Abstract
The brain requires efficient information transfer between neurons and large-scale brain regions. Brain connectivity follows predictable organizational principles. At the cellular level, larger supragranular pyramidal neurons have larger, more branched dendritic trees, more synapses, and perform more complex computations; at the macroscale, region-to-region connections display a diverse architecture with highly connected hub areas facilitating complex information integration and computation. Here, we explore the hypothesis that the branching structure of large-scale region-to-region connectivity follows similar organizational principles as the neuronal scale. We examine microscale connectivity of basal dendritic trees of supragranular pyramidal neurons (300+) across 10 cortical areas in five human donor brains (1 male, 4 female). Dendritic complexity was quantified as the number of branch points, tree length, spine count, spine density, and overall branching complexity. High-resolution diffusion-weighted MRI was used to construct white matter trees of corticocortical wiring. Examining complexity of the resulting white matter trees using the same measures as for dendritic trees shows heteromodal association areas to have larger, more complex white matter trees than primary areas (p < 0.0001) and macroscale complexity to run in parallel with microscale measures, in terms of number of inputs (r = 0.677, p = 0.032), branch points (r = 0.797, p = 0.006), tree length (r = 0.664, p = 0.036), and branching complexity (r = 0.724, p = 0.018). Our findings support the integrative theory that brain connectivity follows similar principles of connectivity at neuronal and macroscale levels and provide a framework to study connectivity changes in brain conditions at multiple levels of organization.
SIGNIFICANCE STATEMENT Within the human brain, cortical areas are involved in a wide range of processes, requiring different levels of information integration and local computation. At the cellular level, these regional differences reflect a predictable organizational principle with larger, more complexly branched supragranular pyramidal neurons in higher order regions. We hypothesized that the 3D branching structure of macroscale corticocortical connections follows the same organizational principles as the cellular scale. Comparing branching complexity of dendritic trees of supragranular pyramidal neurons and of MRI-based regional white matter trees of macroscale connectivity, we show that macroscale branching complexity is larger in higher order areas and that microscale and macroscale complexity go hand in hand. Our findings contribute to a multiscale integrative theory of brain connectivity.
Introduction
Efficient transfer of information between neurons and regions of the human brain is essential for perception, cognition, and behavior. Understanding how brain wiring on both cellular and regional scales is shaped to facilitate such efficient information processing is a central topic in the field of neuroscience.
Variation in neuronal connectivity is widely believed to be a key factor underlying regional differences in information processing (Schüz and Miller, 2002) and regional functional specialization (Amunts and Zilles, 2015). The role and capacity of a region for information processing is suggested to be an important mix of cortical type (Barbas and Rempel-Clower, 1997), neurotransmitter receptor densities (Zilles and Palomero-Gallagher, 2017), intracortical myelination (García-Cabezas et al., 2017), cortical and glial density (Collins, 2011), and dendritic complexity of pyramidal neurons (Elston, 2003b). Pyramidal neurons in particular are an important class of neurons in the context of neural integration and corticocortical connectivity (Schüz and Miller, 2002). Larger, more complex layer II/III pyramidal neurons are thought to display a higher capacity for information integration and are typically found in regions involved in multimodal cognitive processing (Jacobs et al., 2001; Elston, 2003a). Pyramidal neuron connectivity and dendritic branching result from a balance between space and energy cost (Wen and Chklovskii, 2008; Wen et al., 2009; Cuntz et al., 2010; Schröter et al., 2017) and maximization of the connectivity repertoire of the neuron (Wen et al., 2009). Importantly, the axonal projections of supragranular pyramidal neurons comprise a large portion of the white matter bundles interconnecting large-scale brain regions (Jacobs et al., 1997, 2001; Kaas et al., 2002; Barbas, 2015), forming the macroscale wiring network of the human brain (Bullmore and Sporns, 2009; Sporns, 2009; van den Heuvel and Sporns, 2011). Pyramidal neurons thus form an important class of neurons operating at both the microscale and macroscale level of brain connectivity.
MRI studies in turn have shown brain connectivity and macroscale functional networks to display distinct organizational principles with short communication relays (Bullmore and Sporns, 2009), structural and functional community structure (Meunier et al., 2010; Yeo et al., 2011), and the formation of central hubs for topological global integration of information (Tomasi and Volkow, 2011; van den Heuvel and Sporns, 2011). These network properties too are suggested to be shaped by a trade-off between cost and efficiency (Bassett et al., 2009; Bullmore and Sporns, 2012; van den Heuvel et al., 2012), observations contributing to a central question in the field of how different scales of brain connectivity relate to each other (Perin et al., 2011; Van Wedeen et al., 2012; Barbas, 2015; van den Heuvel and Yeo, 2017; Scholtens and van den Heuvel, 2018; García-Cabezas et al., 2019; Goulas et al., 2019).
We examined the relationship between these two levels of brain organization with the goal to examine whether and, if so, how the organizational principles that govern neuronal branching and complexity also shape the macroscale region-to-region connectivity. Across 10 cortical regions, we reconstructed a three-dimensional morphology of microscale dendritic branching and macroscale white matter wiring (Elston and Rosa, 1997; Jacobs et al., 1997; Elston et al., 1999; Jacobs et al., 2001). We explore the cross-scale association of different measures of branching complexity, together contributing to the broader picture of shared structural and functional organization across scales in the brain. We show that macroscale connectivity is organized according to principles similar to those of microscale neuronal connectivity, with larger, more branched connectivity in association cortex. In a direct comparison, our findings show that microscale and macroscale branching complexity strongly go hand in hand across brain areas, supporting the hypothesis of shared principles of connectivity across scales.
Materials and Methods
Human postmortem tissue
Cortical brain samples of five human donors were collected from the Netherlands Brain Bank (NBB; www.brainbank.nl; Tables 1, 2, demographics of the donors). Informed consent to perform autopsies and for the use of tissue and clinical data for research purposes was obtained by the NBB from donors and approved by the Ethical Committee of the Amsterdam University Medical Center. A total of 42 postmortem samples was obtained across the following 10 cortical regions: gyrus frontalis superior, gyrus frontalis medius, gyrus cingularis anterior, precuneus, gyrus insularis brevis, gyrus precentralis, gyrus postcentralis, gyrus occipitalis superior, occipital pole, lateral gyrus parietalis superior (Fig. 1A, Table 2), covering heteromodal association cortex, unimodal association cortex, primary areas, and paralimbic cortex.
Donor demographics
Included tissue samples
Tissue processing
The donor brains were dissected into standardized tissue blocks, first dividing the brain into the primary gyri (e.g., superior frontal gyrus), then into a predetermined number of equally sized blocks. From the selected blocks, 0.5 cm thin blocks were cut perpendicular to the gyral surface and spanning the width of the gyrus of interest (Fig. 1A) while avoiding any unnecessary pressure on the tissue to minimize staining artifacts later in the Golgi–Cox (Cox, 1891; Ramon-Moliner, 1970) staining process. Brain tissue was moved directly into light-proof jars containing Golgi–Cox solution for simultaneous fixation and impregnation (Glaser and Van der Loos, 1981). After 2.5–3 weeks, test sections of each tissue sample were processed to assess the progression and quality of the Golgi–Cox impregnation. In 39 of the 42 tissue samples, the full extent of dendritic branches was stained, and spines were visible on distal branches, but the remaining three samples showed substantial staining artifacts and could not be used for neuron reconstruction. The remaining 39 well-stained samples (Fig. 1B) were removed from the Golgi–Cox solution, rinsed and taken to the next processing steps, dehydrating and celloidin embedding of the tissue in preparation for sectioning. The resulting tissue blocks were cut in 180 µm sections perpendicular to the cortical surface using a sledge microtome (Leica/Reichert Jung Polycut S Microtome) and transferred to ethanol 70% in individual compartments of a Teflon disk for the free-floating development and dehydration process.
Cell selection and quantification of neuronal complexity
Localization
Neuron reconstruction was performed in a region-blinded manner, where for each donor the tissue blocks were assigned different randomly generated region numbers (R1–R10). Every sixth section of each tissue block was additionally Nissl stained to visualize all neuronal cell bodies in the section (rather than the random subset stained by the Golgi–Cox procedure), thus enabling assessment of the cortical layer structure. Using these Golgi–Nissl double-stained sections, the lower boundary of layer III was marked. Neurons in adjacent Golgi-stained sections were selected for morphologic reconstruction using the marked lower layer III boundary for reference, selecting neurons at a cortical depth just above the boundary line. Neurons were reconstructed in exclusively Golgi–Cox stained sections, in which the layered structure is much less apparent, further ensuring the observers were blind to the cortical region identity during neuron reconstruction.
Neuron morphology
Pyramidal cells from each sample were selected from the set of neurons with their soma located in supragranular layer III. Pyramidal neurons were included when (i) the soma was located in the center of the tissue section, (ii) at least two basal dendrites could be observed that each branched at least twice, (iii) there were no broken branches, and (iv) no branches were occluded by staining artifacts or other cells. A 3D reconstruction of the neurons was made using Neurolucida (version 11, Microbrightfield), using a 40× oil-immersion objective (Carl Zeiss Axioskop microscope). Pyramidal cells were manually traced using the neuron tracing tools included in Neurolucida, automatically registering the dimensions and branch order of each reconstructed neuron part. Tracing began with the unmyelinated initial segment of the axon as a landmark point for placement of the traced reconstruction, followed by an outline of the soma of the neuron, after which the main branch of the apical dendrite was reconstructed for orientation purposes. Finally, the basal dendrites including all branches were traced. Neuron reconstruction was performed by trained experts L.H.S. and R.P. and resulted in, on average, 75.2 neurons per donor (ranging from 49 to 99) and 37.6 per included brain region (ranging from 18 to 50), yielding a total of 376 reconstructed pyramidal neurons.
Spine count and spine density
The quality of the Golgi–Cox impregnation was assessed for each sample, evaluating the extent to which spines were fully impregnated, evidence of overimpregnation that could impede accurate quantification of spines, and/or staining artifacts occluding spines. Samples of the first included donor (donor 1) were used for optimization of the spine quantification protocol and not further included in the analysis. For the samples of the remaining four donors, where quality of staining was scored to be sufficient (91% of samples), the best visible, longest basal dendrite of each included neuron was selected for spine quantification, resulting in a total of 281 neurons with spine quantification included. Spines were marked when (i) they had a visible head and (ii) were visibly attached to the dendrite.
Analysis of neuronal morphology
Information on dendritic branching was extracted, and Sholl analysis was performed on the reconstructed neuron morphology using the tools natively included in Neurolucida. The following representative measures were selected based on previous studies (Scholtens et al., 2014): (i) total basal dendritic length, (ii) number of branch points, (iii) Sholl peak branching complexity, (iv) spine count, and (v) spine density. Spine density was computed by dividing the number of counted spines by the length (in µm) of the dendrite on which the spines were quantified and multiplying by 10 to obtain spine density per 10 µm. Neuron reconstructions are available from the NeuroMorpho.org repository of digitally reconstructed neurons.
Macroscale connectivity
MRI preprocessing and connectome reconstruction
High-resolution diffusion-weighted scans of 489 subjects of the Human Connectome Project (HCP; Van Essen et al., 2013; Q3 Subjects Release, age 22–35 years, male and female combined) were used to reconstruct maps of the macroscale human connectome, mapping white matter pathways between cortical areas (de Reus and van den Heuvel, 2014). Diffusion-weighted imaging (DWI) parameters included 1.25 mm isotropic voxel size, TR/TE 5520/89.5 ms, and 270 diffusion directions with diffusion weighting 1000, 2000, or 3000 s/mm2 (Van Essen et al., 2013). For each individual, preprocessing of the diffusion-weighted images included realignment, eddy current, and susceptibility distortion correction, followed by a voxel-wise reconstruction of diffusion profiles using generalized q sampling imaging and whole-brain streamline tractography using Connectivity Analysis TOolbox (CATO; de Lange and van den Heuvel, 2021). Fiber tracing started from all white matter voxels with eight seeds per voxel, with fiber tracing following the main diffusion direction of each voxel until reaching one of the stopping criteria (i.e., the fiber was about to exit the brain mask, made a turn of >45°, and/or reached a voxel with low fractional anisotropy <0.1). High-resolution anatomic T1 scans were used for cortical segmentation and FreeSurfer (Fischl et al., 2004) parcellation of the cortical sheet of each individual subject into 114 distinct regions (57 regions per hemisphere, a spatial resolution similar to that of the dissection protocol for the cortical tissue samples) using the Cammoun 114 subdivision of the Desikan–Killiany atlas (Desikan et al., 2006; Cammoun et al., 2012). The individual cortical parcellation was overlaid with the subject's whole-brain tractography to form a 114 × 114 connectivity matrix, describing all pairs of cortical regions and their reconstructed pathways (van den Heuvel et al., 2019). To match the microscale neuronal reconstructions sampled exclusively from the left hemisphere, connections of the left hemisphere were taken for further analysis, resulting in a 57 × 57 connectivity matrix.
White matter branch reconstruction
To examine parallels between the branching structure of pyramidal neurons and the branching structure of the white matter fibers, we introduced the same format used to describe neuron morphology to now describe the organization of reconstructed white matter trees. We developed the following approach.
In step 1, for each of the 57 left hemisphere regions, a subject's whole-brain tractography was overlaid with the individual regional parcellation to select all reconstructed streamlines for that region.
In step 2, the trajectory of each streamline of a region was followed in steps of 5 mm to determine where streamlines could be grouped into a white matter branch, drawing a sphere of radius 5 mm at each step. Neighboring streamlines were bundled in locations where the radii of two spheres intersect, and the angle between two potentially neighboring fibers is smaller than 90°, indicating that they project into the same direction. To account for streamlines that follow a more winding trajectory than others and thus traverse a smaller distance in a given number of steps, spheres were allowed to intersect with neighboring spheres of equal or lower step count. Branches consisting of more than three fibers were included for further processing, computing their branch diameter and coordinates.
In step 3, all reconstructed branches of a region were used to construct a white matter tree of that cortical area. The center of the cortical area was then placed at the zero coordinate point of the white matter tree. Each white matter branch was described at intervals of 5 mm, listing at each point the coordinates relative to the central point, together with the branch diameter and the previous linked point along the branch. The resulting white matter tree was saved in the widely used SWC format for neuron morphologies.
In step 4, per area, the branches of the connectivity tree were examined for their level of complexity, an analysis parallel to the branching complexity analyses performed to determine connectivity complexity at the neuronal scale. White matter trees were analyzed using the TREES toolbox for quantitative analysis (Cuntz et al., 2010), computing a white matter tree equivalent of total length, number of branch points, and peak branching complexity. Total length was taken as the sum of the length of all branches of the white matter tree (in mm). The number of branch points represents the number of locations where previously parallel streamlines diverge. Peak branching complexity was computed as the maximum number of intersections of the branches of the white matter tree with concentric circles around the center point of the tree (cortical parcel).
Cross-scale analyses
Postmortem cortical sample locations were manually mapped to FreeSurfer's fsaverage brain to determine corresponding macroscale Cammoun 114 FreeSurfer areas (Fig. 1A, Table 2; Fischl, 2012). For the precuneus tissue samples (Fig. 1, Table 2) the resection as shown on the autopsy photographs was located on or close to the border between two cortical parcels. These samples were assigned to both precuneus subparcels, and their macroscale connectivity data were averaged in the subsequent analyses, ensuring a single macroscale data point per tissue sample to match the other samples. All left-hemisphere cortical parcels were grouped according to their cortical region category, belonging to heteromodal association cortex, unimodal association cortex, primary areas, or paralimbic cortex (see Fig. 4A; Mesulam, 1998). Given the unequal number of cortical parcels contained in each region category (10 for paralimbic cortex, 10 primary cortex, 8 unimodal association cortex, 29 heteromodal association cortex), differences between cortical region categories were assessed by means of a nonparametric Kruskal–Wallis analysis of variance, with post hoc Wilcoxon rank sum tests for pairwise region category differences (Bonferroni corrected).
Verification analyses
Effect of donor on neuron complexity metrics
Generalized linear mixed-effects models were computed with the microscale neuron complexity measures as outcome variable, cortical region category as a fixed effect, and donor as a random factor. Additional nested models without donor as a random factor were included and log-likelihood ratio testing was performed to assess whether adding donor improved the model fit.
Spatial autocorrelation
We computed the spatial autocorrelation for each of the four macroscale brain maps (intrahemispheric degree, WM tree branch points, WM tree length, WM Sholl peak), and generated 1000 surrogate maps randomized while preserving spatial autocorrelation (Burt et al., 2020). Next, each surrogate map was correlated with the microscale neuronal complexity data, thus generating null distributions of surrogate correlation coefficients to compare with our observed micro-macro associations. Finally, to determine the p value of the likelihood that the empirically observed micro-macro association was driven by spatial autocorrelation, we calculated the proportion of surrogate maps with a correlation coefficient exceeding the observed association.
Alternative MRI atlases
To verify that the observed micro-macro associations were not dependent on the specific choice of atlas resolution implemented in this study, we repeated the main analyses on 50 HCP subjects using the Desikan–Killiany atlas (Desikan et al., 2006) and its Cammoun 219-region subdivision (Cammoun et al., 2012), providing respectively 34 and 111 cortical parcels for the left hemisphere. In addition, HCP multimodal parcellation (HCP MMP; Glasser et al., 2016) was implemented. As in the main analysis, the locations of the donor samples were mapped to FreeSurfer parcels on fsaverage. Because of their, on average, smaller parcel size, sample-parcel mappings in Cammoun 219 and HCP MMP more often included multiple cortical parcels per sample (Table 3). White matter trees were generated for all parcels, measures of white matter tree branching were computed, and cross-scale association with the microscale measures was assessed using Pearson's correlation.
Mapping of donor tissue sample locations
Alternative tractogram reconstruction methodology
White matter branching complexity was additionally computed using tractograms generated with MRtrix (Tournier et al., 2019). We processed 50 randomly selected datasets of HCP subjects using MRtrix, applying FreeSurfer segmentation and generating tractograms using deterministic streamlines tractography based on spherical deconvolution. We performed spherical-deconvolution-based streamline postfiltering using SIFT. Using the resulting 500,000 filtered streamlines, we generated connectome maps (same Cammoun 114 atlas as in our main analysis, including all left-hemisphere corticocortical connections), which were then used to reconstruct MRtrix-based white matter trees and to compute branching complexity metrics. MRtrix-based white matter tree complexity was compared with the original CATO-based metrics using Pearson's correlation analysis, and the main analyses were repeated to assess the consistency of the results across tractogram reconstruction methods.
Alternative white matter tree reconstruction settings
White matter trees were reconstructed with alternative settings to explore the effect on the branchedness of the resulting white matter tree. We explored the influence of the size of the radius within which streamlines are grouped into a branch of the white matter tree by generating new white matter trees with a radius of 1, 3, 7, and 9 mm and comparing the resulting white matter tree complexity with the default (5 mm). Similarly, we generated new trees using alternative step sizes of 1, 3, 7, and 9 mm (default is 5 mm) and with alternative grouping cutoff angle settings of 10, 30, 60, and 120° (default 90°).
Results
Significant variation in microscale morphology of supragranular pyramidal neurons
Basal dendritic tree length measured on average 1.89 × 1003 µm across reconstructed neurons (SD 0.82 × 1003). Dendritic tree length showed significant variation across cortical areas (χ2(3) = 16.17, p = 0.001, Kruskal–Wallis analysis of variance), with post hoc tests (Wilcoxon rank sum test, Bonferroni corrected) confirming significantly larger basal dendritic trees in heteromodal association cortex compared with primary areas (T = 2.662 × 1004, z = 2.887, p = 0.019), unimodal association cortex (T = 1.779 × 1004, z = 2.662, p = 0.044), and paralimbic cortex (T = 1.920 × 1004, z = 3.066, p = 0.019; see Fig. 4B). Pyramidal neurons with the largest dendritic trees were found in superior frontal gyrus (2.39 × 1003 µm average total dendritic length) and the smallest in the occipital pole (962.1 µm; Fig. 1E).
Regional variation in microscale basal dendritic tree organization. A, Schematic depiction of the locations of the 10 included cortical samples: gyrus precentralis (GPRC3), gyrus postcentralis (GPOC3), gyrus frontalis superior (GFS2), gyrus frontalis medius (GFM2), lateral gyrus parietalis superior (LPS2), gyrus insularis brevis (GIB3), gyrus occipitalis superior (GOS2), gyrus cingularis anterior (GCIA), precuneus (PSM), occipital pole (OCP). B, The number of reconstructed neurons in each tissue sample for each of the five donor brains. White squares indicate samples not obtained from that donor, darker shades of green indicate a larger number of neurons included in a sample (maximum 10 per sample). C, A representative example of Golgi–Cox stained layer III pyramidal neuron imaged at 10× (left), 20× (middle), and 40× (right). D, An example neuron reconstruction for each of the included brain regions of donor 4. E, The region ranking in terms of the five measures of wiring complexity, from left to right, top to bottom, basal dendritic length (in µm), dendritic branch points, peak dendritic Sholl intersections, spine count and spine density (in spines per 10 µm), with the white circle indicating the median and the dark gray bar the interquartile range.
Pyramidal neurons measured on average 19.10 (SD 6.82) branch points, showing significant variation across primary areas and unimodal and heteromodal association cortex (χ2(3) = 32.97, p = 3.26 × 10−07). Basal dendritic trees in heteromodal association cortex were significantly more branched than those in primary areas (T = 2.857 × 1004, z = 5.600, p = 9.57 × 10−08) and paralimbic cortex (T = 1.904 × 1004, z = 2.638, p = 0.046) but not more than unimodal association cortex (T = 1.762 × 1004, z = 2.119, p = 0.282, n.s.; Fig. 4B). Basal dendrites with the highest number of branch points were located in superior frontal gyrus (21.76), and the neurons with the least branched basal dendrites in the occipital pole (12.79; Fig. 1E).
Peak Sholl branching complexity (i.e., the maximum density of the basal dendritic field of a neuron; see Fig. 6) was measured to be on average 11.24 (SD 9.20). Branching complexity varied consistently across brain areas (χ2(3) = 26.17, p = 5.89 × 10−06), with post hoc testing showing heteromodal association cortex to have a higher peak branching complexity than primary areas (T = 2.561 × 1004, z = 4.673, p = 7.89 × 10−06) and paralimbic cortex (T = 1.783 × 1004, z = 3.027, p = 0.018) but not unimodal association cortex (T = 1.647 × 1004, z = 2.618, p = 0.058; Fig. 4B). Highest peak branching complexity was observed for superior frontal gyrus (17.56 on average), and lowest peak branching complexity was found in anterior cingulate cortex (4.53 on average; Fig. 1E).
Spine count estimated an average 183.1 spines per neuron (quantified on the best visible dendritic branch; SD 115.3), with no significant differences between cortical categories (χ2(3) = 8.583, p = 0.058, ns; see Fig. 4B). Highest average number of spines was observed in postcentral gyrus (227.6), and lowest in occipital pole (110.3; Fig. 1E).
Spine density was on average 3.61 spines/10 µm (SD 1.39 spines/10 µm) across all included neurons. Spine density was observed to be significantly different between region categories (χ2(3) = 8.363, p = 0.039), with paralimbic cortex having higher spine density than primary areas (T = 0.308 × 1004, z = 2.967, p = 0.023; Fig. 4B). In line with this observation, of the 10 included cortical sample locations, anterior cingulate gyrus was observed to have neurons with on average the highest spine density (4.35 spines/10 µm), whereas precentral gyrus had the lowest (2.96 spines/10 µm; Fig. 1E).
Significant variation in macroscale morphology of white matter structure
We continued by examining the white matter trees for each of the cortical areas (Fig. 2B,C), with our reconstructed white matter trees describing the organization of the macroscale connections of that region and providing a template to study regional differences in macroscale branching complexity (Figs. 2D, 3, for all n = 57 left-hemisphere regions). Across all individual white matter trees, average tree length was 545.78 mm (SD 279.45), average number of branch points was 14.62 (SD 6.72), and peak branching complexity was on average 12.09 intersections (SD 4.73). Pooling left-hemisphere brain regions of all subjects showed an overall average degree of 16.80 (SD 6.26).
Regional variation in macroscale white matter tree complexity. A, The location of the cortical parcels overlapping with the postmortem sample locations. B, Schematic illustration of the reconstruction of a toy white matter tree from a toy set of DWI tractography streamlines, following the trajectory of each streamline in steps of 5 mm (left) and at each step grouping streamlines when they are within 5 mm distance and projected in the same direction (middle), resulting in a 3D tree suitable for analysis in terms of branching characteristics (right). C, Left to right, visualization of all streamlines touching the superior parietal parcel (LPS2; A) of a randomly selected HCP subject (ID 107422); the 3D white matter tree resulting from grouping neighboring streamlines; visualization of the branching pattern of the white matter tree. D, Regional distributions for all matched-to-neuron samples MRI regions, sorted by region average, depicting white matter tree length (top left), number of branch points (top right), peak Sholl complexity (bottom left) and intrahemispheric degree (bottom right), with the white circle indicating the median and the dark gray bar the inter quartile range. Figure 3 shows ranked measures of all 57 left hemisphere regions. E, Scatter plots showing cross-correlations between degree and the measures of white matter tree complexity, showing a significant correlation with white matter tree length (r = 0.820, p = 1.02 × 10−14), branch points (r = 0.712, p = 7.40 × 10−10), and peak Sholl complexity (r = 0.413, p = 0.002).
Ranking of macroscale Cammoun-114 parcels by neural branching complexity. Showing A, intrahemispheric degree; B, white matter tree length; C, white matter branch points; D, white matter peak Sholl intersections. Within each violin, individual points represent HCP datasets, with the dot indicating the median value. Colored violins highlight regions included in the multiscale branching analysis.
We emphasize that white matter tree complexity measures were associated to network degree (i.e., the total number of connections of a region), white matter tree length (r = 0.8204, p = 1.02 × 10−14), branch points (r = 0.712, p = 7.40 × 10−10), and Sholl complexity (r = 0.413, p = 0.0015; Fig. 2E) but are by no means identical (Fig. 3). An example is the middle frontal gyrus falling near the middle of the ranking for degree (30th) but scoring much higher on white matter tree complexity (e.g., top nine in branch points and top four in Sholl complexity).
Next, nonparametric Kruskal–Wallis tests showed significant variation in region categories in all four white matter branching complexity measures (degree χ2(3) = 3.018 × 1003, branch points χ2(3) = 4.412 × 1003, white matter tree length χ2(3) = 4.282 × 1003, peak Sholl complexity χ2(3) = 2.375 × 1003, all p values < 0.0001; Fig. 4C). Post hoc comparison (Wilcoxon rank sum test) showed significant differences between all pairs of region categories in intrahemispheric degree, white matter tree length, number of white matter branch points (all T values > 0.152 × 1008, z > 8.892, and p < 0.0001). Post hoc comparison of white matter peak complexity showed paralimbic cortex and primary areas to have lower Sholl peak complexity than unimodal and heteromodal association cortex (all T values > 0.187 × 1008, z > 4.836, p < 0.0001).
Cross-scale comparison of branching complexity. A, Region category assignment into paralimbic cortex (GIB3, anterior insula; GCIA, anterior cingulate cortex), primary areas (GPRC3, precentral gyrus; GPOC3, postcentral gyrus; OCP, occipital pole), unimodal association cortex (GOS2, superior occipital gyrus), and heteromodal association cortex (GFS2, superior frontal; GFM2, middle frontal; LPS2, lateral superior parietal; PSM, medial superior parietal cortex (precuneus)). B, Comparison of microscale basal dendritic tree morphology across cortical region categories, with Kruskal–Wallis nonparametric one-way ANOVA showing significant differences between cortical categories in dendritic tree length (χ2(3) = 16.17, p = 0.001), dendritic branch points (χ2(3) = 32.97, p = 3.26 × 10−07), peak Sholl branching complexity (χ2(3) = 26.17, p = 5.89 × 10−06), and spine density (χ2(3) = 8.363, p = 0.039) but not spine count. Post hoc paired comparisons showed that heteromodal association cortex has neurons with larger dendritic trees than paralimbic cortex, primary areas, and unimodal association cortex, and more branch points and higher peak Sholl complexity than paralimbic cortex and primary areas. Spine density was higher in paralimbic cortex than in primary areas. C, Comparison of macroscale white matter tree morphology across cortical region categories, with Kruskal–Wallis nonparametric one-way ANOVA showing significant differences between groups in all four measures (degree χ2(3) = 3.018 × 103, branch points χ2(3) = 4.412 × 103, white matter tree length χ2(3) = 4.282 × 103, peak Sholl complexity χ2(3) = 2.375 × 103, all p values < 0.0001). Post hoc comparison showed significant differences between all pairs of region categories in white matter tree length (top left) and number of white matter branch points (top right). White matter Sholl peak complexity (bottom left) was lower in paralimbic cortex and primary areas than unimodal and heteromodal association cortex. Paralimbic and unimodal association cortex were observed to have similar degree (bottom right), with significant differences between all other pairs of cortical region categories. White circle in each violin indicates the median value, the gray bar the interquartile range.
We ranked the 10 matched-to-histology macroscale cortical regions (Fig. 2A, Table 2) in terms of the branching complexity of their associated white matter trees (Fig. 2D). Rostral middle frontal gyrus displayed on average the most branched (19.79, SD 5.04) white matter trees with the highest Sholl complexity (16.24, SD 4.26), whereas superior frontal gyrus had the longest white matter trees (862.65 mm, SD 276.60 mm) and superior parietal cortex the highest macroscale degree (25.25, SD 5.76). Smallest white matter tree structures were observed in anterior cingulate (length, 349.31 mm, SD 128.14 mm; branch points, 8.48, SD 2.90; Sholl complexity, 8.81, SD 3.47; degree, 11.66, SD 1.95), followed by precentral gyrus [second lowest Sholl complexity and degree, respectively, 11.16 (SD 2.80) and 12.80 (SD 2.77)] and occipital pole [second lowest total tree length and branch points, 406.41 mm (SD 180.55 mm) and 12.33 (SD 3.68; Fig. 2D)].
Cross-scale analysis shows micro-macro correlation in connectivity
Significant cross-scale associations were found in the regional patterns of branching complexity, with the number of branch points showing strongest correspondence across scales (r = 0.797, p = 0.006 (uncorrected; Fig. 5F). Number of connections as quantified by microscale spine count and macroscale degree (r = 0.677, p = 0.032; Fig. 5H), total tree length (r = 0.664, p = 0.036; Fig. 6) and peak Sholl branching complexity (r = 0.724, p = 0.018; Fig. 6) also showed significant multiscale association. Number of branch points showed a particularly strong association across scales, also with other nonequivalent measures of branching complexity (Table 4). Branch points of pyramidal neuron basal dendrites was associated with macroscale regional degree (r = 0.677, p = 0.032), white matter branch points (as reported above, Fig. 5E,F) and white matter tree length (r = 0.798, p = 0.006). In turn, white matter branch points showed strong association with microscale spine count (r = 0.728, p = 0.017), dendritic branch points (as reported above), dendritic length (r = 0.762, p = 0.010), and dendritic peak Sholl branching complexity (r = 0.766, p = 0.009) but not spine density.
Cross-scale associations between neuronal and white matter branching structure
Similar regional dendritic and white matter branching. A, Schematic illustration of the locations of the matching donor sample locations (small rectangles) and FreeSurfer parcels (colored patches) for the 10 included brain regions. B, Two measures of neural branching, the total length and the number of branch points. C, D, Dendrograms of exemplary neurons for all regions depicted in A (same neurons as in Figure 1C), exemplary dendrograms for white matter trees of those same 10 brain regions (of HCP participant 100307, D). E, The regional variation in number of branch points in the reconstructed neurons (left) and in the macroscale white matter trees (right), with white circles indicating the median value and the gray bars the interquartile range. F, The regional number of branch points shows considerable multiscale overlap (r = 0.797, p = 0.006). G, H, The regional distribution of dendritic spine count (left) and macroscale degree (right), with the cross-scale association (r = 0.677, p = 0.032) depicted in H.
Similar regional Sholl complexity and tree length across scales. A, Schematic illustration of the locations of the matching donor sample locations (small rectangles) and FreeSurfer parcels (colored patches) for the 10 included brain regions. B, Schematic depiction of Sholl analysis on neuron basal dendrite (left) and white matter tree branching (right). C, D, The regional variation in peak Sholl intersections in the reconstructed neuron basal dendrites (left) and in the macroscale white matter trees (right), with considerable multiscale overlap (r = 0.724, p = 0.018; D). E, The regional distribution of total dendritic length (left) and macroscale white matter tree length (right). F, Scatter plot shows the cross-scale association, with dendritic length on the x-axis and white matter tree length on the y-axis (r = 0.6639, p = 0.0363). White circle in each violin indicates the median value, the gray bar the interquartile range.
Verification analyses
Effect of donor on neuron complexity metrics
Log-likelihood ratio testing indicated that including donor as a random factor in generalized linear mixed-effects models for the microscale neuron complexity metrics improved model fit for dendritic number of branch points (log-likelihood ratio statistic (LRstat) = 8.645, p = 0.003), spine count (LRstat = 6.191 p = 0.013), and spine density (LRstat = 8.242, p = 0.004) but not for dendritic tree length (LRstat = 0.4466, p = 0.504) and peak Sholl complexity (LRstat = 0.563, p = 0.453). Incorporating age or postmortem delay in the model instead of donor produced the same model outcomes. After incorporating donor as a random factor, cortical region category still significantly contributed to the model for the number of dendritic branch points (T = 5.543, p = 5.619 × 10−08), dendritic tree length (T = 3.615, p = 3.415 × 10−04), and peak Sholl complexity (T = 4.792, p = 2.420 × 10−06) but not spine count (T = 0.077, p = 0.939) and spine density (T = −0.698, p = 0.486). Because the included cortical regions differ per donor, the above result could be driven by either the difference in regions sampled per donor or by a donor characteristic, such as age, neuropathological status, or postmortem delay. Comparing those samples available for all donors showed that even though older donors had a lower overall dendritic complexity (Fig. 7), excluding the oldest two donors from the across-scale association analysis indicated that our main findings were not driven by this age effect (Fig. 8).
Effect of donor age and postmortem delay. A, B, The individual neuron data aggregated over the four regions for which data of all donors was available, that is, precentral gyrus, postcentral gyrus, superior frontal gyrus, and middle frontal gyrus, sorted by donor age (A) and by postmortem delay (B). Kruskal–Wallis nonparametric one-way ANOVA shows significant differences between the donors in dendritic tree length (χ2(4) = 11.24, p = 5.2 × 10−5, with the oldest donor having shorter basal dendrites), dendritic branch points (χ2(4) = 24.92, p = 0.024), peak Sholl complexity (χ2(4) = 14.77, p = 0.0052; older donors tend to have neurons with lower complexity), and spine density (χ2(3) = 12.53, p = 0.0058) but not spine count (χ2(3) = 6.18, p = 0.081). There appear to be no systematic differences in neuron branching complexity related to postmortem delay. White circle in each violin indicates the median value, the gray bar the interquartile range.
Analysis without Alzheimer's disease diagnosed donors. The cross-scale association between neural and white matter branching complexity, performed with leaving out the donors diagnosed with Alzheimer's disease (donor 2 and 3, also the oldest included donors at 88 and 98 years old). Leaving out these donors yielded results highly similar to the main analysis, although without significant association between dendritic and white matter tree length.
Spatial autocorrelation
BrainSMASH-generated surrogate maps (Burt et al., 2020) with similar spatial autocorrelation to the original white matter complexity measures did not result in comparable across-scale associations (lower average Pearson correlation coefficients and p < 0.05 for all significant micro-macro associations of the main analysis), indicating that the observation of across-scale similar branching organization is not primarily driven by spatial autocorrelation (Table 5).
Results of verification analysis with null distributions of spatial autocorrelation-conserved surrogate brain maps of the macroscale measures
Alternative MRI atlases
Verification analyses showed that the Desikan–Killiany atlas replicates all micro-macro associations observed in the main analysis, showing similar or stronger across-scale associations than Cammoun 114 (Fig. 9, Table 6). The Cammoun 219-region subdivision of the Desikan–Killiany atlas (when excluding the pericalcarine parcel, which is a relative outlier in terms of size compared with all other parcels) shows across the board weaker micro-macro associations than the main analysis, although still with the strongest associations for WM branch points (with spine count, dendritic branch points, dendritic length, and dendritic Sholl peak) as well as dendritic branch points (with WM tree branch points and WM tree length; Fig. 9, Table 6). Finally, when disregarding parcels V1 and V3, which are proportionally much larger than the majority of other parcels, HCP MMP also reveals cross-scale associations of WM tree branch points to appear most consistent with the main findings, although again weaker than Cammoun 114 and Desikan–Killiany (Fig. 9, Table 6).
Cross-scale associations with alternative cortical parcellation
Cross-scale associations between neuronal and MRtrix-based white matter branching structure
Alternative MRI atlases. A, Micro-macro associations between equivalent measures of branching complexity, with Desikan–Killiany (top row), the Cammoun 219 region subdivision of the Desikan–Killiany atlas (middle row), and HCP MMP (bottom row). B, Associations of white matter tree branch points with dendritic branching complexity for the same atlases (Table 6).
Alternative tractogram reconstruction methodology
Comparison of MRtrix-based and the original CATO-based tractograms used in the main analysis showed highly similar branching complexity in terms of the number of branch points, white matter tree length, and white matter Sholl peak (r = 0.8074, p = 3.294 × 10−14; r = 0.820, p = 6.274 × 10−15; r = 0.831, p = 1.295 × 10−15; Fig. 10A). Furthermore, MRtrix-based white matter tree complexity significantly differed across cortical region categories for all three measures (WM tree branch points, χ2(3) = 416.5, p = 5.9044 × 10−90; WM tree length, χ2(3) = 609.6, p = 8.162 × 10−132; WM tree peak Sholl complexity, χ2(3) = 173.4, p = 2.319 × 10−37; Fig. 10B). As in the original analysis, post hoc testing (Wilcoxon rank sum test) showed that white matter tree length was significantly different between all pairs of region categories (all T values > 0.126 × 1006, ∣z∣ > 5.687, p < 0.001), and peak Sholl complexity of respectively unimodal and heteromodal association cortex was higher than that of primary and paralimbic cortex (for all significantly different pairs of region categories, T > 0.180 × 1006, z > 6.00, p < 0.001). Differences between cortical region categories in terms of the number of white matter tree branch points were slightly less pronounced than in the original analysis (where all pairs of region categories differed significantly), with all but paralimbic and primary cortex significantly different (all T values > 0.162 × 1006, z > 4.930, p < 0.001). Comparing branching complexity across scales of organization revealed a similar trend toward a shared organization between dendritic and white matter tree complexity, although not as strong as in the main analysis (number of branch points, r = 0.650, p = 0.042; tree length, r = 0.436, p = 0.208 (n.s.), peak Sholl complexity, r = 0.659, p = 0.038; Fig. 10C, Table 7). As in the main analysis, microscale dendritic number of branch points showed the most consistent association with macroscale branching complexity (WM branch points, as above; WM tree length, r = 0.712, p = 0.021; WM peak Sholl complexity, r = 0.893, p < 0.001; Fig. 10D). An additional association, not observed in the main analysis, was that of MRtrix-based WM peak Sholl complexity with microscale dendritic length (r = 0.745, p = 0.013; Fig. 10D, compare Tables 4 and 5).
Verification analysis with alternative tractography data. A, Scatter plots comparing the white matter tree complexity metrics from the main analysis with metrics derived from tractograms generated using MRtrix as a verification of the applicability of the white matter tree pipeline on alternative tractography data. All three white matter tree complexity metrics were highly correlated between the two reconstruction methods (number of branch points, r = 0.807, p = 3.294 × 10−14; white matter tree length, r = 0.820, p = 6.274 × 10−15; and white matter Sholl peak, r = 0.831, p = 1.295 × 10−15). B, The complexity metrics for the four cortical region categories, with strong differences between cortical categories for WM tree branch points (χ2(3) = 416.5, p = 5.904 × 10−90) WM tree length (χ2(3) = 609.6, p = 8.162 × 10−132), and WM tree peak Sholl complexity (χ2(3) = 173.4, p = 2.32 × 10−37), consistent to the main analysis. C, Scatter plots depicting the cross-scale association between equivalent measures of white matter and dendritic branching complexity (number of branch points, r = 0.650, p = 0.042; tree length, r = 0.436, p = 0.207 (ns), peak Sholl complexity: r = 0.659, p = 0.038). ns, Not significant. D, As in the main analysis, microscale dendritic branch points are associated with all three WM complexity measures, number of branch points (C, left), WM tree length (r = 0.712, p = 0.021; D, top), and WM peak Sholl complexity (r = 0.893, p < 0.001; D, middle). An additional association not seen in the main analysis was observed between WM tree Sholl peak and dendritic length (r = 0.745, p = 0.038; D, bottom).
Alternative white matter tree reconstruction settings
Exploring the branchedness of the dendrograms resulting from alternative white matter tree reconstruction settings showed that decreasing the fiber grouping radius lowered the threshold for branching and resulted in white matter trees with more branch points (Fig. 11A). Varying step size had a relatively modest effect, with somewhat more branched trees resulting from smaller step sizes (Fig. 11B). Alternative cutoff angles (determining the maximum angle between two streamlines to still be considered as traveling in the same direction) had an effect similar to that of grouping radius, with smaller cutoff angles resulting in fewer streamlines grouped and thus more branch points in the tree (Fig. 11C).
Dendrograms produced using alternative settings in white matter tree reconstruction. Dendrograms illustrate the effect of changes to the white matter tree reconstruction settings on the branchedness of the resulting white matter tree. Shown is the same superior parietal parcel also included in Figure 2C, with the dendrogram produced with the default settings emphasized in bold. A, Left to right, The dendrograms made using streamline grouping radius 1, 3, 5 (default), 7, and 9 mm. B, Results of a step size of 1, 3, 5 (default), 7, and 9 mm. C, The output resulting from a cutoff angle of 10, 30, 60, 90 (default), and 120 degrees.
Discussion
Our findings point to similar characteristics of brain connectivity across scales. White matter macroscale wiring patterns display similar connectivity characteristics as the neuronal microscale. Well-known aspects of neuronal organization that have an impact on local information processing such as the length of dendritic wiring, branch points, and branching complexity were found to scale in a similar manner across macroscale areas.
Our observed cross-scale relationship in connectivity complexity is supported by empirical and computational modeling studies at both the cellular and whole-brain level. Both cellular and neuroimaging studies have separately shown brain connectivity to demonstrate a trade-off between a minimization of wiring cost and maximizing processing efficiency (Bullmore and Sporns, 2012; van den Heuvel et al., 2012). Wiring economy in microscale pyramidal neuron morphology balances efficient signal integration across the neuron with local processing segregating signals in dendritic segments and branches (Branco and Häusser, 2010). Shorter dendritic trees are more suited for rapid coincidence detection, whereas larger more branched dendritic trees increase the capacity for (slower) temporal integration (Amatrudo et al., 2012; Papoutsi et al., 2017). At the system level, heteromodal association cortex shows both higher neuronal complexity and macroscale connectivity while generally characterized by a highly diverse structural and functional connectivity profile, ideal for a central role in integrative information processing in the brain (Felleman and Van Essen, 1991; van den Heuvel and Sporns, 2013; Zhang et al., 2020). Cortical regions involved in primary sensory processing have a more focused connectivity profile, ensuring segregated initial processing of sensory inputs (Felleman and Van Essen, 1991), in line with our current observations of relatively less complex microstructural and macrostructural connectivity profiles of primary areas, indicating that indeed micro and macroscale branching complexity go hand in hand.
Studies modeling macroscale region-to-region connectivity of the brain have suggested that the connections of a cortical region are shaped by multiple factors. One guiding factor is the spatial proximity of brain regions (Beul et al., 2017; Goulas et al., 2019), with the likelihood of being connected further increasing for regions of similar cortical type (Barbas and Rempel-Clower, 1997). A second proposed mechanism involves physical processes such as tension-based morphogenesis, in which the compact structure of white matter bundles results from the physical tension along developing axons and dendrites, reaching across scales of organization to bring connected regions closer together (Essen, 1997). Alternatively, the three-dimensional structure of brain wiring has been described as a spatially continuous grid, established along three primary chemotactic gradients of early embryogenesis and deformed by cerebral expansion later in development (Van Wedeen et al., 2012).
On the microscale, primate studies have shown that the size of the (basal) dendritic tree of a pyramidal neuron and the reach of its local, intra-areal axonal projections are strongly related (e.g., Lund et al. (1993)) and increase along the cortical hierarchy (Elston, 2003a). The extent of the local and long-range connectivity of a neuron go hand in hand, and together show an increase in the reach of neuronal connectivity from primary areas to association cortex (Amir et al., 1993). These characteristics, together with our observation of consistent branching complexity across the micro and macroscale of the connectome, support the notion of a continuous, consistently organized wiring structure of the human brain across scales.
A number of practical limitations should be considered. The majority of brain donors are of relatively advanced age (mean 78.8, range 55–98 years), and it should be noted that the three oldest tissue donors displayed at least some degree of neuropathological changes (Table 1). The oldest two tissue donors were later diagnosed with respectively early-stage and advanced Alzheimer's disease. Although age-related changes in pyramidal morphology have been reported, supragranular pyramidal neurons are relatively unaffected by aging compared with infragranular neurons (de Brabander et al., 1998), and the global pattern of larger dendrites and more spines in higher order areas, which is the focus of the current study, remains rather robust (Jacobs et al., 1997, 2001). In our data, donor was predictive of variation between neurons in dendritic branch points, spine density, and spine count. We also note that dendritic spines are more sensitive to differences in fixation and staining quality than the overall dendritic branching structure, likely resulting in more interdonor variability in spine count and spine density. Overall, of the measures of microscale branching complexity, dendritic branch points, dendritic length, and peak Sholl complexity are most robust across donors.
Basal dendritic branches of supragranular pyramidal neurons were reconstructed in sections of Golgi–Cox stained tissue samples (Glaser and Van der Loos, 1981). This procedure stains a random subset of neurons, leaving the majority unstained and translucent, thus enabling the visual tracing and 3D reconstruction of neuron morphology. The mechanism by which some neurons are stained and others are not remains elusive (Ramon-Moliner, 1970; Buell, 1982; Swaab and Uylings, 1988). The Golgi–Cox stained tissue allowed for reconstruction of primarily lateral basal dendrites projecting more or less horizontally from the soma and thus results in a partial reconstruction of the dendritic tree.
Within a neuron, the different aspects of its morphology are interrelated and together contribute to overall neuronal complexity (Lund et al., 1993). Similarly, we observe that the different measures of macroscale branching complexity are not independent, and each highlight a different aspect of white matter complexity. Here, we explore the across-scale association of different measures of branching complexity in an attempt to capture the broader picture of shared organization across scales in the brain and to assess which measure(s) of branching complexity are most consistent (and thus informative) across scales, regardless of reconstruction method or specific settings in the white matter tree pipeline.
Although our observation of shared organization of neural wiring across scales was indeed similar across different tractography reconstruction methods and cortical atlas resolutions, we note that diffusion-weighted imaging provides an indirect reconstruction of anatomic connectivity compared with more direct methods such as histologic tract tracing (Donahue et al., 2016). Fiber orientations modeled using diffusion-weighted imaging have, however, been shown to correspond well with histologic myelinated fiber orientation in the human (Seehaus et al., 2015) and primate brain (Kaufman et al., 2005). Metrics of DWI-derived metrics of connectivity strength tend to show overlap with metrics derived from gold-standard tracer-based connectivity measurements in, for example, the macaque cortex (van den Heuvel et al., 2015; Delettre et al., 2019). Regarding the Desikan–Killiany atlas and its variants (Cammoun 114, Cammoun 219) employed in our analyses, these reflect primarily a sulcal-based subdivision of the cortical mantle and thus do not inherently inform on, for example, functional or cytoarchitectural characteristics of the included regions.
Our analyses were performed using a combination of in vivo high-resolution diffusion-weighted MRI datasets of relatively young (aged 20–35 years) participants of the Human Connectome Project (Van Essen et al., 2013) and ex vivo cortical samples of older donor brains (age 55–98 years). Because of this cross-cohort study design, our observations should be taken as a general indication of consistency in organization between neuronal microscale and region-to-region macroscale wiring. Future combined neuronal and white matter reconstruction within the same individuals would be of high interest to further substantiate the observations made.
Our findings suggest the existence of multiscale principles of connectivity organization in the human brain. A first question of high interest is how such observed shared principles might come to be, both from a perspective of evolution and of brain development. Future comparisons incorporating the interplay of connectivity organization with other neuronal and cell types are also of great interest. Other cell types (e.g., inhibitory neurons, astrocytes) have a crucial role in regulating activity within a brain region and play an important role in the branching and spiking behavior of pyramidal neurons in health (Bonifazi et al., 2009; Kanner et al., 2018; Hafizi et al., 2021; Kajiwara et al., 2021) as well as in disease (Marín, 2012; Dossi et al., 2018).
Cross-scale investigations of connectivity may help to better understand disease processes that affect brain connectivity. Many neurologic and psychiatric conditions are reported to show changes in brain connectivity at both the cellular and macroscale level, for example, in Alzheimer's disease (Braak and Braak, 1996; Uylings and de Brabander, 2002; He et al., 2008; Stam et al., 2009) and schizophrenia (Garey et al., 1998; Glantz and Lewis, 2000; Whitfield-Gabrieli et al., 2009; van den Heuvel et al., 2010). Studies examining neurobiological correlates of brain disorders have shown that regional patient-control differences in neuroimaging-based markers of cortical morphometry and connectivity coincide with microscale differences in those areas, for example, in Alzheimer's disease (Buckner et al., 2009; Prescott et al., 2014) and schizophrenia (van den Heuvel et al., 2016). White matter tree complexity is a novel measure of interest to study patient-control differences in brain wiring. In particular, neurodevelopmental disorders could benefit from examining potential developmental perturbations of the branching structure of white matter bundles, as suggested for changes in neural connectivity at the microscale level (Selemon and Goldman-Rakic, 1999; Zikopoulos and Barbas, 2013; Petanjek et al., 2019).
Footnotes
M.P.v.d.H. was supported by an Aard- en LevensWetenschappen OPen (ALWOP.179) and VIDI (452-16-015) grant from the Netherlands Organization for Scientific Research (NWO) and a European Research Council grant CONNECT (ERC-CoG 101001062). Human data were provided by the Human Connectome Project, Washington University–University of Minnesota Consortium (Principal investigators, David Van Essen and Kamil Ugurbil; 1U54MH091657) funded by the 16 National Institutes of Health and centers that support the Blueprint for Neuroscience Research, and by the McDonnell Center for Systems Neuroscience at Washington University. We thank Dr. Corbert van Eden for guidance in the lab, for sharing knowledge on the staining protocols, and for his insightful comments that helped make this research possible.
The authors declare no competing financial interests.
- Correspondence should be addressed to Lianne H. Scholtens at l.h.scholtens{at}vu.nl
