{"paper_id":"2ad6b1b8-593a-457f-bfb3-81d6b08f2f31","body_text":"1 \nDevelopment of functional topography of the default mode subnetworks 2 \nrevealed by precision mapping 3 \n 4 \n Ying He1†*, Jonah Kember1†, Hongxiu Jiang1, Xiaoqian J. Chai1 5 \n 6 \n1. McGill University, Department of Neurology and Neurosurgery, McGill University, 7 \nMontréal, QC H3A 2B4. 8 \n†       These authors contributed equally to this work 9 \nCorresponding author: Ying He (ying.he3@mcgill.ca) 10 \n 11 \n 12 \nAbstract 13 \nThe default mode network (DMN) is a functionally and anatomically heterogeneous 14 \nnetwork comprising distinct subsystem s that support internally directed processes 15 \nsuch as memory and self -related thought. Specialization and segregation of 16 \nfunctional brain networks are critical for cognitive maturation, yet how DMN 17 \nsubnetworks develop remains unclear . Group-average approaches obscure 18 \nindividual variability and blur spatial details, limiting precise  characterization of 19 \nnetwork topography. Using precision mapping in 547 participants aged 5 –21 years, 20 \nwe delineated individualized DMN subnetworks and examined their development 21 \nacross multiple aspects, including functional segregation and spatial topograp hy. 22 \nWe found that DMN subnetworks become increasingly functionally segregated 23 \nfrom each other and from non -DMN networks with age, and exhibit greater 24 \ntopographic distinctiveness through boundary sharpening and reduced spatial 25 \noverlap. Furthermore, we obser ved a selective spatial contraction of the memory -26 \nrelated subnetwork, which correlated with better episodic memory . Our findings 27 \nunderscore the importance of characterizing multidimensional functional and 28 \ntopographical features at the individual level to b etter understand typical 29 \nneurodevelopment and neurodevelopmental and psychiatric conditions.  30 \n 31 \n 32 \n 33 \n 34 \n 35 \n 36 \n 37 \n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted February 25, 2026. ; https://doi.org/10.64898/2026.02.24.707618doi: bioRxiv preprint \n\nIntroduction 38 \nThe default mode network (DMN) comprises interconnected and distributed brain 39 \nregions and is thought to be involved in internal mental processes, such as 40 \nremembering, thinking about future, and self/other -related thoughts1, 2. Disruptions 41 \nof the DMN have been shown in many neuropsychiatric and developmental 42 \ndisorders3, 4, 5. Emerging evidence suggests that the DMN is a complex network that 43 \nconsists of distinct subnetworks, each supporting specific cognitive processes 6, 7, 8. 44 \nDuring child development, the broader functional brain networks become 45 \nincreasingly segregated —characterized by stronger within -network and weaker 46 \nbetween-network connectivity)9, 10, 11. The DMN as a whole becomes more internally 47 \nintegrated while becoming increasingly segregated from other functional networks 48 \nduring development 12, 13, 14 . This growing  segregation supports cognitive 49 \ndevelopment by enabling more specialized network functioning13, 15. However, how 50 \nfunctional segregation between distinct DMN subnetworks develop in children 51 \nremains unknown.  Delineating th e development of DMN subnetworks in a 52 \nneurotypical population is essential, as it provides a crucial baseline for 53 \nunderstanding alterations in related neuropsychiatric disorders.  54 \nTo date, investigations of functional networks in children have largely relie d on 55 \ngroup-average (one -size-fits-all) parcellations. These group -average approaches 56 \nintroduce spatial blurring and fail to capture individual -specific features of brain 57 \norganization. This limitation has motivated the growing use of precision mapping 58 \napproaches, which identify functional networks at the individual level. By capturing 59 \nindividual variability, these methods offer improved reliability in linking brain 60 \norganization to behavior and symptoms 16, 17 . Furthermore, inter -individual 61 \nvariability in functional brain networks may change across development 18. In 62 \naddition, studies using within -person precision mapping approaches have shown 63 \nthat the individual topography of functional networks — including their size, shape, 64 \nand spatial location — differs markedly from the group average 19, 20 . These 65 \ntopographical variations become more refined during youth development and are 66 \nassociated with individual differences in cognition21, 22, with comparable alterations 67 \nreported in clinical populations 23, 24 . Thus, in addition to assessing within - and 68 \nbetween-network connectivity strength and other graph -based measures, this 69 \napproach also enables the examination of individual topographic features and spatial 70 \nextent of DMN subnetworks.  71 \n 72 \nIn the present study, we performed individual-level network parcellation in a cohort 73 \nof 547 participants aged 5–21 years, each with more than 21 minutes of resting-state 74 \nfMRI data. Using the Yeo 17 -network template as a reference, we identified three 75 \ndistinct DMN subnetworks (DMNA, DMNB, DMNC) based on individual 76 \n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted February 25, 2026. ; https://doi.org/10.64898/2026.02.24.707618doi: bioRxiv preprint \n\nfunctional connectivity profiles at the vertex level. A previous meta -analysis of the 77 \nfunctional pro files of the three DMN subnetworks found that DMNA was more 78 \nstrongly associated with self -related processes, emotion and evaluation, as well as 79 \nsocial and mnemonic processes shared with the other two subnetworks 25. DMNB 80 \nwas linked to mentalizing and social cognition, as well as story comprehension and 81 \nsemantic/conceptual processing. In contrast, DMNC showed strong associations 82 \nwith past and future autobiographical thought, episodic memory, and contextual 83 \nretrieval. To understand the development of these three DMN subnetworks, we first 84 \nexamined functional segregation of the three DMN subnetworks, quantified by the 85 \nsegregation index derived from the average within -network and between -network 86 \nfunctional connectivity. This allowed us to assess whether the subnetworks became 87 \nincreasingly segregated from one another and from other non -DMN networks with 88 \nage. Second, we mapped the topography of the three subnetworks individually and 89 \nexamined the topographical segregation, me asured by boundary refinement and 90 \nspatial differentiation between the subnetworks. Third, we examined age -related 91 \nchanges in the surface area of each subnetwork to determine whether their cortical 92 \nrepresentations expanded or contracted during development. After examining age 93 \neffects on these segregation indices and spatial extent, we tested whether these 94 \ndevelopmental changes in brain network organization could predict corresponding 95 \ncognitive development. Given that DMNC is relatively specific to episodic memory, 96 \nwe examined correlations between memory performance and DMNC metrics.  97 \n 98 \nResults 99 \nDMN subnetworks show increasing functional segregation with age  100 \nHere we employed a template -matching method to obtain individual parcellations 101 \nof three DMN subnetworks (Figure 1). After obtaining the individually mapped  102 \n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted February 25, 2026. ; https://doi.org/10.64898/2026.02.24.707618doi: bioRxiv preprint \n\nFigure 1. Overview of the precision mapping method. A. First, we constructed the functional 103 \nconnectivity profile for each vertex by calculating the Pearson correlation between that vertex and 104 \nall other vertices. Next, we computed the Dice similarity metric to quantify the spatial similarity 105 \nbetween each vertex’s connectivity profile and each of Yeo’s 17 networks. Finally, each verte x 106 \nwas assigned to the network with which it showed the highest similarity. B. The left panel shows 107 \nthe template of the DMN subnetworks based on the Yeo 17-network parcellation. The right panel 108 \ndisplays three representative participants selected from different age groups, illustrating individual 109 \ndifferences in the topographical distribution of the DMN subnetworks. 110 \n 111 \nfunctional networks, we first calculated the within -network homogeneity and 112 \ncompared it to that obtained using a group -level template. The homogeneity of 113 \nindividual networks was significantly higher than that of the group template ( t546 = 114 \n15.42, p < 0.001), indicating the effectiveness of our precision mapping method. In 115 \naddition, we computed the Adjusted Rand Index between each subject’s parcellation 116 \nunder a given threshold and the parcellations of the same and other subjects under 117 \ndifferent thresholds. We observed that parcellations from different thresholds within 118 \nthe same subject showed higher similarity than those between different subjec ts 119 \n(Figure S1 ), suggesting that our precision mapping method produces consistent 120 \nresults within individuals and is robust to parameter variations.  121 \n 122 \nWe quantified functional segregation by computing a segregation index based on 123 \nwithin-network and between -network functional connectivity. Within the DMN, 124 \neach subnetwork showed significantly increased functional segregation from the 125 \nother two subnetworks with age (Figure 2A, DMNA: β = 0.005, t541 = 6.37, pcorrected 126 \n< 0.001; DMNB: β = 0.008, t541 = 7.65, pcorrected < 0.001; DMNC: β = 0.008, t541 = 127 \n6.76, pcorrected < 0.001). These findings demonstrate progressive functional 128 \nsegregation among all three DMN subnetworks across development.  129 \nIn addition, we examined the age-related changes in functional segregation between 130 \neach DMN subnetwork and the other 16 large-scale functional networks. Similar to 131 \nsegregation relative to other subnetworks of DMN, all DMN subnetworks showed 132 \nsignificant increases in segregation from the other 16 networks with age (Figure 2B, 133 \nDMNA: β = 0.010, t541 = 8.13, pcorrected < 0.001; DMNB: β = 0.007, t541 = 6.42, 134 \npcorrected < 0.001; DMNC: β = 0.006, t541 = 6.61, pcorrected < 0.001). We also tested 135 \nwhether increasing functional segregation of DMNC from the other two DMN 136 \nsubnetworks and from all other 16 networks correlated with PSM scores, but no 137 \nsignificant correlations were found (all ps ≥ 0.42). These findings further highlight 138 \nthe developmental emergence of functional differentiation of DMN subnetworks. 139 \n 140 \n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted February 25, 2026. ; https://doi.org/10.64898/2026.02.24.707618doi: bioRxiv preprint \n\n 141 \n 142 \nFigure 2. DMN subnetworks show increasing functional segregation across development. A. The 143 \nsegregation of each DMN subnetwork relative to other DMN subnetworks significantly increased 144 \nwith age. B. The segregation of each DMN subnetwork from all other large -scale functi onal 145 \nnetworks also showed a significant age -related increase. Shaded bars reflect 95% confidence 146 \nintervals of the regression line. 147 \n 148 \nTopographical Boundaries of DMN Subnetworks Become Sharper with Age  149 \nHere, we examined the spatial distribution of the three DMN subnetworks among 150 \nthe participants using the density map (Figure 3A). We found that the highest density 151 \nregions of each subnetwork were primarily located within their respective canonical 152 \ncortical zones, but also extended into regions typically associa ted with other 153 \nsubnetworks, albeit at lower densities. For instance, the posterior cingulate cortex 154 \nand precuneus regions were typically assigned to DMNA, although some individuals 155 \nshowed assignments to DMNB. The anterior temporal regions were generally 156 \nassociated with DMNB, but a substantial proportion of participants had anterior 157 \ntemporal regions that were assigned to DMNA. Similarly, the ventral medial 158 \nprefrontal cortex was most often assigned to DMNA, yet assignments to DMNB and 159 \nDMNC were also observed in some individuals. This pattern reflects both individual 160 \nvariability and their shared characteristics, with the three networks being distributed 161 \nyet adjacent. 162 \nFurthermore, we aimed to investigate whether the topographical boundaries between 163 \nDMN subnetworks become sharper with age. To do so, we first identified boundary 164 \nvertices based on their proximity (within 5 mm) to the borders between adjacent 165 \nsubnetwork clusters (Figure 3B). Boundary sharpness was defined as the extent to 166 \nwhich these vertices showed stronger functional connectivity with their own 167 \nsubnetwork compared to the neighboring one, indicating a clearer distinction 168 \n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted February 25, 2026. ; https://doi.org/10.64898/2026.02.24.707618doi: bioRxiv preprint \n\nbetween adjacent subnetworks. Here, we present two examples illustrating a blurred 169 \nversus a sharp boundary (Figure 3C). We examined  age-related changes in mean 170 \nboundary sharpness for each DMN subsystem, calculated by averaging the 171 \nsharpness values of the boundaries between the given subnetwork and the other two 172 \nsubnetworks. DMNA showed a significant increase in mean boundary sharpness  173 \nwith age ( β = 0.01, t541 = 4.91, pcorrected < 0.001, Figure 3D), driven by sharpness 174 \nincreases at both its interfaces with DMNB ( β = 0.01, t541 = 3.65, pcorrected < 0.001) 175 \nand DMNC ( β = 0.01, t541 = 2.78, pcorrected = 0.010, Figure S2). We didn’t find 176 \nsignificant age-related changes in mean boundary sharpness in DMNB ( β = 0.006, 177 \nt541 = 1.77, pcorrected = 0.114) or DMNC (β = 0.002, t541 = 0.34, pcorrected = 0.731) after 178 \nFDR correction. This pattern of results suggests that DMNA boundaries undergo 179 \nprogressive sharpening during development, whereas DMNB and DMNC 180 \nboundaries remain relatively stable.   181 \n 182 \nFigure 3 Topographical boundaries of DMN subnetworks became sharper with development. A. 183 \nThe density maps indicated the proportion of subjects for whom a giv en vertex was assigned to 184 \neach of the three DMN subnetworks. The spatial locations of the three DMN subnetworks were 185 \nprimarily centered around their respective centroids within each cortical zone, but they also 186 \nextended into regions typically associated with other subnetworks. B. Two clusters from different 187 \nDMN subnetworks were selected as examples to illustrate how boundary sharpness was quantified. 188 \nFor each border vertex between the two clusters, we identified all vertices located within 5 mm 189 \nand used them for the analysis. C. Two example subjects illustrating boundary sharpness between 190 \nDMNA and DMNB. Red outline indicates the DMNB cluster, and the yellow outline indicates the 191 \nDMNA cluster. The color bar represents the functional connectivity strength of each vertex to the 192 \nmean time series of the DMNA cluster. The top panel shows a subject with a blurred boundary, 193 \nwhere vertices along the border exhibit highly similar connectivity to both subnetworks. The 194 \nbottom panel shows a subject with a sharper boundary , where border vertices display distinct 195 \n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted February 25, 2026. ; https://doi.org/10.64898/2026.02.24.707618doi: bioRxiv preprint \n\nconnectivity profiles between the two subnetworks. Note that while most cluster vertices are 196 \nshown for visualization, sharpness was calculated only for border vertices within 5 mm (panel B). 197 \nD. The scatter plot show s that DMNA boundaries with both other subnetworks became 198 \nsignificantly sharper with age. 199 \n 200 \nSpatial overlap of DMN subnetworks decreases with age  201 \nAfter demonstrating the refinement of boundaries with age, we next sought to 202 \ninvestigate the spatial differentiation among DMN subnetworks. We defined spatial 203 \noverlap between DMN subnetworks as vertices that simultaneously showed high 204 \nfunctional connectivity with the mean time series of two subnetworks, where \"high\" 205 \nwas defined as connectivity exceeding t he subject -specific mean by at least one 206 \nstandard deviation. We then examined how the surface area of these overlapping 207 \nregions changed with age. Among the subnetworks, the spatial overlap between 208 \nDMNA and DMNB exhibited the highest surface area proportion , and this overlap 209 \nwas significantly negatively correlated with age ( β = -0.003, t541 = -2.53, pcorrected = 210 \n0.017, Figure 4A). Similarly, the overlap between DMNB and DMNC also showed 211 \na significant negative correlation with age (β = -0.002, t541 = -2.71, pcorrected = 0.017, 212 \nFigure 4B). The spatial overlap between DMNA and DMNC did not show a 213 \nsignificant relationship with age ( β = 6.96E -05, t541 = 0.07, pcorrected = 0.947). 214 \nAdditional analyses across different correlation thresholds for defining high spatial  215 \noverlap (from the subject -specific mean, to two standard deviations above) 216 \nconfirmed the robustness of the age-related decrease in overlap between DMNA and 217 \nDMNB (Figure S3). These findings indicate that the spatial distribution of the DMN 218 \nsubnetworks becomes more differentiated during development, especially between 219 \nDMNA and DMNB. 220 \n 221 \n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted February 25, 2026. ; https://doi.org/10.64898/2026.02.24.707618doi: bioRxiv preprint \n\n 222 \nFigure 4. Overlap of DMN subnetworks decreased with age. A. Two examples illustrate the spatial 223 \noverlap between DMNA and DMNB from a younger and an older participant, respectively. Only 224 \nthe left hemisphere is displayed for visual clarity. Analyses were conducted on both hemispheres. 225 \nThe accompanying scatter plot demonstrates a significant decrease in overlap with age. B. Two 226 \nexamples show the spatial overlap between DMNB and DMNC from a younger and an older 227 \nparticipant. The scatter plot again indicates a significant age-related decline in overlap.  228 \nSurface area of DMNC decreases with age 229 \nWe first examined age -related changes in the surface area of DMN subnetworks 230 \nusing a GLM, controlling for sex and scanning sites. After correction for multiple 231 \ncomparisons, we observed a significant age -related decrease in the surface area of 232 \nthe DMNC subn etwork ( β = –0.04, t541 = –4.00, pcorrected < 0.001). No significant 233 \nassociations with age were found for the DMNA ( β = 0.01, t541 = 0.58, pcorrected = 234 \n0.63) or DMNB (β = –0.01, t541 = –0.94, pcorrected = 0.45) subnetworks (Figure 5A). 235 \nWhen treating the DMN as a whole network, its overall surface area does not exhibit 236 \nsignificant age-related changes ( β = -0.02, t541 = -1.53, pcorrected = 0.19). To further 237 \nillustrate age -related changes in DMNC surface ar ea, subjects were divided into 238 \nthree age groups based on age: children (≤12 years, n=147), adolescents (12 -18 239 \nyears, n=262), and adults (≥18 years, n=138). The corresponding density maps show 240 \nthat as age increases, the surface area of the DMNC gradually de creases and 241 \n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted February 25, 2026. ; https://doi.org/10.64898/2026.02.24.707618doi: bioRxiv preprint \n\nbecomes more spatially constrained (Figure 5B). We also observed a significant age-242 \nrelated reduction of DMNC in the ventromedial prefrontal cortex (vmPFC), a region 243 \ntypically associated with the DMNA (Figure 5B). This pattern suggests a spatial  244 \nreorganization of DMN subnetworks during development, with DMNC territory 245 \npotentially redistributing to other subnetworks in adulthood.  246 \nTo further characterize the spatial organization of age -related changes in 247 \nindividualized DMNC, we compared each participant's DMNC map with the group-248 \nlevel template as a reference. This approach enabled us to identify which networks 249 \noccupy the regions where the DMNC contracts during development, relative to the 250 \ncanonical network spatial distribution . We first  quantified t he spatial overlap 251 \nbetween each participant's DMNC and the group -level DMNA. As expected, the 252 \nproportion of individual DMNC overlapping with group-level DMNA (overlap area 253 \n/ individual DMNC area) significantly decreased with age (β = –0.001, t403 = –4.92, 254 \np < 0.001, Figure 5C) . This indicates that regions initially occupied by DMNC in 255 \nyounger children are progressively reassigned to canonical DMNA territory.  In 256 \ncontrast, when examining the overlap between youth DMNC and adult group -level 257 \nDMNB, nearly all subjects showed minimal to no overlap (Figure S4), confirming 258 \nthat the developmental changes in DMNC are relatively spatially specific to DMNA-259 \nassociated regi ons. Collectively, these findings demonstrate age -related spatial 260 \nreorganization among DMN subnetworks, whereby cortical territories undergo 261 \nsystematic reassignment during development. 262 \n 263 \nTo further examine whether the DMNC subnetwork contracting is associat ed with 264 \nthe development of episodic memory, we also examined the relationship between 265 \nPSM scores and the surface area of DMN subnetworks, controlling for age, sex, and 266 \nscanning site. A significant positive association was observed between age and PSM 267 \nscore (β = 0.71, t396 = 8.05, p < 0.001), indicating that episodic memory performance 268 \nimproves with development. We found that the surface area of the DMNC 269 \nsubnetwork was significantly negatively correlated with PSM scores ( β = -1.17, t395 270 \n= -2.27, p = 0.02), indicating that a reduced surface area was associated with better 271 \nepisodic memory performance. These findings suggest that during development, 272 \n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted February 25, 2026. ; https://doi.org/10.64898/2026.02.24.707618doi: bioRxiv preprint \n\nDMNC exhibits a more compact cortical organization compared to other 273 \nsubnetworks, which may underlie its link to memory performance (Figure 5D). 274 \n 275 \nFigure 5. Surface area of DMNC decreased with age and memory performance. A. This bar plot 276 \nillustrates the effect of age on the surface area of DMN subnetworks, based on a general linear 277 \nmodel in which age was treated as th e independent variable and surface area as the dependent 278 \nvariable, while controlling for confounding factors (sex, site, etc.). The width of each bar reflects 279 \nthe regression coefficient for age. Asterisks indicate levels of statistical significance after F DR 280 \ncorrection: **p < 0.001; *p < 0.01; p < 0.05. B. Panel B shows the density maps for different age 281 \ngroups. Density values represent the proportion of subjects within each group in which a given 282 \nvertex was assigned to the DMNC. C. The left panel shows two example subjects, illustrating the 283 \nspatial overlap between individual DMNC (blue) and the group -level DMNA from template 284 \n(yellow). The scatter plot demonstrates an age-related decrease in the overlap proportion (overlap 285 \narea / individual's total DMNC area), indicating that as DMNC shrinks during development, it is 286 \nprimarily the regions that correspond to canonical DMNA that are lost.  D. A negative association 287 \nwas observed between the DMNC surface area proportion (individual’s DMNC area / individual’s 288 \ntotal cortical area) and Picture Sequence Memory performances. 289 \n 290 \nDiscussion 291 \nIn this study, we systematically examined the devel opmental segregation and 292 \nspecialization of DMN subnetworks in 547 participants aged 5 -21 years, using 293 \nprecision mapping to obtain individualized parcellations of DMN subnetworks. Our 294 \nfindings reveal developmental segregation occurring across multiple dimen sions. 295 \nFirst, the three DMN subnetworks exhibited increased functional segregation from 296 \n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted February 25, 2026. ; https://doi.org/10.64898/2026.02.24.707618doi: bioRxiv preprint \n\neach other and from all other networks with age. Second, topographical segregation 297 \nwas observed: boundaries between DMNA and the other two subnetworks became 298 \nsignificantly sharper with age, accompanied by reduced spatial overlap between 299 \nDMNA and DMNB. Third, we observed a selective contraction of DMNC that 300 \ncorrelated with better episodic memory, with the contracted regions predominantly 301 \noverlapping with canonical DMNA te rritory. Together, these findings provide a 302 \ncomprehensive characterization of DMN subnetwork development, encompassing 303 \nprogressive functional and topographical segregation, and selective spatial 304 \nfocalization. 305 \n  306 \nHow to identify the DMN subnetworks remains a n area of active research 26, 27. A 307 \ncommonly used approach is to divide it into three subnetworks, as proposed by 308 \nAndrews-Hanna et al. (2010 , 2014)7, 25, with similar parcellations also re ported by 309 \nYeo et al. (2011) 28. DMNA is associated with self -related processes, emotion, and 310 \nevaluation, while also sharing social and mnemonic functions with DMNB and 311 \nDMNC. DMNB is linked to social processing, whereas DMNC shows relative 312 \nspecificity to episodic memory 25. Based on these prior findings,  we decided to 313 \nparcellate the DMN into three subsystems, following Yeo et al.  Previous studies 314 \nhave examined the functional segregation of the DMN by treating it as a single 315 \nnetwork. For example, one study reported increased within-network connectivity of 316 \nthe entire DMN during rest from childhood to adolescence 12. Similar effects have 317 \nbeen observed in children during task states 13. More recently, a large -scale 318 \npopulation study found that functional segregation of the canonical DMN, measured 319 \nusing the same index, increases rapidly with age and peaks at the end of the third 320 \ndecade14. Only one previous study has investigated the segregation of DMN 321 \nsubnetworks in a developmental sample, comparing the modularity of DMN 322 \nsubsystems between individuals with autism and neurotypical controls29. However, 323 \nthis study employed an ROI-to-ROI approach with relatively limited ROI coverage 324 \nand the same ROI definitions across all participants, which could have overlooked 325 \ninter-individual variations and thereby contributed to the finding of no marked age -326 \nrelated changes in the neurotypical group29. Our approach overcomes this limitation 327 \nand reveals developmental effects showing that the DMN, a heterogeneous network 328 \nas established by functional –anatomical evidence 7, 25, 28, 30, 31, 32 , becomes 329 \nincreasingly heterogeneous and functionally segregated over the course of 330 \ndevelopment. Segregation of ass ociative networks is consistently associated with 331 \nenhanced cognitive performance, while the loss of such segregation in later life has 332 \nbeen implicated in cognitive decline 9, 33, 34, 35, 36, 37, 38 . DMN subnetworks share a 333 \ncommon organizational motif that supports internally generated representations, 334 \nwith each network specialized for distinct processing domains 27, and their 335 \n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted February 25, 2026. ; https://doi.org/10.64898/2026.02.24.707618doi: bioRxiv preprint \n\nsegregation with development may reflect the enhancement of specific cognitive 336 \nfunctions. 337 \n  338 \nWe also observed an age-related increase in segregation in the topography of DMN 339 \nsubnetworks.  Our precision mapping results reveal individual parcellations with 340 \nclear and specific spatial boundaries. Given that the three DMN subnetworks are 341 \nspatially adjacent to one another we examined the age -related effects on boundary 342 \nsharpness between subnet works and found that the boundaries around DMNA 343 \nbecome progressively sharper across development. At the same time, we found that 344 \nthe spatial overlap between DMNA and DMNB decreases with age, suggesting that 345 \nDMNA gradually emerges as a more spatially distin ct and clearly delineated 346 \nsubnetwork. Notably, DMNA showed the most pronounced age -related changes in 347 \ntopographical segregation among all three DMN subnetworks. DMNA encompasses 348 \nthe so -called midline core regions of the DMN, including the anterior medial 349 \nprefrontal cortex, angular gyrus, and posterior cingulate cortex, which are among 350 \nthe most consistently engaged regions within the network 7, 25 . Previous group -351 \naveraged studies examining functional connectivity between pairs of DMN nodes 352 \nhave reported age-related increases primarily in midline core regions, which is 353 \nconsistent with our findings 12. Convergent structural and functional connectivity 354 \nanalyses in prior work also suggest that PCC –mPFC connectivity (largely 355 \noverlapping with DMNA) represents the most immature link within the DMN in 356 \nchildren, further highlighting the pro nounced age -related effects 39. Importantly, 357 \nprevious studies have demonstrated that behavioral phenotypes across cognition, 358 \npersonality, and emotion can be predicted by individual -specific network 359 \ntopography with modest accuracy, comparable to predictions based on connectivity 360 \nstrength40. Our study advances this line of research by introducing two novel 361 \nmeasures of topographical segregation —boundary sharpness and spatial overlap —362 \nthat uncover new insights into the developmental segregation of DMN subnetworks. 363 \nThese topographical measures, enabled by within -person precision mapping, may 364 \nprovide new biomarkers for predicting behavioral outcomes  and developmental 365 \ntrajectories at the individual level, complementing traditional connectivity -based 366 \napproaches. 367 \nFinally, we examined age-related effects on the surface area of DMN subnetworks. 368 \nWhile many studies have investigated developmental trajectories of total or regional 369 \ncortical surface area based on group averages 41, 42, 43 , research focusing on the 370 \nsurface area of individual functional networks remains limited. One study on human 371 \nlifespan functional connectome changes, using personalized ma pping, reported a 372 \nslight expansion of the whole DMN size during the first month of life but no 373 \nsignificant change thereafter14. These findings are consistent with ours. Specifically, 374 \n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted February 25, 2026. ; https://doi.org/10.64898/2026.02.24.707618doi: bioRxiv preprint \n\nwhen we first tested the proportion of the whole DMN surface area relative to the 375 \ntotal cortical surface area, we found no significant age -related effects. This result 376 \nhighlights that treating the DMN as a single entity may obscure the heterogeneous 377 \ndevelopmental trajectories of its subnetworks. When we examined each subnetwork 378 \nseparately, we found that the surface area of DMNC progressively decreased with 379 \nage. This reduction  was predominantly localized to regions that were more 380 \nfrequently assigned to canonical DMNA, indicating spatial reorganization within 381 \nDMN subnetworks during development. Critically, this contraction was correlated 382 \nwith episodic memory performance, suggesting that its areal specialization supports 383 \nongoing cognitive development. 384 \nOur study has several limitations. First, we divided the DMN into three subnetworks; 385 \nhowever, the exact number of DMN subnetworks remains unclear due to the 386 \ncomplexity of DMN functio n, and our study does not address this question, which 387 \nwarrants further investigation. Second, when examining DMN -related cognitive 388 \nperformance—for example, the close association between the DMNA and self -389 \nrelated thoughts —we lacked corresponding behavioral  measures to explore these 390 \nrelationships, which may require a broader behavioral phenotype. Finally, our 391 \ndiscussion of cognitive functions associated with DMN subnetworks primarily relies 392 \non previous literature rather than functional validation using speci fic tasks, which 393 \nshould be addressed in future studies. 394 \nIn summary, our study establishes a novel, multi -dimensional framework for 395 \ninvestigating DMN subsystem segregation during development by incorporating 396 \nmeasures of functional segregation, topographical  segregation, and spatial extent 397 \nalterations through individualized precision mapping. We demonstrate that 398 \nprogressive functional segregation of DMN subnetworks is accompanied by 399 \nboundary sharpening and spatial differentiation —manifested as reduced overlap  400 \nbetween subnetworks—as well as selective spatial refinement of DMNC that relates 401 \nto episodic memory performance. By transcending group -averaged approaches and 402 \nexamining individual DMN subnetworks, our findings emphasize the importance of 403 \ncharacterizing mu lti-dimensional functional and topographical features at the 404 \nindividual level to elucidate network segregation and specialization processes 405 \nessential for typical neurodevelopment and relevant to neuropsychiatric and 406 \ndevelopmental disorders. 407 \n 408 \nMaterials and Methods 409 \nData 410 \nAll participants are from the Human Connectome Project in Development (HCP-D). 411 \nHCP-D is a cross-sectional project including participants aged 5 –21 years. HCP-D 412 \nMRI data were acquired at four sites: Harvard University, University of California–413 \n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted February 25, 2026. ; https://doi.org/10.64898/2026.02.24.707618doi: bioRxiv preprint \n\nLos Angeles, University of Minnesota, and Washington University in St. Louis 44. 414 \nOur analysis is based on Release 2.0 (N = 625). All participants provided informed 415 \nassent or consent. Individuals exhibiting excessive head motion (frame -wise 416 \ndisplacement > 0.25 mm for more than one -third of the scanning duration) were 417 \nexcluded from further analysis. The final sample consisted of 547 subjects aged 5 to 418 \n21 years (mean age = 15 ± 3.8 years, 285 females).  The sample was predominantly 419 \nWhite (63.4%), followed by individuals identifying as more than one race (15.8%), 420 \nBlack or African American (10.8%), Asian (7.5%), and Other (2.5%). The median 421 \nannual household income of the sample was $119,000. 422 \nImage acquisition 423 \nMRI scans were acquired at 4 sites on 3T Siemens Prisma scanners with a 32 -424 \nchannel Prisma head coil. T1w images were acquir ed with a 3D multi -echo 425 \nMPRAGE (Magnetization Prepared Rapid Gradient Echo) sequence with an in -426 \nplane acceleration factor of 2. The scanning parameters included: repetition time of 427 \n2,500 ms, echo times of 1.8, 3.6, 5.4, and 7.2 ms, inversion time of 1,000 ms, flip 428 \nangle of 8 degrees, 208 slices, and a voxel resolution of 0.8 mm isotropic. The fMRI 429 \nscans are acquired with a 2D multiband gradient -recalled echo (GRE) echo -planar 430 \nimaging (EPI) sequence (MB8, TR/TE = 800/37 ms, flip angle = 52) and 2.0 mm 431 \nisotropic voxels covering the whole brain (72 oblique-axial slices). To mitigate bias 432 \nassociated with a specific phase-encoding direction, functional scans were acquired 433 \nin pairs, with one run using anterior-to-posterior (AP) and the other using posterior-434 \nto-anterior (PA) phase encoding. Participants aged 8 years and older underwent 26 435 \nminutes of resting -state scanning, consisting of four runs of 6.5 minutes each. For 436 \nyounger participants aged 5 to 7 years, a shorter protocol was used, comprising six 437 \nruns of 3.5 minutes each, totaling 21 minutes. 438 \nImage preprocessing 439 \nIn our study, we utilized preprocessed T1-weighted and fMRI images obtained from 440 \nthe NIMH Data Archive (NDA). The data were processed using the HCP -441 \nrecommended preprocessing pipeline, fMRIVolume, which includes realignment, 442 \ncorrection for spatial distortio ns, and registration of the functional images to the 443 \nstructural images with minimal smoothing (only due to interpolation, with no 444 \nexplicit smoothing). Artifact removal was performed using sICA+FIX, which 445 \nregresses out motion - and artifact -related ICA compo nents. Finally, fMRISurface 446 \nwas used to project gray matter voxels onto the cortical ribbon and register them to 447 \nthe fs_LR 32k surface space. 448 \nPrecision mapping  449 \nWe employed a template -matching approach similar to that used in previous 450 \nstudies45 . Compared with other precision mapping methods—such as Infomap, non-451 \nnegative matrix factorization, and Overlapping MultiNetwork Imaging mapping —452 \n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted February 25, 2026. ; https://doi.org/10.64898/2026.02.24.707618doi: bioRxiv preprint \n\ntemplate matching requires less scan time to obtain reliable personalized maps 45. In 453 \nthis study, we used Yeo’s 17-network parcellation as the template. For each vertex, 454 \nwe computed its functional connectivity profile with all the other vertices a nd 455 \nretained the top 5% of connections. We also examined the top 1% and 3% of 456 \nconnections and compared the consistency of parcellation across different thresholds 457 \nusing the Adjusted Rand Index, to ensure that the results were not driven by the 458 \nspecific choi ce of threshold. We then calculated the Dice similarity coefficient 459 \nbetween each vertex’s connectivity profile and each of the 17 Yeo networks, 460 \ngenerating a vertex -by-network similarity matrix. Each vertex was assigned to the 461 \nnetwork with which it showed t he highest similarity. Subsequently, we applied the 462 \nConnectome Workbench (https://www.humanconnectome.org/- 463 \nsoftware/connectome-workbench) command to identify and remove small isolated 464 \nclusters smaller than 40 mm2. These small clusters were then spatially reassigned to 465 \nthe nearest neighboring network to ensure spatial contiguity. Finally, we obtained 466 \nthe individual network parcellation for each subject. 467 \nHomogeneity of functional networks 468 \nTo evaluate the effectiveness of our precision mapping approach, we calculated the 469 \nwithin-network homogeneity, following procedures used in previous studies 40, 46. 470 \nSpecifically, we computed the Pearson correlation coefficients among all vertices 471 \nwithin each network and then averaged these values across all networks for both the 472 \nprecision-mapped parcellation and the group-level Yeo's 17-network parcellation. A 473 \npaired t -test was conducted to assess whether within -network homogeneity was 474 \nsignificantly higher for individual -specific parcellations compared to the group -475 \nlevel parcellation. 476 \nFunctional segregation of DMN subnetworks 477 \nTo investigate the functional organization of DMN subnetworks, we calculated the 478 \nfunctional segregation index, a metric that has been used in previous studies 34. We 479 \nfirst extracted the time series of all vertices and computed pairwise Pearson 480 \ncorrelations, resulting in a vertex-wise functional connectivity matrix. Next, we 481 \napplied Fisher’s z-transformation to the connectivity values. For each network, we 482 \ncalculated the segregation index as the difference between its average within -483 \nnetwork connectivity (FC W) and its average  between-network connectivity (FC B), 484 \nnormalized by the within -network connectivity, as shown in the formula. We 485 \ncomputed the within -DMN segregation (each DMN subnetwork vs. the other two 486 \nDMN subnetworks) and whole-brain segregation (each DMN subnetwork vs. all 16 487 \nother networks). 488 \nFunctional\tsegregation = \t FC1 − FC3\nFC1\n 489 \nDensity map of DMN subnetworks 490 \n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted February 25, 2026. ; https://doi.org/10.64898/2026.02.24.707618doi: bioRxiv preprint \n\nTo examine individual differences in the topography of DMN subnetworks, we 491 \ncomputed the density map of the DMN subnetworks. The density map was generated 492 \nby calculating the proportion of individuals for whom a given vertex belonged to a 493 \nspecific functional network. Higher values indicate that a greater proportion of 494 \nparticipants assigned that vertex to the given network. To better visualize age-related 495 \nchanges in the spatial distribution of functional networks, subjects were divided into 496 \nthree age groups based on age: children (≤12 years, n=147), adolescents (12 -18 497 \nyears, n=262), and adults (≥18 years, n=138).  For each group, we generated density 498 \nmaps to visualize the spatial location of the network. 499 \nSharpness of the boundary between DMN subnetworks  500 \nTo quantify the topographical distinctiveness of DMN subnetworks, we calculated 501 \nthe boundary sharpness between adjacent networks. First, we used the Connectome 502 \nWorkbench co mmand to identify spatially contiguous clusters within each 503 \nfunctional network. Then, we generated borders outlining each cluster, and extracted 504 \nthe vertex indices corresponding to these borders. For each cluster, we computed its 505 \nmean time series and ident ified nearby vertices within a 5 mm distance of each 506 \nborder vertex. These nearby vertices were classified as either inside vertices 507 \n(belonging to the same cluster) or outside vertices (belonging to different clusters). 508 \nWe then calculated the Pearson correl ation between the cluster’s mean time series 509 \nand the time series of each nearby border vertex. To assess boundary sharpness, we 510 \ncomputed Cohen’s d for the difference in correlation values between inside and 511 \noutside vertices. The outside vertices were furth er categorized into different 512 \nfunctional networks based on individual -level precision mapping. This allowed us 513 \nto compute boundary sharpness between each cluster and each adjacent functional 514 \nnetwork. If the number of outside vertices belonging to a given n etwork was fewer 515 \nthan 5, the sharpness was not calculated for that network pair. Finally, we averaged 516 \nthe sharpness values of all clusters belonging to the same network to obtain network-517 \nto-network sharpness estimates. We separately computed the sharpness between 518 \neach DMN subnetwork and the other two DMN subnetworks.  519 \nSpatial overlap between DMN subnetworks 520 \nTo quantify spatial overlap between DMN subnetworks, we built upon the 521 \nindividualized precision mapping results. While precision mapping assigns each 522 \nvertex exclusively to a single subnetwork through a winner -take-all procedure, 523 \nfunctional overlap between subnetworks may exist in practice. To capture this 524 \npotential overlap, we adopted the following approach. For each subnetwork, we first 525 \ncomputed the mean BOLD time series by averaging all vertices within its subject -526 \nspecific precision mapping -defined spatial extent. We then calculated Pearson 527 \ncorrelation coefficients between this mean time series and every cortical vertex 528 \nacross the brain. The resulting correlation maps underwent z-score normalization to 529 \n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted February 25, 2026. ; https://doi.org/10.64898/2026.02.24.707618doi: bioRxiv preprint \n\nstandardize the distribution. To allow vertices to be associated with multiple 530 \nsubnetworks based on their functional connectivity strength, each subnetwork was 531 \nredefined as the set of vertices exceeding a functional-connectivity threshold, which 532 \nwe systematically varied from 0.0 to 2.0. Primary results are reported at a threshold 533 \nof 1, with additional results provided in the Supplementary Materials. We selected 534 \nthis threshold because, after z -score normalization, a value of 0 corresponds to the 535 \nmean correlation and 2 represents two standard deviations above the mean —536 \nindicative of high connectivity. Higher thresholds retain very few vertices and 537 \nproduce minimal overlap across regions. For each subnetwork pair, overlapping 538 \nvertices satisfying the threshold criterion for both subnetworks were identified. We 539 \ncalculated the surface area of each cortical vertex to determine both the overlap area 540 \nand the combined area of the two subnetworks. The overlap ratio, compute d as the 541 \noverlap area divided by the combined area, provided a normalized measure of spatial 542 \ndifferentiation. 543 \nSurface area of DMN subnetworks 544 \nTo quantify the spatial extent of each DMN subnetwork, we first calculated the 545 \nsurface area (in mm²) for each cort ical vertex. The relative proportion of each 546 \nnetwork was then calculated by summing the surface areas of its constituent vertices 547 \nand dividing by the total cortical surface area. After examining age effects on surface 548 \narea proportion, we further examined d evelopmental changes in the topography of 549 \nDMN subnetworks. To characterize how the spatial distribution of individualized 550 \nnetwork parcellations changes with age, we compared individual maps with the Yeo 551 \ntemplate as a reference, which represents the canonic al spatial distribution of each 552 \nnetwork. By comparing individual network maps against this reference on a vertex-553 \nby-vertex basis, we identified vertices that showed different subnetwork 554 \nassignments across development. For example, by comparing an individua l's 555 \nDMNC map against the group-level DMNA and DMNB maps, we could determine 556 \nwhether regions that contract from DMNC during development are subsequently 557 \noccupied by other DMN subnetworks (i.e., DMNA or DMNB) in the reference map. 558 \n 559 \nAssessing Age-Related Effects on Brain Measures 560 \nWe employed a general linear model (GLM) to examine the relationship between 561 \nage and brain measures, including functional segregation, boundary sharpness, 562 \nspatial overlap and surface area of DMN subnetworks. In these models, age was 563 \ntreated as the independent variable, while the brain measures served as dependent 564 \nvariables. Sex and scanning sites were included as covariates. To account for 565 \nmultiple comparisons, we applied false discovery rate (FDR) correction to the p -566 \nvalues associated with the age variable. 567 \n 568 \n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted February 25, 2026. ; https://doi.org/10.64898/2026.02.24.707618doi: bioRxiv preprint \n\nDMN subnetwork development and cognitive development  569 \nWe examined whether developmental changes in the segregation and spatial extent 570 \nof DMN subnetworks are associated with cognitive development.  We used scores 571 \nfrom the Picture Seq uence Memory (PSM) test as a cognitive measure of episodic 572 \nmemory performance. We expected the PSM scores to increase with age, and 573 \ncorrelate with the memory -linked DMNC network. In the PSM task, a series of 574 \nillustrated objects and activities were presente d on an iPad in a specific order. The 575 \nsequence length varied from 6 to 18 pictures depending on the participant's age. 576 \nParticipants were asked to recall the correct order of the pictures and received points 577 \nfor each correctly recalled adjacent pair. This t est assesses the acquisition, storage, 578 \nand effortful recall of newly learned information. Higher scores indicate better 579 \nepisodic memory performance relative to age-matched peers. 580 \n 581 \n 582 \n 583 \n 584 \n 585 \n 586 \n 587 \n 588 \n 589 \n 590 \n 591 \n 592 \n 593 \n 594 \n 595 \n 596 \n 597 \n 598 \n 599 \n 600 \n 601 \n 602 \n 603 \n 604 \n 605 \n 606 \n 607 \n 608 \n 609 \n 610 \n 611 \n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted February 25, 2026. ; https://doi.org/10.64898/2026.02.24.707618doi: bioRxiv preprint \n\n 612 \n 613 \n 614 \n 615 \nReference: 616 \n1. Buckner RL, Andrews-Hanna JR, Schacter DL. The brain's default network: Anatomy, function, 617 \nand relevance to disease. Ann N Y Acad Sci 1124, 1-38 (2008). 618 \n 619 \n2. Menon V. 20 years of the default mode network: A review and synthesis. Neuron 111, 2469-2487 620 \n(2023). 621 \n 622 \n3. Anticevic A, Cole MW, Murray JD, Corlett PR, Wang XJ, Krystal JH. The role of default 623 \nnetwork deactivation in cognition and disease. Trends Cogn Sci 16, 584-592 (2012). 624 \n 625 \n4. Padmanabhan A, Lynch CJ, Schaer M, Menon V. The Default Mode Network in Autism. 626 \nBiological Psychiatry: Cognitive Neuroscience and Neuroimaging 2, 476-486 (2017). 627 \n 628 \n5. Konrad K, Eickhoff SB. Is the ADHD brain wired differently? A review on structural and 629 \nfunctional connectivity in attention deficit hyperactivity disorder. Hum Brain Mapp 31, 904-916 630 \n(2010). 631 \n 632 \n6. Gordon EM, et al. Default-mode network streams for coupling to language and control systems. 633 \nProc Natl Acad Sci 117, 17308-17319 (2020). 634 \n 635 \n7. Andrews-Hanna JR, Reidler JS, Sepulcre J, Poulin R, Buckner RL. Functional-Anatomic 636 \nFractionation of the Brain's Default Network. Neuron 65, 550-562 (2010). 637 \n 638 \n8. Braga RM, Buckner RL. Parallel Interdigitated Distributed Networks within the Individual 639 \nEstimated by Intrinsic Functional Connectivity. Neuron 95, 457-471.e455 (2017). 640 \n 641 \n9. Sherman LE, Rudie JD, Pfeifer JH, Masten CL, McNealy K, Dapretto M. Development of the 642 \nDefault Mode and Central Executive Networks across early adolescence: A longitudinal study. 643 \nDev Cogn Neurosci 10, 148-159 (2014). 644 \n 645 \n10. Chai XJ, Ofen N, Gabrieli JDE, Whitfield-Gabrieli S. Selective development of anticorrelated 646 \nnetworks in the intrinsic functional organization of the human brain. J Cogn Neurosci 26, 501-647 \n513 (2014). 648 \n 649 \n11. Gu S, et al. Emergence of system roles in normative neurodevelopment. Proc Natl Acad Sci 112,  650 \n(2015). 651 \n 652 \n12. Fan F, et al. Development of the default-mode network during childhood and adolescence: A 653 \nlongitudinal resting-state fMRI study. Neuroimage 226, 117581-117581 (2021). 654 \n 655 \n13. Wang C, Hu Y, Weng J, Chen F, Liu H. Modular segregation of task-dependent brain networks 656 \ncontributes to the development of executive function in children. Neuroimage, 116334-116334 657 \n(2019). 658 \n 659 \n14. Sun L, et al. Human lifespan changes in the brain's functional connectome. Nat Neurosci,  660 \n(2025). 661 \n 662 \n15. Baum GL, et al. Modular Segregation of Structural Brain Networks Supports the Development of 663 \nExecutive Function in Youth. Curr Biol 27, 1561-1572.e1568 (2017). 664 \n 665 \n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted February 25, 2026. ; https://doi.org/10.64898/2026.02.24.707618doi: bioRxiv preprint \n\n16. Kraus B, Zinbarg R, Braga RM, Nusslock R, Mittal VA, Gratton C. Insights from personalized 666 \nmodels of brain and behavior for identifying biomarkers in psychiatry. Neurosci Biobehav Rev 667 \n152, 105259-105259 (2023). 668 \n 669 \n17. Gratton C, et al. Defining Individual-Specific Functional Neuroanatomy for Precision Psychiatry. 670 \nBiol Psychiatry 88, 28-39 (2020). 671 \n 672 \n18. Demeter DV, et al. Precision functional mapping reveals less inter-individual variability in the 673 \nchild vs. adult human brain. In: bioRxiv). 20250724 edn (2025). 674 \n 675 \n19. Gordon EM, et al. Precision Functional Mapping of Individual Human Brains. Neuron 95, 791-676 \n807.e797 (2017). 677 \n 678 \n20. Seitzman BA, et al. Trait-like variants in human functional brain networks. Proc Natl Acad Sci 679 \n116, 22851-22861 (2019). 680 \n 681 \n21. Cui Z, et al. Individual Variation in Functional Topography of Association Networks in Youth. 682 \nNeuron 106, 340-353.e348 (2020). 683 \n 684 \n22. Keller AS, et al. Personalized functional brain network topography is associated with individual 685 \ndifferences in youth cognition. Nat Commun 14,  (2023). 686 \n 687 \n23. Lynch CJ, et al. Frontostriatal salience network expansion in individuals in depression. Nature 688 \n633, 624-633 (2024). 689 \n 690 \n24. Dickie EW, et al. Personalized Intrinsic Network Topography Mapping and Functional 691 \nConnectivity Deficits in Autism Spectrum Disorder. Biol Psychiatry 84, 278-286 (2018). 692 \n 693 \n25. Andrews-Hanna JR, Smallwood J, Spreng RN. The default network and self-generated thought: 694 \ncomponent processes, dynamic control, and clinical relevance. Ann N Y Acad Sci 1316, 29-52 695 \n(2014). 696 \n 697 \n26. Smallwood J, Bernhardt BC, Leech R, Bzdok D, Jefferies E, Margulies DS. The default mode 698 \nnetwork in cognition: a topographical perspective. Nat Rev Neurosci 22, 503-513 (2021). 699 \n 700 \n27. Buckner RL, DiNicola LM. The brain's default network: updated anatomy, physiology and 701 \nevolving insights. Nat Rev Neurosci 20, 593-608 (2019). 702 \n 703 \n28. Thomas Yeo BT, et al. The organization of the human cerebral cortex estimated by intrinsic 704 \nfunctional connectivity. J Neurophysiol 106, 1125-1165 (2011). 705 \n 706 \n29. Bathelt J, Geurts HM. Difference in default mode network subsystems in autism across childhood 707 \nand adolescence. Autism 25, 556-565 (2021). 708 \n 709 \n30. Doucet G, et al. Brain activity at rest: a multiscale hierarchical functional organization. J 710 \nNeurophysiol 105, 2753-2763 (2011). 711 \n 712 \n31. Greicius MD, Supekar K, Menon V, Dougherty RF. Resting-state functional connectivity reflects 713 \nstructural connectivity in the default mode network. Cereb Cortex 19, 72-78 (2009). 714 \n 715 \n32. Schmahmann JD, et al. Association fibre pathways of the brain: parallel observations from 716 \ndiffusion spectrum imaging and autoradiography. Brain 130, 630-653 (2007). 717 \n 718 \n33. Wang C, Hu Y, Weng J, Chen F, Liu H. Modular segregation of task-dependent brain networks 719 \ncontributes to the development of executive function in children. Neuroimage 206,  (2020). 720 \n 721 \n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted February 25, 2026. ; https://doi.org/10.64898/2026.02.24.707618doi: bioRxiv preprint \n\n34. Chan MY, Park DC, Savalia NK, Petersen SE, Wig GS. Decreased segregation of brain systems 722 \nacross the healthy adult lifespan. Proc Natl Acad Sci 111, E4997-5006 (2014). 723 \n 724 \n35. Fair DA, et al. Development of distinct control networks through segregation and integration. 725 \nProc Natl Acad Sci 104, 13507-13512 (2007). 726 \n 727 \n36. Lopez KC, Kandala S, Marek S, Barch DM. Development of Network Topology and Functional 728 \nConnectivity of the Prefrontal Cortex. Cereb Cortex 30, 2489-2505 (2020). 729 \n 730 \n37. Ng KK, Lo JC, Lim JKW, Chee MWL, Zhou J. Reduced functional segregation between the 731 \ndefault mode network and the executive control network in healthy older adults: A longitudinal 732 \nstudy. Neuroimage 133, 321-330 (2016). 733 \n 734 \n38. Keller AS, Sydnor VJ, Pines A, Fair DA, Bassett DS, Satterthwaite TD. Hierarchical functional 735 \nsystem development supports executive function. Trends Cogn Sci 27, 160-174 (2023). 736 \n 737 \n39. Supekar K, Uddin LQ, Prater K, Amin H, Greicius MD, Menon V. Development of functional 738 \nand structural connectivity within the default mode network in young children. Neuroimage 52, 739 \n290-301 (2010). 740 \n 741 \n40. Kong R, et al. Spatial Topography of Individual-Specific Cortical Networks Predicts Human 742 \nCognition, Personality, and Emotion. Cereb Cortex 29, 2533-2551 (2019). 743 \n 744 \n41. Vijayakumar N, et al. Brain development during adolescence: A mixed-longitudinal investigation 745 \nof cortical thickness, surface area, and volume. Hum Brain Mapp 37, 2027-2038 (2016). 746 \n 747 \n42. Bethlehem RAI, et al. Brain charts for the human lifespan. Nature 2022,  (2022). 748 \n 749 \n43. Asato MR, Terwilliger R, Woo J, Luna B. White matter development in adolescence: A DTI 750 \nstudy. Cereb Cortex 20, 2122-2131 (2010). 751 \n 752 \n44. Somerville LH, et al. The Lifespan Human Connectome Project in Development: A large-scale 753 \nstudy of brain connectivity development in 5–21 year olds. Neuroimage 183, 456-468 (2018). 754 \n 755 \n45. Hermosillo RJM, et al. A precision functional atlas of personalized network topography and 756 \nprobabilities. Nat Neurosci 27, 1000-1013 (2024). 757 \n 758 \n46. Gordon EM, Laumann TO, Adeyemo B, Huckins JF, Kelley WM, Petersen SE. Generation and 759 \nEvaluation of a Cortical Area Parcellation from Resting-State Correlations. Cereb Cortex 26, 760 \n288-303 (2016). 761 \n 762 \n 763 \n 764 \n 765 \n 766 \nAcknowledgments 767 \nFonds de recherche du Québec FRQNT 2021-NC-283571 (XJC) 768 \nCanada First Research Excellence Fund (XJC)  769 \nCanada Research Chairs program (XJC) 770 \nJeanne Timmins Fellowship (YH) 771 \nAuthor contributions: 772 \nConceptualization: YH, JK, XJC 773 \nMethodology: YH, JK 774 \nInvestigation: YH, JK, HJ, XJC 775 \n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted February 25, 2026. ; https://doi.org/10.64898/2026.02.24.707618doi: bioRxiv preprint \n\nVisualization: YH 776 \nSupervision: XJC 777 \nWriting—original draft: YH 778 \nWriting—review & editing: YH, JK, HJ, XJC 779 \nCompeting interests 780 \nThe authors declare that they have no competing interests. 781 \nMaterials & Correspondence 782 \n Ying He (ying.he3@mcgill.ca) 783 \n 784 \nData and materials availability: 785 \nRaw data are freely available through the Human Connectome Project, upon completion of a  786 \ndata-usage agreement: [https://nda.nih.gov/general-query.html?q=query=featured-datasets 787 \n:HCP%20Aging%20and%20Development]. All the analysis and visualization code are 788 \navailable on GitHub: https://github.com/Chai-Neuro-Lab/Segregation-of-DMN-subnetworks 789 \n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted February 25, 2026. ; https://doi.org/10.64898/2026.02.24.707618doi: bioRxiv preprint","source_license":"CC-BY-4.0","license_restricted":false}