Introduction

Progressive supranuclear palsy (PSP), a sporadic, late-onset and fatal neurodegenerative disease, is clinically characterized by progressive postural instability and early falls, vertical supranuclear gaze palsy, levodopa-resistant parkinsonism and cognitive impairment (CI) with predominant frontal executive dysfunction1,2. Owing to heterogeneous clinical manifestations, a definitive diagnosis of PSP can be confirmed only at autopsy3. The severity and neuroanatomical distribution of PSP, which is defined pathologically by the accumulation and dissemination of 4-repeat tau protein in neurons, astrocytes, and oligodendrocytes4, differ significantly across individuals3. The common occurrence rate of classic Richardson syndrome in PSP patients is approximately 6 per 100,0005,6,7,8. Various clinical phenotypes, beyond the initially identified Richardson syndrome phenotype, could result from PSP pathology such as parkinsonism mimicking Parkinson’s disease (PSP-P), frontotemporal dementia (PSP-F), and corticobasal syndrome (PSP-CBS), among others9,10,11,12. Despite the increasing number of clinical studies, there is still no effective disease-modifying therapy for PSP13,14. Therefore, studying the neuropathological mechanisms underlying PSP is highly important and may provide a crucial theoretical basis for the early neurological rehabilitation of PSP.

The way we perceive and act, in both healthy and diseased states, hinges on the interconnections among different brain areas. Generally, it is presumed that the robustness of a structural link between two brain regions correlates with the intensity of their respective functional connections15. Recent findings revealed that the link between structural and functional connectivity varies across the healthy human brain and is structured according to distinct hierarchical and cytoarchitectural principles15. Notably, primary sensory (unimodal) cortices exhibit strong structural‒functional coupling (SFC), whereas the apexes of the processing hierarchy (transmodal association cortices) show variation in networks such as the default mode network (DMN)16,17,18. A hypothesis suggests that the comparative separation in advanced regions facilitates abstract thinking, shielded from detailed signals in the initial phases of sensory processing19. Alterations in the SFC take place throughout brain development16, as well as in psychiatric20,21 and neurological disorders22,23,24,25. Nevertheless, whether and how the SFC is disrupted in PSP patients remain unclear.

Hierarchy is a core tenet of organization within the human brain, facilitating the transformation of basic sensory data into advanced mental processes26,27. Traditional studies in neuroanatomy, which focus on the brain after death, have recorded tiered processes in structuring cortical microstructures, including the arrangement of neuronal cells and myelinated nerve fibers27,28,29. The progression of nonintrusive neuroimaging methods30 has enabled the breakdown of connectivity information from the human brain31. Macroscopic brain structures are organized hierarchically along a “sensory‒fugal” axis, extending from sensory‒motor to multimodal regions and facilitating the encoding and merging of information from sensory experiences to cognitive processes27,32. Recently, the emergence of the diffusion map embedding technique has facilitated the transformation of large-scale brain characteristics into simpler, multidimensional forms, specifically gradients31, which act as tiered organizing axes for depicting the continuous range of brain features. Within this range, gradients of functional connectivity represent the spectrum from sensory perception and action to conceptual cognition26, indicating its vital role in underpinning intricate cognitive processes and actions26,31. Concurrently, gradients in structural connectivity reflect slow changes in morphology or microstructure, offering insights into the processes of neurodevelopment26,33. Even with recent advancements in the study of structural and functional gradients31,34,35,36, alterations in PSP remain ambiguous, and the relationships between SFC and these gradients still need to be investigated.

Grasping the structure of the brain necessitates a perspective that spans multiple scales. Differences in cell types and distributions across regions form the basis of large-scale patterns, and histological examination of postmortem tissue has been instrumental in revealing the microscopic structure of the brain. Given the restricted spatial clarity still imparted to scientists by live MRI, cellular-level deductions are impossible, underscoring the importance of identifying links between macroscale patterns and cellular formations to validate physiological observations from MRI and advanced brain-related computational models. BigBrain represents a unique three-dimensional volumetric model of a segmented, cell-body stained postmortem human brain37. The tool enables a detailed computational examination of the whole organ, focusing on cell staining, with resolutions reaching up to 20 µm. This is specifically designed for those who use neuroimaging techniques. Numerous studies have utilized this distinctive tool for comprehensive histological-neuroimaging studies38,39,40,41,42,43.

The structure and features of neurons in the cerebral cortex, known as cortical cytoarchitecture, significantly vary throughout the cortical mantle44,45,46,47. Recent advancements in methodology have led to a move toward more quantitative studies of cortical cytoarchitecture, relying on statistical evaluations of 2D histological sections48,49,50. The main goal of these investigations has been to measure the “cortical profile,” or the change in the intensity of the cell body-stained picture throughout the cortical depth. First, a technique known as Microstructural Profile Covariance (MPC) for examining cortical architecture measures the microstructural resemblance between regions at the individual subject level38. Past studies have shown how useful MPC is for comprehending large-scale cortical architecture patterns; in particular, they have been shown to illustrate a sensory‒fugal gradient of microstructural differentiation in both in vivo and postmortem datasets38.The axis illustrates a steady shift from primary sensory and motor areas characterized by pronounced laminar differentiation, passing through the association cortex to paralimbic regions exhibiting progressively abnormal granular features27. The examination of how cortical profiles differ throughout the cerebral cortex and delineate area boundaries is the next step; this analysis is focused mostly on the central moments, which include the mean, standard deviation, kurtosis, and skewness44,48,49. Essentially, the analysis of the average BigBrain intensity distribution across the entire cortical mantle allows for the detection of neuron density. Skewness analysis contrasts the deep and superficial layers of the cortex, indexing the uneven distribution of cells, which is a key dimension of laminar differentiation51.

Analyzing cortical profiles quantitatively has enhanced our comprehension of the cytoarchitectural diversity of the human cerebral cortex. Neocortical layers are typically characterized by six horizontally overlapping gray matter stripes, each with unique attributes such as neuron size, type, and density, which can be further divided into several sublayers44,47. A key cytoarchitectural characteristic differing in the cerebral cortex is its layered configuration, characterized by the thickness of the laminar layer and the size and density of neurons in each stratum. Certainly, the study of laminar characteristics has been a key aspect of numerous qualitative analyses of human cytoarchitectural differences46,47,52. Nonetheless, comprehending the stratified structure of the human neocortex could shed more light on the role of intracortical circuits in facilitating function53,54.

In this study, we hypothesized that there would be significant differences in SFC between patients with PSP and healthy controls (HC). We further hypothesized that these differences are related to structural and functional gradients and histological characteristics. To test these hypotheses, we first describe replicable PSP-related SFC abnormalities in two independent cohorts. Second, we combined diffusion map embedding methods to construct structural and functional gradients and evaluated the hierarchical differences in gradients between PSP patients and HC. Furthermore, we utilized BigBrain histology data, first constructing an MPC framework to obtain histological gradients, then parameterizing BigBrain’s cellular structural characteristics on the basis of statistical moments previously proposed for histology data. We subsequently describe the data-driven axes of laminar thickness and intensity covariance by quantifying the regional covariation in layer thickness and intensity within BigBrain37,55. Finally, in our exploratory analysis, we examined the correlation between SFC and various gradients and histological features, thereby linking gradients, histological features, cellular features, and the cortex to further interpret our findings.

Results

Data sample

Our study included two separate cohorts: one for the discovery cohort and another for replication. In particular, we utilized a duplicate cohort to corroborate the case‒control alterations in SFC. In the discovery cohort, four patients and two healthy controls were omitted because of excessive movement. Finally, 47 PSP and 80 HC were included. In the replication cohort, three patients and three healthy controls were excluded. 50 PSP and 52 HC were included. Supplementary Tables 1, 2 present the demographic and clinical details of both the patients and the control subjects. No notable differences (p > 0.05) were observed in age or sex among the groups in either cohort. To minimize errors in the model, age and gender served as covariates in the linear analysis of differences between groups. The findings we present are derived from the discovery cohort, unless specified otherwise. Figure 1 presents an overview of the analysis framework.

Fig. 1: Study design and analytical procedure.
figure 1

A Structural‒functional coupling (SFC) analysis. Analyses were conducted via the Schaefer whole-brain parcellation, which included 400 cortical regions. Structural connectivity (SC) and functional connectivity (FC) matrices were derived for each participant from diffusion-weighted imaging (DWI) and resting-state functional MRI (rsfMRI) data respectively. For each participant, regional connectivity profiles were extracted from each row of the structural or functional connectivity matrix and represented as vectors of connectivity strength from a single network node to all other nodes in the network. SFC was then measured as the Spearman rank correlation between non-zero elements of regional structural and functional connectivity profiles. SFC was then compared between groups. B Structural and functional gradient construction. For each subject’s structural and functional connectivity matrices, gradients of connectivity covariance were constructed via diffusion map embedding, a non-linear intricate algorithm that organizes nodes by affinity, with a normalized angle serving as the affinity metric. We focused our analyses on the first 2 principal structural and functional gradients. Gradient scores and SFC may be mapped back to the cortical surface. We then correlated functional and structural gradient scores with SFC differences. C Histology construction. Extracting cell staining changes and high resolution cortical profiles from the BigBrian database. First, building a microstructure profile covariance (MPC) matrix is constructed on the basis of cortical profiles to extract the first two major histological gradients; second, calculating the central moments are calculated and parameterized via cell staining variation; third, the cortex is subdivided into 6 layers on the basis of previous studies, and laminar thickness and intensity covariance (LTC/LIC) matrices are constructed to extract cell features (cellular size and density) from these 6 layers; finally, the SFC is correlated with all extracted histological features to understand the interpretability of SFC from a cellular histological perspective.

Full size image

PSP-related changes in SFC

SFC was quantified by correlating structural connectivity derived from diffusion MRI with functional connectivity obtained from resting-state fMRI. Similar to other studies16,31, the healthy control group presented cortical SFC variations, with higher coupling in the prefrontal cortex and lower coupling in the lateral temporal and frontoparietal areas (Fig. 2A). The PSP maintained this particular pattern.

Fig. 2: Structural‒functional coupling (SFC) differences between participants with progressive supranuclear palsy (PSP) and healthy controls (HC).
figure 2

A The SFC pattern in patients with PSP and healthy controls. The prefrontal cortex exhibited higher coupling whereas the lateral temporal and frontoparietal areas showed lower coupling values. The PSP maintained this particular pattern. B The histogram shows the distributions of the mean SFC values in the PSP and control groups after regressing out the effects of age, sex, and age × sex. C Case‒control comparison (t-map) of regional SFC (first row, unthresholded). Twenty-seven cortical regions showed statistically significant differences (bottom row, p  <  0.05, threshold by BH-FDR correction). D A density scatterplot of the mean regional SFC values of healthy controls (x-axis) and the case‒control t statistic (y-axis) (r  =  −0.1928, pspin  = 0.0286). Most cortical regions have positive SFC in healthy controls, which increases in PSP patients compared with healthy controls (72.7% of regions), or positive SFC in healthy controls, which decreases in PSP patients compared with healthy controls (26.9% of regions).

Full size image

We analyzed the differences in SFC between PSP patients and HC via a GLM with age, sex, and age × sex as covariates. As depicted in Fig. 2B, a notable difference was observed in the mean SFC distribution between cases and controls, following the exclusion of age, sex, and the interaction of age and sex (using a two-sample Kolmogorov–Smirnoff test, p = 0.0013). Region-wise comparisons (BH-FDR corrected p < 0.001) showed that SFC increased in PSP in the left visual network (parts 16, 17, 18, 19, 22, 24, 25, and 29), right visual network (parts 5, 6, 11, 12, 14, 15, 17, 18, 19, 20, 22, 23, 25, 26, and 29), right somatomotor network (parts 19, 23, and 28), and right precuneus and posterior cingulate cortex of the default model network (part 1) (Fig. 2C and Supplementary Table 3). Increased SFC in the visual and somatomotor networks suggests heightened coordination between structural and functional connectivity in these regions, potentially reflecting compensatory mechanisms or disease-specific pathology.

An increase in regional SFC indicates a tighter connection (or increased coupling) between the structure and functional connectivity in these regions, which is interpreted as coordinated fluctuations in neural activity in adjacent cortical areas changing along with anatomical structures (with their consistency elevated); conversely, a decrease in regional SFC implies reduced coupling. The case‒control t-map is significantly correlated with the average control region SFC spatially (Pearson r = –0.1928, pspin = 0.0286, Fig. 2C), indicating that the more connected regions are, the greater the case‒control difference56,57. Negative regional t values and positive average SFC values represent decoupling in PSP individuals relative to HC in 26.9% of the regions, whereas 72.7% of the regions have positive t values and positive average SFC values, indicating higher coupling in PSP individuals than in HC.

We assessed the relationship between case‒control changes in SFC and symptoms via Pearson’s correlation analysis. We included seven clinical variables in this study: the Progressive Supranuclear Palsy Rating Scale (PSPRS) score and its subitems such as daily activities and behavioral symptoms, ocular motor deficits, motor impairment causing bulbar symptoms, limb motor deficits, and gait deficits. We found that the right visual network cortex (part 17) exhibited a significant positive correlation with ocular motor scores (Supplementary Table 4). No significant correlations were found between the SFC and other symptoms.

Structural and functional gradient analysis

Our study focused on determining whether the spatial differences in separating structure and function correspond with the essential characteristics of cortical organization. By employing diffusion map embedding to reduce dimensionality in a nonlinear manner58, we deduced the structural and functional gradients of cortical organization for the structural and functional connectivity matrices of each control participant. Similar to earlier research31,38,59,60, our analysis concentrated on the initial two principal gradients. The first principal gradient accounted for 13.91% of the variance in structural gradients and 17.6% in functional gradients, whereas the second principal gradient explained 12.04% of the variance in structural gradients and 16.04% of the variance in functional gradients.

We evaluated the variability in connectivity represented by the primary two gradients in healthy control subjects. The first gradients, both structural and functional, were anchored on one end in the frontal area and on the opposite end in the occipital areas (Fig. 3A, B). We conducted correlations between each brain area’s weighting in the first gradient (using the mean value for the control group alone) and the associated A‒P axis coordinate for that region to verify this A‒P alignment. The investigation showed no significant correlation for the first structural gradient (r = –0.4805, pspin = 0.1921) or the functional gradient (r = –0.5486, pspin = 0.1137) (Supplementary Fig. 2A, B).

Fig. 3: Relationships between structural‒functional coupling (SFC) and structural and functional gradients.
figure 3

The first two principal gradients derived from the averaged control structural and functional connectivity matrices are presented. Gradient scores are projected back onto the cortical surface. The first principal structural (A) and functional (B) gradients revealed a dissociation between the posterior and anterior regions. The second principal structural (C) and functional (D) gradients revealed a dissociation between unimodal and transmodal regions. E The functional gradient map is from Margulies et al. and is similar to our constructed second principal structural (C) and functional (D) gradients. F We plot the correlation between the SFC t-statistic and the five different gradient scores to understand the relationship and found that SFC is positively correlated with the first functional gradient and negatively correlated with the second structural gradient and functional gradient.

Full size image

Conversely, in the control group, the secondary principal gradients were located in unimodal areas (primary sensory cortex) at one extremity and transmodal zones at the opposite end (Fig. 3C, D). For verification, we categorized each brain area into a hierarchical level on the basis of its respective functional network, transitioning from unimodal (level 1) to transmodal (level 4)61. The weighting of each brain area in the second principle gradient and its hierarchy level were then correlated. The structural (r = 0.4657, pspin = 0.016) and functional second principal gradients (r = 0.7944, pspin < 0.0001) were significantly correlated with this unimodal–transmodal axis (Supplementary Fig. 2C, D).

Subsequently, we assessed the relationships between the case‒control differences in SFC and the structural and functional gradients. Moreover, we compared the relationships between the mean SFC and the gradients in the PSP group and the HC group respectively.

We found that SFC was positively correlated with the first functional gradient (r = 0.4894, pspin = 0.0017), indicating stronger coupling in the posterior regions but less coupling in the anterior regions.The SFC is negatively correlated with the second structural gradient (r = –0.3618, pspin = 0.0416) and functional gradient (r = –0.4758, pspin = 0.0016); that is, unimodal sensory activity regions exhibit stronger SFC, whereas transmodal activity regions show weaker coupling. However, there was no significant correlation between SFC and the first structural gradient (r = 0.2913, pspin = 0.1149). Furthermore, we used the functional gradients of Margulies et al.’s31 previous work for validation and reported that the SFC differences in PSP are consistent with our results, with a clear correlation (r = −0.4938, pspin = 1 × 10−4). In addition, differences were found only in the first functional gradient of the PSP group in both the PSP and HC groups (Supplementary Fig. 3), indicating that the SFC along the A‒P axis of PSP was highly coupled, whereas the other SFC values were not significantly different.

BigBrain histology-based microstructure profile covariance and principal gradient analysis

We extracted the cortical-cortical microstructural similarity of the BigBrain 3D histological reconstruction from the BigBrainWarp toolbox (https://bigbrainwarp.readthedocs.io/en/latest/index.html). Using a technical approach consistent with previous research in Materials and Methods38,62, we ultimately generated an MPC matrix capable of capturing the similarity of the cellular structure between cortical regions.

The method of diffusion embedding, as mentioned in the gradient analysis section, has been recently utilized not only for identifying the intrinsic functional segregation of cortical areas on the basis of resting-state functional MRI31 but also for application to the histology-based MPC matrix. The first main gradient, contributing to 39.98% of the variation, is anchored at one end in the frontal lobe and at the other end in the occipital region (Fig. 4A) which is similar to the first structural gradient mentioned earlier. The second gradient accounts for 29.78% of the variance, anchored at one end by the primary sensory and motor areas and at the other end by the transmodal association and limbic cortices (Fig. 4B), namely the sensory‒fugal gradient axis. We then correlated the statistical t values of the SFC with the two histological gradients and found a positive correlation with the first histological gradient (r = 0.4397, pspin = 4.0 × 10–4, Fig. 4D), but no correlation was found with the second gradient (r = –0.1979, pspin = 0.1063, Fig. 4E). Moreover, we used the histological gradient processed by the ENIGMA toolbox to compare its correlation with that of SFC and found that there was no correlation (Fig. 4C, F). Furthermore, Fig. 4G shows the different distributions of the two histological gradients in the Yeo 7 networks. In the first histological gradient, we found higher gradient values in the visual area, whereas in the second histological gradient, the gradient values of the limbic system were higher.

Fig. 4: Relationships between structural‒functional coupling (SFC) and histological gradients as well as various histological parameters.
figure 4

The top two main histological gradients extracted from BigBrain based on the MPC matrix are displayed on the cortical surface. A The first histological gradient is anchored at one end in the frontal lobe and at the other end in the occipital region which is similar to the first structural gradient mentioned earlier. B The second histological gradient anchored at one end by the primary sensory and motor areas and at the other end by the transmodal association and limbic cortices. C We present histological gradients processed via the ENIGMA toolbox. Panels (D–F) show the correlation between SFC changes and the first and second histological gradients, whereas we validated the gradient patterns via the ENIGMA toolbox, which revealed that only the first histological gradient was related to SFC changes. G Differential distribution of histological first and second gradients in the Yeo 7 network. H Central moment parameters calculated on the basis of cellular staining variations were projected onto the cortical surface. I Different cellular parameters are significantly correlated with SFC, providing new insights at the cellular histological level into the neuropathological mechanisms of PSP. J Different central moment parameters and externopyramidization distributions in the Yeo 7 network.

Full size image

Moments of the intensity profiles

Previous studies63 have shown that the staining intensity curve of the cortex increases nonlinearly from the superficial layer near the dural boundary to the deep layer near the white matter boundary. By utilizing known techniques, we encapsulated the staining intensity profile of each vertex within approximately 1012 nodes of similar size and parameterized them through central statistical moments64. We calculated four metrics, including the mean, standard deviation, skewness, and kurtosis (Fig. 4H), and compared the correlations between the SFC and the central moment parameters, revealing that all the metrics are related to the SFC. Specifically, the mean and kurtosis are positively correlated with the SFC (r = 0.2646, pspin = 0.0055; r = 0.42, pspin = 7 × 10–4; Fig. 4I), whereas the standard deviation and skewness are negatively correlated with the SFC (r = −0.4527, pspin = 5 × 10−4; r = −0.2495, pspin = 0.0382; Fig. 4I). We focus primarily on the first moment (the mean of all cortical depths) and the third moment (the skewness of cortical depth). The mean is used to explore cell or neuron density, whereas skewness analysis contrasts the deep and superficial cortical layers, indexing the uneven distribution of cells, a key dimension of laminar differentiation. Moreover, we calculated externopyramidization65, and the relationship with SFC was positive (r = 0.2233, pspin = 0.0052), indicating a progressive variation in intensity profiles among cortical layers and implying that it distinguished lower cortical hierarchy zones from higher ones via the hierarchical transition in laminar projection patterns62. Furthermore, we compared the different distributions of these indicators in the Yeo 7 network and found that the visual network was higher in terms of the mean and kurtosis, but lower in terms of the standard deviation and skewness, with a slightly higher in externopyramidization (Fig. 4J).

Laminar thickness and intensity covariance with principal gradients and laminar neuronal density and size

Our study employed a BigBrain-derived cortical map, and in alignment with current scholarly methods66, the laminar thickness covariance (LTC) matrix was calculated, illustrating the resemblance in laminar thickness patterns across the cortical areas. Our focus was on the first and second gradients, which account for approximately 31.84% and 17.69% of the observed variance within the LTC respectively. The first gradient of the LTC extends from the lateral frontal area, moving toward the medial frontal area, the temporal area, and the primary visual area, terminating in the parietal and occipital regions (Fig. 5A). Moreover, the second gradient of the LTC transitions from unimodal areas to transmodal areas (Fig. 5B). We employed the same methodology to construct the laminar intensity covariance (LIC) matrix and its first and second gradients (27.6% and 17.98% of the observed variance respectively), with gradient patterns similar to those of the LTC (Fig. 5C, D).

Fig. 5: Relationships between structural‒functional coupling (SFC) and Laminar organization and cell parameters and sensitivity analysis.
figure 5

The laminar thickness covariance (LTC) matrix and laminar intensity covariance (LIC) matrix were constructed on the basis of the segmented 6-layer cortical tissue structure, and their first and second principal gradients were extracted. Panels (A–D) present the first and second principal gradients of the LTC and LIC matrices, respectively. The first gradient of the LTC and LIC extends from the lateral frontal area, moving toward the medial frontal area, the temporal area, and the primary visual area, terminating in the parietal and occipital regions. Moreover, the second gradient of the LTC and LIC transitions from the unimodal areas to the transmodal areas. E We find that SFC is associated mainly with the first and second principal gradients of the LIC. No correlation is found in LTC. F We extracted the cellular size and cellular density features of the 6-layer cortical tissue to explore their relationship with the SFC, primarily finding that the cellular size was negatively correlated with the SFC, whereas the cellular density was positively correlated, most prominently in the fourth layer. G Correlation analysis of all the gradients used in this study to compare their reliability. H The use of Neurosynth for meta-analysis revealed that the PSP-related SFC changes are associated primarily with early visual processing, sighted, and the primary visual cortex.

Full size image

Meanwhile, we calculated the first and second gradients of the relationship between the SFC and LTC with the LIC and discovered that the SFC has no significant correlation with the LTC (r = 0.2671, pspin = 0.0901; r = –0.1292, pspin = 0.1317) but is related to the LIC (r = 0.4042, pspin = 3 × 10–4; r = −0.3271, pspin = 0.0014).

Finally, we explored the relationships between SFC and the density and size measurements of six-layer neurons in the brain. By doing so, we aimed to understand how coupling is associated with other cellular structural features. We observed changes in SFC and laminar neuronal features (Supplementary Fig. 6), most prominently in the fourth layer, with an increase in neuronal density (r = 0.7344, p < 0.001, Supplementary Table 6) and a decrease in neuronal size (r = –0.7813, p < 0.001) (Fig. 5F).

Reproducibility analysis

The pattern of SFC changes in PSP was confirmed through an independent replicate cohort. The identically derived PSP and HC mean SFC maps from the replication cohort were spatially concordant with those from the discovery cohort (r  =  0.6367, pspin  <  0.0001; r = 0.5323, pspin < 0.0001; Supplementary Fig. 1).

Second, we conducted Spearman correlation analysis to further validate the reliability of the differences between the case‒control groups and their relationships with gradients and histological features. The results showed that most correlation indicators based on Spearman correlation analysis were highly reproducible (see Supplementary Table 5).

Third, we conducted a correlation analysis of all gradients (Fig. 5G) and found strong correlations among the first functional gradient and the first structural gradient, the second functional gradient and Margulies’ first functional gradient, the first histological gradient and the LIC’s first gradient whereas the second functional gradient and the second structural gradient, the second histological gradient and the ENIGMA’s histological gradient, the first histological gradient and LTC’s first gradient, the LTC’s first gradient and the LIC’s first gradient showed relatively strong correlations. Finally, there was a relatively strong negative correlation between the second structural gradient and the first histological gradient. Additionally, in Supplementary Fig. 4, we can observe the distribution of differences in case‒control t values, the first and second histological gradients across different cell classifications, with an increase in SFC differences in the granular layer, an increase in the distribution of the first histological gradient in the granular and agranular layers, and a decrease in the distribution of the second histological gradient in the granular layer.

Fourth, we calculated the correlation between SFC and the thickness of the six-layer laminar organization and found that they were not significantly correlated (Supplementary Fig. 5).

Finally, we utilized Neurosynth67 for meta-analysis to understand the correlation between PSP-related SFC and topic terms, revealing that PSP-related SFC changes are associated primarily with early visual processing, sighted, and the primary visual cortex (Fig. 5H).

Discussion

In this study, we explored abnormal changes of the SFC in PSP patients and HC. We identified that the SFC changes map of case‒control differences was spatially correlated with the structural and functional gradient map. Furthermore, we found that SFC changes are associated with the spatial first gradient of histology, not only with various measures of central moments in histology but also with the covariance gradient strength of the hierarchical structure. These findings provide a new perspective on the coordinated changes in structure and function in PSP pathogenesis and their relationships with gradients and the microstructural cellular structure hierarchy.

Previous studies68,69 have shown that detailed anatomical information about the brain system helps constrain descriptions of its function. The SFC describes the structural support for functional communication and occurs when the connectivity between areas of white matter connections in the cortical regions predicts the strength of functional connections between areas16. We found similar SFC patterns in PSP patients and HC, with relatively higher SFC in the primary sensory and medial prefrontal cortex and relatively lower coupling in the lateral temporal and frontoparietal regions. This is consistent with previous research findings16,36. Additionally, we found that the SFC differences in PSP patients were mainly concentrated in the high coupling of visual areas. One of the hallmark signs of PSP is the inability of both eyes to focus correctly, making it difficult to look down, or experiencing blurred vision and diplopia, which specifies its interaction with visual function2. Previous studies70,71 have shown that lesions in the primary visual cortex lead to the loss of conscious visual perception, which has a significant impact on human patients. Our findings indicate that as PSP patients experience structural atrophy changes in the visual network region, the functionality of their neural coordination activities also decreases (i.e., higher consistency). These findings also suggest that PSP patients may experience visual dysfunction. Additionally, abnormalities in the sensory motor area have been observed, indicating that PSP may lead to disturbances in the sensory perception system, which is consistent with the clinical presentation of PSP patients losing balance while walking2. In our current work, we found that anomalies in the visual network and sensory motor-related areas are considered part of the low-level sensory perception system, possibly indicating that the changes of SFC in the perceptual system play important roles in the neuropathological mechanism of PSP.

The hierarchical structure of the cortex provides a unified principle for understanding the anatomy and function of the cortex in primates at multiple scales31,68,72. We found that changes of the SFC in PSP follow a unimodal to transmodal gradient. Several studies16,17,38 in the field of health have shown that the SFC of the unimodal sensory cortex is stronger and that the relative decoupling of the transmodal association cortex is consistent with improvements in executive ability and abstract reasoning. Our second major gradient also reflects the hierarchical structure from unimodal to transmodal and is negatively correlated with the SFC differences in PSP. This further supports the tethering hypothesis that the transmodal association cortex is not constrained by the molecular gradients of the early sensory cortex73. Furthermore, we verified our results via Margulies et al.’s functional gradient31, which was consistent with our research findings. In addition, we observed a significant increase in the differences in SFC along the A‒P axis in PSP (first functional gradient). Observations have shown a gradient from the front to the back in gene expression within the adult human brain74,75,76 and in prenatal stages77,78. Particular patterns of gene expression at this scale might lead to susceptibility during the occurrence of degeneration. Nonetheless, the A‒P gradient reflects not only transcriptional variances but also alterations in cortical microstructural features such as increases in neuronal quantity and density, alongside a reduction in neuron and arbor dimensions along the A‒P axis75. A rising count of neurons in the posterior (visual network) areas might increase their susceptibility to the expansion of alpha-synuclein across neurons.

PSP-related alterations in SFC may be caused by many factors, such as genetic, molecular, and neuronal changes. Recently, histological examination of postmortem tissue has become a powerful strategy for understanding the microstructure and molecular basis of the brain connectome38,62,66. By projecting the histological features of postmortem tissue into the spatial space of the neocortex, we can understand how these axes are determined by intersecting cell type specificity and cell structure gradients. Here, we focus primarily on the first two major histological gradients of cortical tissue and find that the changes of SFC are related mainly to the first histological gradient (anterior‒posterior axis), which combines multiple local gradients and functional topologies, such as the ventral visual stream from the occipital pole to the temporal pole, which achieves the sensory‒semantic dimension of perceptual processing79,80, and the rostrocaudal gradient in the prefrontal cortex, which describes the transition from cognitive processes supporting action preparation to cognitive processes tightly coupled with movement execution81,82,83,84, indicating that we can further understand the high coupling abnormal performance of PSP in the visual network from a cellular histological perspective. The second gradient (sensory-fugal axis) represents an overall organizational principle that combines these local processing streams, while this axis supports the separation of low- and high-order components in the assumed cortical hierarchy structure38. Unfortunately, we did not find any correlation between SFC changes and this parameter. Next, we systematically analyzed the cortical intracellular strength in the direction of the cortical columns and parameterized it through the central moments of the profiles to quantify the cellular structure in vivo. We found that the correlation trends of mean, kurtosis, and externopyramidization were the same, whereas the trends of skewness and standard deviation were the same; thus, we obtained supportive confirmation of the above-mentioned associations between cortical microstructure and SFC changes in PSP patients. In particular, we found a significant opposite correlation between the coupling changes and the mean and skewness of the laminar profiles; these two complementary features exploit the depth-dependent changes in the cell density distribution65,85. As a complementary feature of laminar organization, the externopyramidization index indicates the neuron density ratio between the upper and lower cortical layers of the granule. When the cortex is more differentiated in terms of its cellular structure, it increases, which occurs in the main area65 with a distinct layer IV. Therefore, these depth-dependent cortical cellular structural features suggest greater laminar differentiation in visual network areas sensitive to changes of SFC in PSP. Finally, we attempt to expand upon previous quantitative studies on cortical structural variability by focusing on the layered structure of the cerebral cortex38,44 and evaluating its relationship with changes in SFC. We found that the SFC is related to the first principal axis of laminar intensity covariance, which represents an overall increase in the relative strength of the upper layers from the lateral frontal lobe to the posterior parietal region. This finding is consistent with a previous animal study that showed a relative increase in the height of the hidden columns of the upper layer along the rostrocaudal axis of the cerebral cortex in several rodent and non-human primate species86. We also observed an increase in neuron density, with cell size negatively related to SFC changes, and found the most prominent associations in layer IV. Granule cells, as the main cellular component of layer IV, contribute to the formation of specific sensory cortical areas. This layer serves as a primary target for cortical afferent connections; thus, it is particularly developed within sensory regions. We established a close correspondence between in vivo models and histological measurements. This work enhances the concept of an intrinsic relationship between cell structure and SFC changes in PSP.

PSP is currently considered a 4-repeat tauopathy. According to current research12, the biological/biochemical criteria for PSP are not yet well defined. Although several biomarkers have been identified, including the tau protein87, NfL88, and microRNA89, they are still under exploration. Currently, the clinical diagnostic criteria for PSP, according to the latest 2017 standards2, are mainly divided into four aspects: oculomotor dysfunction, postural instability, movement impairment, and cognitive dysfunction. These features are manifested primarily through clinical symptoms. Neuroimaging plays a supportive role in diagnosing PSP, and our research revealed that when PSP in the VN region, which is sensitive to SFC changes, laminar differentiation is more pronounced. This is mainly related to the higher cell density and smaller size of the internal granular layer. Our findings provide new insights into the consistency of structural and functional changes in the hierarchical structure and cellular histology of PSP pathogenesis.

There are several limitations to the current study that should be mentioned. First, the estimation of structural connectivity employs diffusion tractography streamlines, which are prone to both false positives and negatives90. For an optimal approximation of structural connections, we employed data from multiple shells and enhanced post-processing techniques, such as constrained spherical deconvolution91 and the SIFT2 process92. Second, estimates of functional connectivity are obtainable from rsfMRI data, which are prone to artifacts, especially motion. To mitigate this, rigorous quality assurance and strict exclusion criteria were implemented93. Third, drug use, comorbid conditions, or lifestyle factors may affect outcomes. We performed an initial screening at the beginning of the inclusion process to ensure that PSP patients were on stable medication within one month of undergoing MRI while also ensuring that they did not have other comorbidities. For lifestyle factors, we excluded participants with a history of long-term (over 10 years) alcohol abuse or drug misuse. Fourth, owing to the relative rarity of PSP, the sample size in the current study is relatively small, necessitating future studies with larger samples to further validate SFC in PSP patients. Fifth, in this study, we utilized whole-brain maps of the cortex from the individual BigBrain37,55. This is currently the only available cortex-wide and high-resolution map, and until similar maps are available, it remains unclear to what extent our findings will generalize to other individuals. Furthermore, while we have optimized estimates of structural and functional connectivity, these still remain indirect measures of brain structure and function, factors that need to be considered when interpreting our research results. In addition, our findings did not distinguish between PSP subtypes in the discovery cohort, preventing further study. Moreover, the number of PSP patients with different subtypes in the replication cohort was limited. Therefore, we plan to continue collecting data to increase the sample size of PSP subtypes for subtype analysis. Finally, although our study used cross-sectional data, which provide valuable insights, longitudinal studies on the progression of PSP may offer further insights into the temporal sequence of SFC in PSP patients.

Our research results indicate that the SFC of the entire cortex is severely disrupted in PSP patients, with higher coupling in the visual network being more prominent. We demonstrate that the coupling differences in PSP follow a macroscopic organizational principle from unimodal to transmodal gradients. Finally, we elucidated the histological relevance of SFC changes in PSP. In conclusion, our findings propose a framework to explain SFC changes in PSP patients and offer new insights into understanding the structural and functional consistency changes in PSP patients.

Materials and methods

Participants

The discovery cohort included a total of 133 subjects, 51 subjects with probable PSP and 82 HC. Participants with PSP were recruited from the Four Repeat Tauopathy Neuroimaging Initiative (4RTNI)94,95 and healthy subjects were enrolled from the Neuroimaging Initiative for Frontotemporal Lobar Degeneration (FTLDNI; http://4rtni-ftldni.ini.usc.edu/). For the detailed inclusion and exclusion criteria of this dataset, please refer to this article94. To be included, patients with PSP had to fulfill the NINDS-SPSP criteria96 and be aged between 55 and 90 years. The general exclusion criteria were (i) age younger than 18 years or older than 90 years; (ii) left-handedness; (iii) the presence of magnetic resonance contraindications; (iv) organic intracranial lesions and (v) poor image quality checked before and during neuroimaging data preprocessing. The participants received an MRI scan and a clinical evaluation that included a PSPRS score and its subitems. More details are shown in Supplementary Table 1.

The replication cohort included individuals with PSP (n  =  53) and age- and sex-matched HC (n  =  55), and individuals were acquired and scanned at Qilu Hospital of Shandong University. The detailed inclusion and exclusion criteria can be found in our previously published article97. More details are shown in Supplementary Table 2. The replication study was approved by the Medical Ethics Committee of Qilu Hospital of Shandong University. Every protocol of the study adhered to the guidelines set by the 1975 Helsinki Declaration. Every participant gave their knowledgeable agreement and comprehended the study’s objectives, advantages, and possible hazards.

MRI acquisition

All MRI data were acquired on a 3 T Siemens magnetic resonance scanner. Please see the supplementary materials for detailed parameters of the discovery and replication cohorts.

Data preprocessing

Data processing includes handling anatomical, diffusion, and functional data. We also address issues regarding data privacy and secure storage. For detailed processing procedures, please refer to the supplementary materials.

Brain parcellation and identification of brain network modules

The entire cerebral cortex was split into 400 functionally homogeneous nodes via the Schaefer-400 template98 to identify the nodes of the structural and functional networks. This template has been widely used in previous studies, revealing physically meaningful features of brain organization and superior functional and connectional homogeneity compared with other parcellations98,99. It is defined by the similarity of global functional connectivity and local functional connectivity gradients.

We generated whole-brain structural and functional connectivity matrices for each individual. Additionally, we used the classic 7 network parcellation provided by Yeo et al.100 to define functional modules and allocated each node from the Schaefer-400 template to one of the seven modules98,100: the default mode network and the visual, somatomotor, dorsal attention, ventral attention, limbic, and frontoparietal control modules. This allowed us to assess the between-group differences and examine the impact of SFC.

Construction of the structural connectomes

A hybrid surface/volume segmentation was created101. QSIPrep-preprocessed T1w images and brain masks were used. Then, using the iFOD2 tractography method102 and dynamic seeding with streamlines terminated at the gray–white matter interface, we carried out anatomically restricted tractography with 10 million streamlines103. To minimize biases, we used the spherical deconvolution informed filtering of tractograms (SIFT2) technique92. The structural brain network was built via the obtained collection of streamlines. A 400 × 400 undirected, weighted matrix was created by combining the connectivities that had been weighted by the streamline count.

Construction of the functional connectomes

We use DPABISurf software104 to perform batch processing for functional connectivity in each participant’s native space. The Pearson correlation coefficient between the mean regional BOLD time series was used to measure functional connectivity. To avoid arbitrary criteria and reduce the impact of erroneous connections, we used structural connectivity to guide the creation of the functional connectome. In particular, in the absence of anatomical linkages, we disregarded functional connectivities between ROIs that were entirely based on time series correlations. A 400 × 400 weighted adjacency matrix was created for each participant to reflect the functional connectome.

SFC calculation

We measured the extent to which a brain region’s structural connectivity correlates with coordinated variations in neuronal activity between regions to characterize how SFC evolves in PSP. The structural and functional connectivity matrix for each participant was extracted as vectors of connectivity strength from one node to every other node. The SFC was subsequently calculated as the Spearman rank correlation between the non-zero elements of profiles in structural and functional connectivity for each area16,36,105.

Gradient analysis

With the use of the BrainSpace toolbox106, the development of both structural and functional gradients has undergone preliminary processing. In line with earlier research31,38,59,107,108, our approach involved utilizing only the highest 10% of connections in each row of every 400 × 400 structural and functional matrix, and we calculated a cosine similarity matrix to identify resemblances in structural connectivity patterns. Diffusion map embedding, an algorithm for reducing nonlinear manifolds, was executed to delineate gradient elements that account for the variance in the network in a descending sequence; This algorithm has the ability to map a complex network matrix into a common area (termed low-dimensional embedding), in which areas of the brain with varying degrees of connectivity are nearer to one another, whereas regions with minimal or no connections are more distant. Importantly, certain manifold parameters govern diffusion mapping embedding, affecting the concentration of sampling points on the manifold (α = 0 indicates maximum influence; α = 1 indicates no influence). This research yielded 10 gradient elements via the diffusion embedding algorithm, utilizing the BrainSpace toolbox’s standard configuration, and adhered to the suggested manifold parameters, α = 0.5 and t = 0, maintaining worldwide connections among data points within the embedded space. A gradient template, both structural and functional, was subsequently created using an average matrix derived from PSP patients and healthy individuals, followed by Procrustes rotation to synchronize the person’s structural and functional gradient with the template.

Histological data acquisition and preprocessing

A high-definition, Merker-stained three-dimensional histological model of a human brain postmortem taken from a 65-year-old man was acquired from the freely available BigBrain repository (https://bigbrain.loris.ca/main.php37). After death, the brain was encased in paraffin, sectioned coronally into 7400 20-μm slices, stained with silver to visualize the cellular structures109, and then digitized. The process of manually examining artifacts such as rips, tears, shears, and stain crystallization was succeeded by automated repair methods, which included nonlinear alignment with postmortem MRI, normalization of the intensity, and averaging of the blocks110. The process of 3D reconstruction involves a sequential step-by-step hierarchical method111. Our main focus was on examining 100-μm data to evaluate the uniformity of the results over various spatial dimensions. The calculations were conducted on reversed images, where the intensity of staining indicated the density of the cells and the size of the soma. There are also geometric grids available that mimic the external and internal cortical junctions (namely, the GM/CSF and GM/WM boundaries), featuring 163,842 corresponding vertices for each hemisphere112.

Building the MPC matrix and mapping MPC gradients on the basis of histology

The cortical vertices were segmented into 1012 adjacent cortical ‘nodes’, each spanning approximately 1.5 cm2 in surface area, omitting vertices with median intensities exceeding three scaled median absolute deviations from the node’s median intensity. The segmentation method maintains the limits of the Desikan Killany atlas113 and undergoes a transformation from the conte69 surface to the BigBrain midsurface via nearest neighbor interpolation. Profiles of staining intensity, indicative of neuronal density and soma dimensions on the basis of cortical depth, were produced across 160k surface points in each hemisphere. The residual profiles, derived by adjusting the intensity data for the y coordinate to reflect notable intensity changes from front to back, were averaged within 1012 uniformly scaled, spatially adjoining nodes113. Consistent with earlier research31,114, the MPC matrix underwent proportional thresholding at 90% for each row and was transformed into a normalized angle matrix. The intricate methods and techniques employed align with those used in prior research38 and gradient analysis.

BigBrain statistical moment calculation

We parameterized BigBrain cytoarchitectural characteristics via central moments derived from several cortical depths in the ENIGMA Toolbox (https://enigma-toolbox.readthedocs.io/en/latest/index.html). In accordance with previous research115,116, we analyze four indicators, namely, the mean, standard deviation, skewness and kurtosis. Using the intensity metrics, we determined four key moment characteristics: the mean, standard deviation, skewness, kurtosis, and externopyramidization. The average and standard deviation illustrate the general distribution of intensity in cytoarchitecture across various layers, with skewness signifying a shift in intensity toward either supragranular layers (positive skewness) or a flat distribution (negative skewness), and kurtosis determines if the extremities of the intensity distribution have extreme values. The process of externopyramidization demonstrates a steady transition of intensity levels from the infragranular to the supragranular layers.

BigBrain maps of laminar thickness

As part of the BigBrainWarp toolbox (https://bigbrainwarp.readthedocs.io)117, laminar thickness measurements of the BigBrain surface area were utilized. To minimize the computational expense of the analysis, the BigBrain surface mesh and laminar thickness maps underwent a downsampling process, reducing their size from approximately 164k to approximately 10k vertices per hemisphere. A reduction in the size of the surface mesh was achieved by choosing fewer vertices and aligning the surface with the cortical structure while also downsampling surface details (such as laminar thickness and intensity) by allocating the value of each preserved vertex to the mean value of that vertex and its closest eliminated vertices117.

BigBrain layer-specific distribution of neurons

The density and dimensions of neurons specific to each layer in chosen tissue slices of the BigBrain isocortex, with a resolution of 1 μm, were sourced from the siibra Python package (https://siibra-python.readthedocs.io/en/latest). The creation of this dataset involved manually annotating cortical layers and automatically segmenting neuronal cell bodies (https://github.com/FZJ-INM1-BDA/celldetection)118. The dataset encompasses 111 tissue slices, equating to 80 vertices on the downscaled BigBrain surface.

Laminar thickness covariance and laminar intensity covariance construction

Initially, we omitted brain areas characterized by agranular or dysgranular cortical types owing to the ambiguous delineation of layer borders in these zones52. García-Cabezas and team52 developed a cortical type map by categorizing each von Economo region119 into one of six types: agranular, dysgranular, eulaminate I, eulaminate II, eulaminate III, or koniocortex. The replication of the laminar thickness covariance (LTC) matrix was carried out, following the method previously employed by Saberi and team66. The specific process initially excluded agranular and dysgranular areas such as the cingulate gyrus, prefrontal cortex, temporal pole, and perirhinal cortex, along with the homotypical cortex owing to their absence of a distinct 6-layer architecture. We mitigated the immediate effect of curvature on the thickness of the laminar layer by smoothing it with a dynamic disk and then normalizing its thickness relative to the total thickness at each cortical point to establish a proportional thickness. Ultimately, laminar thickness maps were segmented via the Schaefer 400-region atlas98, incorporating 359 areas beyond the a/dysgranular cortex into the analysis. Parcellation involves calculating the median of vertices in each parcel. The LTC among cortical areas was determined through a pairwise partial correlation of laminar thicknesses, taking into account the isocortex’s average laminar thickness, to detect covariance exceeding the average. Following this, the partial correlation coefficients underwent Z-transformation, leading to the creation of the LTC matrix. Following this, a laminar intensity covariance (LIC) matrix was calculated, akin to the method previously outlined for LTC matrices.

Null model

To mitigate the possible misleading impacts of spatial autocorrelations, a spin test was performed120,121. By randomly rotating the spherical projection of spatial maps, one can produce a collection of null Pearson’s correlation coefficients without altering the spatial correlation. Consequently, in our research, we initially executed 10,000 permutations of spin tests on cortical areas to create a null distribution, calculating the pspin value as the ratio of null correlation coefficient values exceeding the actual correlation coefficient values.

Statistical analysis

The gender, age, and clinical features of the participants were compared between the groups. For continuous variables, independent samples t-tests were employed, whereas χ2 tests were utilized for categorical variables. For details, age was compared via an independent samples t test, whereas gender was compared via a chi-square test.

To assess the spatial distribution, the regional SFC was calculated as an average among all HC participants. A general linear model (GLM), which focuses on regional SFC values as the outcome variable, was utilized to investigate the variances between cases and controls. Age and gender were incorporated as additional variables. For every area, this model was applied, and the bilateral t-statistic (contrast = PSP - HC) was derived. To adjust for the importance of each area, network, or category, the Benjamini–Hochberg method for false discovery rate (BH-FDR) was employed, ensuring p < 0.05.

Average measurements of brain areas exhibiting abnormal SFC in the PSP group were taken to conduct Pearson’s correlations between the SFC and clinical factors in PSP patients, taking into account age, gender, and the interplay of age and sex. In addition, we conducted a Pearson’s correlation analysis between the SFC t statistic values and various gradients and histological characteristics. The results of the correlation were adjusted using BH-FDR, with p < 0.05.

Reproducibility analysis

The case‒control changes mentioned in SFC were verified in the replication cohort. With the same method used for the discovery cohort, the case‒control t map was obtained in the replication queue via GLM regression, excluding age and gender. To evaluate the repeatability of regional changes in SFC, spatial similarity analysis of PSP and HC mean coupling maps was carried out between the discovery cohort and the replication cohort. Second, we used Spearman correlation to replicate the above-mentioned Pearson correlation analysis results. Third, we also analyzed the correlation of all gradients and compared their correlation and reliability in this study. To further verify the biological basis of the significant difference in the SFC and histological gradient, we mapped the von Economo and Koskinas cytoarchitectonic atlas47 to the surface and compared the distributions of different cortical regions. Fourth, we extracted a six-layer laminar organization from the BigBrain, projected it onto the cerebral cortex to show its distribution, and calculated the correlation between the SFC and each layer thickness. Finally, we used Neurosynth67 to conduct a meta-analysis to determine which terms are related to the SFC of PSP patients.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.