The architecture of the human default mode network explored through cytoarchitecture, wiring and signal flow

Abstract

The default mode network (DMN) is implicated in many aspects of complex thought and behavior. Here, we leverage postmortem histology and in vivo neuroimaging to characterize the anatomy of the DMN to better understand its role in information processing and cortical communication. Our results show that the DMN is cytoarchitecturally heterogenous, containing cytoarchitectural types that are variably specialized for unimodal, heteromodal and memory-related processing. Studying diffusion-based structural connectivity in combination with cytoarchitecture, we found the DMN contains regions receptive to input from sensory cortex and a core that is relatively insulated from environmental input. Finally, analysis of signal flow with effective connectivity models showed that the DMN is unique amongst cortical networks in balancing its output across the levels of sensory hierarchies. Together, our study establishes an anatomical foundation from which accounts of the broad role the DMN plays in human brain function and cognition can be developed.

Fig. 1. Cytoarchitectural heterogeneity of the DMN.

a, Distribution of cortical types within the DMN. Upper left, the most common atlas of the DMN (used in primary analyses) is shown on the cortical surface. Lower left, cytoarchitectonic atlas of cortical types^,. Upper middle, histogram depicting frequency of cortical types within the DMN. The plus sign indicates significant over-representation and the minus sign, under-representation, relative to whole-cortex proportions. Lower middle, schematic highlighting prominent features that vary across cortical types, including the location/size of largest pyramidal neurons (triangles), thickness of layer IV, existence of sublayers in V–VI (gray dashed lines), regularity of layer I/II boundary (straightness of line). Kon, koniocortical; Eul, eulaminate; Dys, dysgranular; Ag, agranular. Right, circular plot representing the spread of the DMN from externally to internally driven cortical types. The percentage of each type within the DMN is depicted by the amount of the respective line (not the area in between lines) covered by the red shaded violin. Similar schematics may be found in extant literature^,,. b, Three-dimensional reconstructed sliced and stained human brain. Coronal slices of cell-body-stained sections (20-μm thick, n = 7,404) were reconstructed into a 3D human brain model, BigBrain. c, Example cortical patch shows depth-wise variations in cell-body-staining in BigBrain. d, Cytoarchitectural differentiation within the DMN. Principal eigenvector (E1) projected onto the inflated BigBrain surface shows the patterns of cytoarchitectural differentiation within the DMN. PHPC, parahippocampus; Prec., precuneus; IP, inferior parietal; MT, middle temporal; IF, inferior frontal; PFC, prefrontal cortex (superior frontal and anterior cingulate cortex). e, Cytoarchitectural profiles. Line plots represent cell-body-staining intensity by intracortical depth (from pial to white matter (wm) boundary) at different points along E1. Cortical points with lower E1 (blue) have peaked cellular density in mid–deep cortical layers, indicative of pronounced laminar differentiation, whereas cortical points with higher E1 (red) have more consistent cellular density across cortical layers, illustrating lower laminar differentiation. f, Cytoarchitectural landscapes of subregions. (i) Topography of E1 in each subregion shown as 3D surface plots, with E1 as the z axis. The x and y axes are defined by Isomax flattening of each subregion. (ii) Proportion of variance in E1 explained by spatial axes (x,y) for each subregion and for models of increasing complexity (second- to fourth-order polynomial regression). (iii) Waviness of E1 in each subregion. Upper and lower bounds of each box represent the adjusted R² for each hemisphere (n = 2), and the center point is the adjusted R² averaged across hemispheres.

Fig. 2. Organization of DMN connectivity.

a, Top, scatterplots show correlation of cytoarchitectural (Cyto-axis) axis (E1) with average (i) structurally modeled E_nav, (ii) functionally modeled input and (iii) functionally modeled output. Each point represents a node of the DMN; r and P_uncorrected values indicate the statistical outcomes of two-sided product–moment correlation tests (subregion assignment is illustrated in Extended Data Figs. 6a and 7 and Supplementary Table 1). Line plots are presented with 95% confidence interval shading. Bottom, bar plots shows the linear correlation coefficient (r) of E1 with average connectivity to each cortical type. The stability of the correlation coefficient was calculated by repeating the procedure ten times, each including 90% of datapoints. Error bars, s.d. of the r value across repetitions. Asterisks, significant negative r values indicating that DMN nodes with peaked profiles have higher E_nav with externally driven cortical types and stronger input from most cortical types. b, Multimodal model of DMN organization shows the dual character of the DMN, including areas with convergent input and insulated areas. All points in the scatterplot represent units of the DMN, are colored by position along the cytoarchitectural axis (y axis) and are organized along the x axis based on weighted average of type-specific E_nav. The top 75% of functionally defined inputs are shown. c, The DMN is unique amongst functional networks in balancing the strength of output across cortical types. (i) Distributions of strength of input from and output to each type. Colored ridge plots show probability distributions of connectivity between the DMN and each cortical type. For functional output, the DMN exhibits overlapping, normal distributions, whereas for functional input, type-wise differences are evident. (ii) Comparing networks in terms of balance of their output per type. Focusing on functional output, colored ridge plots show distributions for all networks, illustrating more balance between types in the DMN. Right, Imbalance of connectivity to distinct cortical types evaluated as the KL divergence from a null model with equal connectivity to each type. Colored dots show the empirical KL divergence for each network and the gray density plots show the null distribution of KL divergence values based on 10,000 spin permutations. Permutation testing indicated that the DMN is unique among functional networks in balancing output across cortical types (that is, imbalance lower than 95% of permutations). attn., attention network.

Fig. 3. Replication of crossmodal analyses within individual brains using 7-T MRI.

a, Comparison of microstructural axes. The principal eigenvector of microstructural variation in the DMN (E1) was extracted from myelin-sensitive qT1 MRI, in line with the procedure employed on the histological dataset (BigBrain), revealing similar patterns. (i) Primary histological axis, (ii) MRI-derived axis. b, Subregions differ in MRI-derived microstructural axis topography. The roughness of MRI-derived microstructural differentiation varied between subregions in line with histological evidence. (i) Parahippocampal (left) and prefrontal (right) landscapes. (ii) Landscape waviness. The parahippocampus exhibited a graded transition from high-to-low E1, reflected by high smoothness and low waviness, whereas the prefrontal cortex exhibited an undulating landscape with high waviness. For individual replications (n = 8), bars show the median across subjects and error bars depict the maximum and minimum. c, Communication efficiency and functional input decrease along the microstructural axis of the DMN. Using individual-specific measures, we consistently found that cortical points with higher E1 were associated with (i) lower average E_nav, (ii) especially lower E_nav with perceptually coupled cortical types and (iii) lower functional input. Line plots are presented with 95% confidence interval shading. Column plots with error bars, for individual replications, show mean ± s.d. across subjects (n = 8).

Extended Data Fig. 1. Meta-analytic functional decoding of the cortical type atlas.

Meta-analytic functional decoding of the cortical type atlas supports the association, described in literature reviews, between the gradient of cortical types and a shift in function from primary sensory to unimodal to heteromodal to memory-related processes. Using meta-analytic maps of thousands of functional MRI, we extracted terms that were consistently associated with increased activity within the specific cortical type (threshold z-statistic>2). The size of each word reflects the relative strength of its association with the cortical type. Only psychological constructs were retained in the term lists (thus excluding anatomical terms, for example “V1”, and experiment-related terms, for example “healthy controls”). Decoding was performed within spatially contiguous subregions for Kon, Eu-III and Eu-II, because no terms exceeded the threshold when the subregions were combined, due to the distinctive unimodal functions of each subregion.

Extended Data Fig. 2. Cortical types and functional networks.

A) Bar charts illustrate the proportion of cortical types within each functional network (for further details, see Supplementary Table 2. B) Matrix illustrating the outcome of pair-wise Kolmogorov-Smirnov tests, whereby darker colours reflect greater difference in the cortical type make-up of the functional networks. Rows and columns of the matrix are ordered according to the first principal component, thereby showing that the DMN occupies a middle ground between the functional networks skewed towards high granularity and the functional networks dominated by eulaminate-II.

Extended Data Fig. 3. Cytoarchitectural heterogeneity in the DMN replicated with alternative atlases.

A) The diverse cytoarchitectural composition of the DMN was also evident using alternative atlas definitions. Stacked boxplots illustrate the number of vertices assigned to each cortical type within the atlas with increasingly conservative thresholds for inclusion in the DMN represented along the x-axis. i) DMN based on consistency of deactivation during perceptually-driven tasks. Vertex-wise change in the BOLD response were calculated across 787 subjects in Human Connectome Project during fifteen perceptually-driven tasks. Surface projections show the consistency of deactivations (z ≤ -5) across the tasks. ii) Association (z-statistic) of each vertex to the DMN derived from an independent component analysis of 7,342 task contrasts. iii) Probability of the DMN at each vertex, calculated across 1029 individual-specific functional network delineations. Proportion of the DMN assigned to each cortical type, where the DMN is defined variably based on different consistency thresholds. B) Using an intersection of the three approaches in part A, we created a highly conservative delineation of the DMN. Specifically, vertices were included in the conservative atlas if (i) deactivations were observed in more than a quarter of perceptually-driven tasks, (ii) contribution to the task-ICA exceeded a z-statistic of 1 and (iii) assignment to the DMN was observed in more than a quarter of individuals. Subsequently, we replicated the procedure in the primary analysis to extract the principal cytoarchitectural axis. Notably, similar patterns of cytoarchitectural differentiation are evident in this conservative delineation of the DMN. The conservative cytoarchitectural axis also captures a variation from peaked to flat profiles.

Extended Data Fig. 4. Lower-order eigenvectors and comparing E1 to cortical types.

A) First five eigenvectors projected on the inflated BigBrain surface. For line plots on the right, staining intensity profiles were averaged within 100 bins of the respective eigenvector and coloured by eigenvector position. B) i. Raincloud- and box-plots show the distribution of E1 across cortical types (n = 109/3785/3982/2913/282/669 for Kon/Eu-III/Eu-II/Eu-I/Dys/Ag). Box plots represent minimum, quartile 1, median, quartile 3 and maximum. ii. Cortical type assignment (1:6) was rescaled to the range of E1 then subtracted from E1, producing a deviation map that highlights where the type-based and data-driven depictions of DMN cytoarchitecture differ. Negative values indicate lower E1 than expected by a linear relationship with cortical type, whereas positive values indicate higher than predicted E1. Thus, the E1 pattern is distinct to the gradient of laminar elaboration that is captured by the cortical types. Both are anchored by koniocortex on one side and agranular cortex on the other, but they differ in the ordering of Eu-and dysgranular areas.

Extended Data Fig. 5. Landscape simulations of smoothness and waviness.

As expected, smoothness decreases with noise and increases with slope, whereas waviness increases with noise and decreases with slope. A) We simulated 121 landscapes with varied slopes and bumpiness (noise). x and y values were identical in all landscapes, while the z-axis – reflecting E1 topography in the main study – was modulated in each simulation. The z-axis value was calculated as “(x * slope) + (rand * sigma)”, where slope is a value within [0:0.1:1], rand is a vector of normally distributed pseudorandom numbers the length of x and sigma is the product of x and a value within [0:0.1:1]. B-C) Left. Each square of the matrix represents a simulated landscape, with rows reflecting increasing slope and columns reflecting increasing noise. Centre-Right. Line plots show the outcome metrics of simulations per row and column, respectively. r-values represent the outcome of partial product-moment correlations (for example correlation of smoothness with noise, controlling for slope).

Extended Data Fig. 6. Intra- and inter-structural connectivity of the DMN with respect to the cytoarchitectural axis (E1) and cortical type.

Variations in navigation efficiency as a function of the cytoarchitectural axis within the DMN, DMN subregion and cortical type. Panel A) involves connections from each node of the DMN with all nodes outside the DMN (as in the primary analysis), Panel B) connections from each node of the DMN to all other nodes of the DMN and Panel C) connections from each node of the DMN to all other nodes. Far left. Cortical maps show average navigation efficiency. Centre left. Scatterplots show the correlation of the cytoarchitectural axis (E1) with average navigation efficiency, with points coloured by the seed parcel’s position within the DMN. Centre right. Bar plots show the linear correlation coefficient (r) of E1 with average navigation efficiency to each cortical type. Far right. Matrix shows the average navigation efficiency between each subregion of the DMN and each cortical type.

Extended Data Fig. 7. Overview of key cortical maps.

Cortical maps illustrate the key axes of variation in A-B) cytoarchitecture, C) structural connectivity and D-E) signal flow. Exact values for each parcel can be found in Supplementary Table 1.

Extended Data Fig. 8. Comparison of functional networks based on inter-network connectivity to different cortical types.

Coloured ridge plots on the left of each panel show probability distributions of connectivity between the functional networks and non-DMN cortical types. We evaluated the imbalance of connectivity across cortical types using the Kullback-Leibler (KL) divergence from a null model with equal connectivity to each type. On the right of each panel, coloured dots show the empirical KL divergence for each network and the grey density plots show the null distribution of KL divergence values based on 10,000 spin permutations. A) The DMN exhibits the most balanced navigation efficiency across cortical types, compared to other functional networks. The balance of the DMN did not reach a level of significance relative to spin permutations, but spin permutations account for the size and distribution of the network, thus we may infer it is the large size and wide distribution of the network that enable the DMN to strike a balance in communication across cortical types. B) Input to the DMN is not balanced with regards to cortical types. Stronger input comes from heteromodal, Eu-I cortex, which aligns with the over-representation of this cortical type within the DMN. C) The DMN is unique amongst functional networks in exhibiting balanced output to all cortical types, which is further supported by the balance of the DMN reaching significance in spin permutation testing.