribu>on may shig with age as frontal regions become more involved. Consistent with this perspec>ve, our results revealed that the encoding weights for complex emo>ons in pSTC increased with age, sugges>ng a developmental trajectory in the neural representa>on of complex emo>ons in pSTC. This finding aligns with previous behavioral studies showing that social complex emo>on recogni>on does not fully mature un>l young adulthood31–33. In fact, our results suggests that the representa>on of complex facial expressions in pSTC con>nues to develop over the lifespan. As for the correla>on between basic emo>on encoding and age, the lack of a significant effect in our study does not necessarily indicate an absence of developmental change but may instead be due to the limited sample size. In summary, our study provides novel insights into the neural mechanisms underlying the development of facial expression processing. As with any study, several limita>ons should be acknowledged. First, most electrode coverage in our study was in the leg hemisphere, poten>ally limi>ng our understanding of lateraliza>on effects. Second, while our results provide insights into the role of DLPFC during development, we were unable to examine other prefrontal regions, such as the orbitofrontal cortex (OFC) and anterior cingulate cortex (ACC), to examine their unique contribu>ons to emo>on processing. Lastly, due to sample size constraints, we were unable to divide par>cipants into more granular developmental stages, such as early childhood, adolescence, and adulthood, which could provide a more detailed characteriza>on of the neural mechanisms underlying the development of facial expression .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted May 23, 2025. ; https://doi.org/10.1101/2025.05.23.655726doi: bioRxiv preprint processing. Future studies using non-invasive methods, with more age-diverse samples will be essen>al for refining our understanding of how facial emo>on processing develops across the lifespan.

We would like to thank Dr. Julia Berezutskaya for providing the audiovisual film that was used for iEEG data collec>on. This work was supported by fundings from United States Na>onal Ins>tutes of Health (R01-MH127006). Author contribu.ons X.F. and A.T. performed the data analysis. X.F. and K.R.B wrote the paper. Compe.ng interests The author declares no compe>ng interests. Table 1. Demographic informa:on of childhood group. .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted May 23, 2025. ; https://doi.org/10.1101/2025.05.23.655726doi: bioRxiv preprint

1. Barrera, M.E., and Maurer, D. (1981). The percep>on of facial expressions by the three-month- old. Child Dev. 52, 203–206. 10.2307/1129231. 2. Walker-Andrews, A.S. (1997). Infants’ percep>on of expressive behaviors: differen>a>on of mul>modal informa>on. Psychol. Bull. 121, 437–456. 10.1037/0033-2909.121.3.437. Table 2. Demographic informa:on of post-childhood group. .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted May 23, 2025. ; https://doi.org/10.1101/2025.05.23.655726doi: bioRxiv preprint 3. Caron, R.F., Caron, A.J., and Myers, R.S. (1982). Abstrac>on of invariant face expressions in infancy. Child Dev. 53, 1008–1015. 4. Nelson, C.A., Morse, P .A., and Leavin, L.A. (1979). Recogni>on of facial expressions by seven- month-old infants. Child Dev. 50, 1239–1242. 10.2307/1129358. 5. Nelson, C.A., and Dolgin, K.G. (1985). The generalized discrimina>on of facial expressions by seven-month-old infants. Child Dev. 56, 58–61. 6. Kolb, B., Wilson, B., and Taylor, L. (1992). Developmental changes in the recogni>on and comprehension of facial expression: implica>ons for frontal lobe func>on. Brain Cogn. 20, 74–84. 10.1016/0278-2626(92)90062-q. 7. Romani-Sponchiado, A., Maia, C.P ., Torres, C.N., Tavares, I., and Arteche, A.X. (2022). Emo>onal face expressions recogni>on in childhood: developmental markers, age and sex effect. Cogn. Process. 23, 467–477. 10.1007/s10339-022-01086-1. 8. Rodger, H., Vizioli, L., Ouyang, X., and Caldara, R. (2015). Mapping the development of facial expression recogni>on. Dev. Sci. 18, 926–939. 10.1111/desc.12281. 9. Thomas, L.A., De Bellis, M.D., Graham, R., and LaBar, K.S. (2007). Development of emo>onal facial recogni>on in late childhood and adolescence. Dev. Sci. 10, 547–558. 10.1111/j.1467- 7687.2007.00614.x. 10. Johnston, P .J., Kaufman, J., Bajic, J., Sercombe, A., Michie, P .T., and Karayanidis, F. (2011). Facial emo>on and iden>ty processing development in 5- to 15-year-old children. Front. Psychol. 2, 26. 10.3389/fpsyg.2011.00026. 11. Lawrence, K., Campbell, R., and Skuse, D. (2015). Age, gender, and puberty influence the development of facial emo>on recogni>on. Front. Psychol. 6, 761. 10.3389/fpsyg.2015.00761. 12. Duchaine, B., and Yovel, G. (2015). A revised neural framework for face processing. Annu. Rev. Vis. Sci. 1, 393–416. 10.1146/annurev-vision-082114-035518. 13. Gomez, J., Barnen, M.A., Natu, V., Mezer, A., Palomero-Gallagher, N., Weiner, K.S., Amunts, K., Zilles, K., and Grill-Spector, K. (2017). Microstructural prolifera>on in human cortex is coupled with the development of face processing. Science 355, 68–71. 10.1126/science.aag0311. 14. Arcaro, M.J., Schade, P .F., Vincent, J.L., Ponce, C.R., and Livingstone, M.S. (2017). Seeing faces is necessary for face-domain forma>on. Nat. Neurosci. 20, 1404–1412. 10.1038/nn.4635. 15. Bernstein, M., and Yovel, G. (2015). Two neural pathways of face processing: A cri>cal evalua>on of current models. Neurosci. Biobehav. Rev. 55, 536–546. 10.1016/j.neubiorev.2015.06.010. 16. Engell, A.D., and Haxby, J.V. (2007). Facial expression and gaze-direc>on in human superior temporal sulcus. Neuropsychologia 45, 3234–3241. 10.1016/j.neuropsychologia.2007.06.022. .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted May 23, 2025. ; https://doi.org/10.1101/2025.05.23.655726doi: bioRxiv preprint 17. Pereira, M.R., Barbosa, F., de Haan, M., and Ferreira-Santos, F. (2019). Understanding the development of face and emo>on processing under a predic>ve processing framework. Dev. Psychol. 55, 1868–1881. 10.1037/dev0000706. 18. Yurgelun-Todd, D.A., and Killgore, W.D.S. (2006). Fear-related ac>vity in the prefrontal cortex increases with age during adolescence: a preliminary fMRI study. Neurosci. Len. 406, 194–199. 10.1016/j.neulet.2006.07.046. 19. Williams, L.M., Brown, K.J., Palmer, D., Liddell, B.J., Kemp, A.H., Olivieri, G., Peduto, A., and Gordon, E. (2006). The mellow years?: neural basis of improving emo>onal stability over age. J. Neurosci. 26, 6422–6430. 10.1523/JNEUROSCI.0022-06.2006. 20. Tessitore, A., Hariri, A.R., Fera, F., Smith, W.G., Das, S., Weinberger, D.R., and Manay, V.S. (2005). Func>onal changes in the ac>vity of brain regions underlying emo>on processing in the elderly. Psychiatry Res. 139, 9–18. 10.1016/j.pscychresns.2005.02.009. 21. Yurgelun-Todd, D. (2007). Emo>onal and cogni>ve changes during adolescence. Curr. Opin. Neurobiol. 17, 251–257. 10.1016/j.conb.2007.03.009. 22. Hariri, A.R., Bookheimer, S.Y ., and Mazziona, J.C. (2000). Modula>ng emo>onal responses: effects of a neocor>cal network on the limbic system. Neuroreport 11, 43–48. 10.1097/00001756- 200001170-00009. 23. Hariri, A.R., Manay, V.S., Tessitore, A., Fera, F., and Weinberger, D.R. (2003). Neocor>cal modula>on of the amygdala response to fearful s>muli. Biol. Psychiatry 53, 494–501. 10.1016/s0006-3223(02)01786-9. 24. Berezutskaya, J., Vansteensel, M.J., Aarnoutse, E.J., Freudenburg, Z.V., Piantoni, G., Branco, M.P ., and Ramsey, N.F. (2022). Open mul>modal iEEG-fMRI dataset from naturalis>c s>mula>on with a short audiovisual film. Sci. Data 9, 91. 10.1038/s41597-022-01173-0. 25. Desikan, R.S., Ségonne, F., Fischl, B., Quinn, B.T., Dickerson, B.C., Blacker, D., Buckner, R.L., Dale, A.M., Maguire, R.P ., Hyman, B.T., et al. (2006). An automated labeling system for subdividing the human cerebral cortex on MRI scans into gyral based regions of interest. Neuroimage 31, 968– 980. 10.1016/j.neuroimage.2006.01.021. 26. Flack, T.R., Andrews, T.J., Hymers, M., Al-Mosaiwi, M., Marsden, S.P ., Strachan, J.W.A., Trakulpipat, C., Wang, L., Wu, T., and Young, A.W. (2015). Responses in the right posterior superior temporal sulcus show a feature-based response to facial expression. Cortex 69, 14–23. 10.1016/j.cortex.2015.03.002. 27. Fan, X., Wang, F., Shao, H., Zhang, P., and He, S. (2020). The bonom-up and top-down processing of faces in the human occipitotemporal cortex. eLife 9. 10.7554/eLife.48764. 28. Alba-Ferrara, L., Hausmann, M., Mitchell, R.L., and Weis, S. (2011). The neural correlates of emo>onal prosody comprehension: disentangling simple from complex emo>on. PLoS ONE 6, e28701. 10.1371/journal.pone.0028701. .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted May 23, 2025. ; https://doi.org/10.1101/2025.05.23.655726doi: bioRxiv preprint 29. Burnen, S., Thompson, S., Bird, G., and Blakemore, S.-J. (2011). Pubertal development of the understanding of social emo>ons: Implica>ons for educa>on. Learn. Individ. Differ. 21, 681–689. 10.1016/j.lindif.2010.05.007. 30. Russell, J.A., and Paris, F.A. (1994). Do Children acquire Concepts for Complex Emo>ons Abruptly? Int. J. Behav. Dev. 17, 349–365. 10.1177/016502549401700207. 31. Garcia, N.V., and Scherf, K.S. (2015). Emerging sensi>vity to socially complex expressions: A unique role for adolescence? Child Dev. Perspect. 9, 84–90. 10.1111/cdep.12114. 32. Burnen, S., Bird, G., Moll, J., Frith, C., and Blakemore, S.-J. (2009). Development during adolescence of the neural processing of social emo>on. J. Cogn. Neurosci. 21, 1736–1750. 10.1162/jocn.2009.21121. 33. Mona-Mena, N.V., and Scherf, K.S. (2017). Pubertal development shapes percep>on of complex facial expressions. Dev. Sci. 20. 10.1111/desc.12451. 34. Glasser, M.F., Coalson, T.S., Robinson, E.C., Hacker, C.D., Harwell, J., Yacoub, E., Ugurbil, K., Andersson, J., Beckmann, C.F., Jenkinson, M., et al. (2016). A mul>-modal parcella>on of human cerebral cortex. Nature 536, 171–178. 10.1038/nature18933. 35. Kaneshige, T., and Haryu, E. (2015). Categoriza>on and understanding of facial expressions in 4- month-old infants. Jpn Psychol Res 57, 135–142. 10.1111/jpr.12075. 36. Pollak, S.D., Messner, M., Kistler, D.J., and Cohn, J.F. (2009). Development of perceptual exper>se in emo>on recogni>on. Cogni>on 110, 242–247. 10.1016/j.cogni>on.2008.10.010. 37. Ruba, A.L., and Pollak, S.D. (2020). The development of emo>on reasoning in infancy and early childhood. Annu. Rev. Dev. Psychol. 2, 503–531. 10.1146/annurev-devpsych-060320-102556. 38. Bany, M., and Taylor, M.J. (2006). The development of emo>onal face processing during childhood. Dev. Sci. 9, 207–220. 10.1111/j.1467-7687.2006.00480.x. 39. Klasen, M., Chen, Y .-H., and Mathiak, K. (2012). Mul>sensory emo>ons: percep>on, combina>on and underlying neural processes. Rev. Neurosci. 23, 381–392. 10.1515/revneuro-2012-0040. 40. Collignon, O., Girard, S., Gosselin, F., Roy, S., Saint-Amour, D., Lassonde, M., and Lepore, F. (2008). Audio-visual integra>on of emo>on expression. Brain Res. 1242, 126–135. 10.1016/j.brainres.2008.04.023. 41. Kreifelts, B., Ethofer, T., Grodd, W., Erb, M., and Wildgruber, D. (2007). Audiovisual integra>on of emo>onal signals in voice and face: an event-related fMRI study. Neuroimage 37, 1445–1456. 10.1016/j.neuroimage.2007.06.020. 42. Weissman, D.H., Warner, L.M., and Woldorff, M.G. (2004). The neural mechanisms for minimizing cross-modal distrac>on. J. Neurosci. 24, 10941–10949. 10.1523/JNEUROSCI.3669-04.2004. .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted May 23, 2025. ; https://doi.org/10.1101/2025.05.23.655726doi: bioRxiv preprint 43. Jacob, H., Brück, C., Domin, M., Lotze, M., and Wildgruber, D. (2014). I can’t keep your face and voice out of my head: neural correlates of an anen>onal bias toward nonverbal emo>onal cues. Cereb. Cortex 24, 1460–1473. 10.1093/cercor/bhs417. 44. Gunning-Dixon, F.M., Gur, R.C., Perkins, A.C., Schroeder, L., Turner, T., Turetsky, B.I., Chan, R.M., Loughead, J.W., Alsop, D.C., Maldjian, J., et al. (2003). Age-related differences in brain ac>va>on during emo>onal face processing. Neurobiol. Aging 24, 285–295. 10.1016/s0197-4580(02)00099- 4. 45. Monk, C.S., McClure, E.B., Nelson, E.E., Zarahn, E., Bilder, R.M., Leibenlug, E., Charney, D.S., Ernst, M., and Pine, D.S. (2003). Adolescent immaturity in anen>on-related brain engagement to emo>onal facial expressions. Neuroimage 20, 420–428. 10.1016/s1053-8119(03)00355-0. 46. Lobaugh, N.J., Gibson, E., and Taylor, M.J. (2006). Children recruit dis>nct neural systems for implicit emo>onal face processing. Neuroreport 17, 215–219. 10.1097/01.wnr.0000198946.00445.2f. .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted May 23, 2025. ; https://doi.org/10.1101/2025.05.23.655726doi: bioRxiv preprint