Introduction

Humans and macaques share a common ancestor, but they have diverged evolutionarily approximately 25 million years ago (Hill et al. (2010)). As a result of genetic changes, environmental factors, and selective pressures (Lecouvet et al. (1997)), they have developed distinct brain structures and functions. Cortical folds are important features of primate brains and play a crucial role in various cognitive and behavioral processes, including perception, action, and cognition (Fornito et al. (2004); Cachia et al. (2018); Yang et al. (2019); Whittle et al. (2009)). Studies have revealed differences and similarities in fold morphology and brain function between humans and macaques (Semendeferi et al. (2002)). Furthermore, there is a intricate relationship between the similarities and differences in cortical folding morphology and the similarities and differences in brain function. For example, humans possess a larger prefrontal cortex compared to macaques, which gives them executive functions such as planning, decision-making, and working memory (Semendeferi et al. (2002)). The higher cognitive and affective functions observed in human compared to macaque are also associated with the larger proportion of their association cortex in the cortical surface (Glasser et al. (2013); Rilling (2014); Rolls and Grabenhorst (2008)). The variations in cortical folding morphology, as well as the differences in brain function, may reflect the adaptation of species to diverse cognitive, social, and ecological demands. Despite variations in morphological and functional characteristics of the cortical folds among different species, there are also many commonalities, indicating relative conservation in the evolutionary process (Van Essen et al. (2019)). For example, gyrencephalic primates which share many primary sulci, such as the lateral, superior temporal, and (except for the marmoset) central sulci, exhibit similarities in both morphology and brain function (White et al. (1997); Ferrier (1873); Friedrich et al. (2021)). Additionally, by comparing the brain activity of chimpanzees during tasks with nonsocial tasks and at rest, it was found that the cortical midline areas of chimpanzees deactivate during these tasks. This suggests that the DMN of chimpanzees is anatomically and functionally similar to that of humans (Barks et al. (2015)). Some studies have found that in species including humans and monkeys, strongly interconnected regions are consistently separated by outward folds, whereas weakly connected regions are consistently separated by inward folds. This folding pattern is associated with brain connectivity, suggests a certain similarity in the mechanisms underlying cortical folding in humans and monkeys (Essen (1997); Sereno et al. (1995); Sousa et al. (1991)). In summary, the folding patterns and functional profiles of cortical regions demonstrate both similarities and differences across different species. These similarities may reflect evolutionary conserved functions, while the differences may indicate species-specific features (de Lange et al. (2019); Buckner and Krienen (2013); Patel et al. (2015)).

Most of the current cross-species studies are based on one or several anatomical landmarks (Eichert et al. (2019); Goulas et al. (2014); Van Essen and Dierker (2007); Van Essen et al. (2018)), which cannot really solve the cross-species analysis needs of the whole brain. The definition of a whole-brain anatomical landmark across species is a complex task due to differences in brain size, cortical folding patterns, and the relative size and location of different brain regions. As a landmark defined based on cortical fold pattern, the gyral peak, as the maximum in height on the cortex, has been discovered and studied in both humans and macaques (Zhang et al. (2022, 2023)). It is defined as the local highest point of the gyri. In our previous work, there are many similarities between the findings of humans and macaques regarding gyral peaks. For example, both species were able to detect consistent gyral peaks among individuals on the cerebral cortex. And it was even consistent across ages in the longitudinal macaque dataset (Zhang et al. (2022)). In both of these works there is a discussion of peaks’ height and inter-individual consistency. In the macaque study we found that the more consistent peaks are located in the higher and thinner cortex, while in the human study we demonstrate that the higher peaks are more consistent among individuals (Zhang et al. (2023)). These two findings corroborate each other. While some findings are not entirely the same between humans and macaques. For example, the higher consistency peaks in humans possessing smaller structural connectivity properties, while the conclusion is opposite in macaques. In addition, there are some conclusions that have been verified on only one species. Based on the aforementioned advantages of gyral peaks, they are highly suitable as anatomical landmarks for cross-species research to infer the developmental and evolutionary aspects of cortical folding and brain functionality.

Here, we investigated shared and unique peaks across individuals and species. We first identified the group-wise peak clusters of human and macaque brains, respectively, and aligned the macaque peak clusters onto the human brain surface using cross-species registration methods (Xu et al. (2020)). This allowed us to identify the shared and unique peak clusters between the two species. Then, we compared the inter-individual consistency of shared and unique clusters within each species, and investigated whether there was a relationship between the inter-individual consistency of shared clusters between human and macaque. Additionally, we examined the anatomical features of these shared and unique clusters and calculated the functional and structural connectivity matrices of the human and macaque brains. Then we used Brain Connectivity Toolbox (BCT, Rubinov and Sporns (2010)) to compute the node features of shared and unique clusters. Furthermore, we examined the spatial relationships between these clusters and different brain regions of multiple atlases. Finally, we used human brain RNA-seq data to select important genes from shared and unique peaks in classification tasks. Our study provides a medium based on cortical folding patterns for cross-species cortical analysis. Through such a medium, we can explore the derivation and specialization of human and macaque brain and understand the rules of how the brain is constructed during development and evolution (Krubitzer (2007)).

Results

Locations of Shared and Unique Peak Clusters

To obtain shared and unique gyral peaks between species, we first extracted peak clusters for each species. The definition of peaks and the method for extracting peak clusters within each species are described in the section. Then, we employed a cross-species registration method (Xu et al. (2020)) to align the macaque clusters onto the human brain surface. Figure.1 (a) top and middle panel displays the locations of all peak clusters found on both human and macaque brains (Human: LH-96, RH-96; Macaque: LH-42, RH-43). Then we use the cross-species registration method Xu et al. (2020) to register the peak clusters of the macaque brain onto the human brain surface (Figure.1(a) bottom, identical color-coding applied across both templates to signify the same peak cluster). Next, based on the definition of shared peak clusters, we identified shared and unique gyral peaks between the two species. Figure.1 (b) shows the locations of shared peak clusters between the two species, with 25 shared peaks in the left hemisphere while 26 in the right (the locations of all human shared peaks are reported in Table 1). For the purpose of comparison, the shared gyral peak clusters of two species were displayed on the surface of the human brain template (Conte69) with the same color coding for corresponding peak clusters on the two species. Figure.1 (c) shows the locations of unique peak clusters found in each species, with 141 (LH-71, RH-70) unique peak clusters found in the human brain and 34 (LH-17, RH-17) found in the macaque brain. The unique peaks found in the human brain were mapped onto the human brain template surface (Conte69), while those found in the macaque brain were mapped onto the Yerkes19 (Van Essen et al. (2012a)) template surface. It is worth noting that for each species, the union of the clusters in Figure.1(b) and Figure.1(c) is the same as the clusters in Figure.1(a) (including color).

Figure 1.

Table 1.

To investigate the regions where shared and unique peaks are located, we utilized the Cole-Anticevic Brain-wide Network Partition (CA network, Ji et al. (2019)), which includes in total 12 functional networks (Figure.2 (c)) based on the MMP (multimodal parcellation, Glasser et al. (2016)). We then projected the human Cole-Anticevic network on the macaque surface using Xu et al. (2020) to qualitatively compare the differences in the distribution of cluster centers between human and macaque. It is important to emphasize that while the shared peak clusters were obtained through cross-species registration, and the human brain network (Cole-Anticevic) was also transferred to the macaque surface using cross-species registration, it is still meaningful to compare the distribution of shared peak centers between humans and macaques. This is because the intersection of clusters (one of the definition of shared peak clusters) does not necessarily imply that the centers of peak clusters are located in the same brain region. We counted the number of shared and unique peaks distributed in different brain regions. In human brain, most shared gyral peaks are distributed in the lower-order cortex such as somatosensory cortex (SMN), primary visual cortex (V1), and secondary visual cortex (V2) (Figure.2 (a)), while most human-unique gyral peaks are located in the higher-order cortex such as default mode network (DMN), Cingulo-Opercular (CON), and Frontoparietal (FPN) (Figure.2 (b)). In the macaque brain, shared peak cluster centers most distributed in the V2, DMN, and CON (Figure.2 (d)), while unique peak cluster centers most distributed in the DMN, Language (Lan), and Dorsal-attention (DAN) (Figure.2 (e)). In order to eliminate the influence of different brain area sizes on the count of cluster, we normalized the count by the regional surface area and reported in FigureA1 (a). The results were similar to those on the original counts. Our findings indicated that most of the shared peaks are located in low-order sensory and motor areas, while most of the unique peaks are located in higher-order cortices.

Figure 2.

Consistency of Unique/Shared Peak Clusters

In our previous articles, the inter-individual consistency of peaks is a measure to assess whether peaks exist stably in different individuals (Zhang et al. (2022, 2023)). To explore the consistency of shared and unique peak clusters in macaques and humans, we calculated the mean count covered by these clusters and normalized by the number of individuals, as presented in Figure.3(a). In both human and macaque, the consistency of shared peak clusters is significantly greater than that of unique peak clusters (p<0.001). Additionally, the overall consistency of peaks in the macaque brain is much higher than that in the human brain, indicating that the peaks in the different macaque brain are more concentrated in spatial distribution. Furthermore, we performed linear regression analysis on the average counts of all corresponding shared peak clusters of human and macaque. The horizontal and vertical axes of the Figure.3(b) represent the average count of shared peaks in the macaque and human brains, respectively. The Pearson correlation coefficient (PCC) of the inter-species consistency of the left and right brain is 0.20 and 0.26 (p>0.05 for all), respectively.

Figure 3.

Figure 4.

The result of linear regression shows that there is a positive correlation in the inter-individual consistency of shared peaks between macaque and human brains.

Anatomical Features of Shared and Unique Peaks

We then calculate the mean of the anatomical features of shared and unique peaks on all individuals of both species. The shared and unique peaks in each individual were obtained by calculating the intersection between the group-wise shared and unique clusters and the gyral peaks in each individual. We found that, whether in human or macaque, the sulcs and local surface area of shared peaks are larger than those of the unique peaks, but the curvatures are smaller. Due to issues with MRI data quality and technical limitations, we only reconstructed the white matter surface of the macaque brain and did not have the gray matter surface available. Therefore, it was not possible to calculate cortical thickness for the macaque dataset. Additionally, due to the unavailability of T2 data in the macaque dataset, the myelin feature was also missing. For the exclusive anatomical features of human, shared gyral peaks are located in cortical regions with thinner cortex but larger myelin in contrast of unique peaks (Table 2). Our statistical analysis using t-tests demonstrated that the p-value for shared and unique peaks of all features is less than 0.001 except local surface area of macaque.

Table 2.

Functional Connectivity Characteristics of Shared and Unique Peaks

Table 3 shows the mean and standard deviations (std) of node indicators of the functional connectivity for all shared and unique peak clusters in human and macaque, including degree, strength, clustering coefficient (CC), betweenness, and efficiency. In general, our results demonstrate that shared peaks exhibit significantly (p<0.001) larger degree, strength, clustering coefficient, betweenness, and efficiency values than unique peaks (except for betweenness of macaque) for the functional connectivity characteristics. The p-values resulting from t-tests for the shared and unique peaks of each feature are presented in the Table 3, indicated with asterisks. In addition, we also make a comparison between shared and unique peaks on the structural connectivity matrix of human brain, and the results are presented in the Supplementary Information (due to the poor tracking effect of dti fiber tractography in the macaque data, we only calculated the structural connection matrix of human brain).

Table 3.

Spatial Relationship Between Peaks and Functional Regions

To assess the relative spatial relationship between the two types of peaks and different brain regions, we calculated the number of brain regions where each type of peak appeared within a 3-ring neighborhood. We utilized various types of brain atlases, including functional, structural, and cytoarchitectural based ones. These atlases were critical as they enabled the inclusion of various characteristics of the brain and facilitated the outlining of the two peaks’ spatial pattern across multiple references. Table 4 present the results for 10 human brain atlases and 2 macaque brain atlases. The results of the t-test are represented in the table using asterisks. The results indicated that shared peaks are surrounded by a larger number of brain regions compared to unique peaks. Results of other atlases can be found in Supplementary Information.

Table 4.

Gene Analysis of Shared and Unique Peak Clusters Based on Lasso

Finally, to study whether there are significant differences in gene expression between the two types of peaks, we utilized the surface-based gene expression dataset AHBA (Arnatkevičiūtė et al. (2019)) and employed the widely used lasso method for gene selection. The preprocessed AHBA gene data is in the form of region×gene and the region above referred to the parcellation of a certain atlas, such as Aparc, Schaefer100, Schaefer500, Schaefer1000, etc. We finally selected Schaefer500 atlas for this study because high resolution may result in some areas with no gene data (more details refer to Arnatkevičiūtė et al. (2019)), while low resolution may result in multiple categories of clusters being located in the same region. Therefore, Schaefer500 was chosen as the most suitable atlas for our analysis. Before using lasso for feature selection, the determination of the Lambda parameter is necessary to regulate the number of selected features. For parameter selection, we employed 10-fold cross validation is used for parameter selection. By considering the maximization of accuracy (acc) and minimization of mean squared error (MSE) simultaneously, the lambda value was ultimately determined to be 0.027 (Figure.5 (b)). The accuracy of training set was 0.84, and the MSE was 0.64; The accuracy of test set was 0.75, and the MSE was 1.00. Finally, we used lasso to select 28 genes that play a significant role in the classification of shared and unique peaks. For the selected genes, we identified significant differential expression in 7 genes between unique and shared peaks using Welch’s t-test. These genes are: PECAM1, TLR1, SNAP29, DHRS4, BHMT2, PLBD1, SERPINB9P1.

Figure 5.

Discussion and Conclusion

In this study, 192 gyral peaks were detected in the human brain, and 85 gyral peaks were detected in the macaque brain. Additionally, 51 pairs of shared peaks (25 in the left and 26 in the right hemisphere) were identified using cross-species registration, as previously reported by (Xu et al. (2020)). The following findings were observed:

1. Spatial distribution: Through the count and frequency of cluster centers’ position, it became apparent that a large proportion of the shared peak occur in the low-order cortex, while unique peaks of human are mainly located in higher-order cortex.

2. Consistency: The inter-individual consistency of shared peaks within each species was greater than that of unique peaks, and there was a positive correlation in consistency was observed between shared peaks in both human and macaque brains.

3. Anatomy: In both human and macaque, it can be found that the sulcs and local surface area of shared peaks are larger but the curvatures are smaller compared to unique peaks in each species. Furthermore, for the exclusive anatomical features of human, shared gyral peaks are located in cortical regions with thinner cortex but larger myelin in contrast of unique peaks.

4. Brain connectivity: Shared peaks in the structural (human only) and functional (human and macaque) graph metrics exhibited higher values for degree, strength, clustering coefficient, betweeness and efficiency compared to unique peaks in both species (except for betweeness of the macaque).

5. Relationship with brain regions: Shared peaks were located in foci surrounded a greater variety of brain regions than unique peaks in multiple brain atlases in both species.

6. Gene analysis: Using Lasso to perform feature selection on all genes, we found some genes related to brain function played an important role in the classification of shared and unique peaks.

We observed that shared peaks are primarily present in unimodal areas, including sensory and motor regions, with fewer occurrences in the default mode network (DMN) and ventral multimodal cortex. In contrast, unique peaks are mostly localized in regions associated with DMN, language, cingulo-opercular, orbito-affective, and frontoparietal cortex. These findings are in line with previous conclusions that cortical regions associated with motor and sensory functions are relatively conserved across species (Hopkins et al. (2014); Krubitzer (2007); Xu et al. (2020); Teissier and Pierani (2021)), while the unique peaks on human appear in brain regions that are specific to human, such as language areas. One possible explanation is the disproportionate expansion of multiple, distributed regions of association cortex relative to sensory regions during species evolution (Krubitzer (2007); Buckner and Krienen (2013)). The expansion of these regions, which untethered them from the constraints of sensory hierarchies and established species-specific functional associations, is the foundation of the ‘tethering hypothesis’ (Buckner et al. (2013)). Evolutionary psychology and neuroscience indicate that this differential regional allometric growth arises from developmental constraints and represents an adaptive adjustment by the brain to optimize its functional organization (Montgomery (2013); Montgomery et al. (2016); Willemet (2015)). Recent research supports this postulation from the perspective of brain function, indicating that the DMN as an important region within the cortex whose function changed throughout the process of evolution (Xu et al. (2020)). Based on these studies, interspecies conservation of sensorimotor regions and uniqueness of higher-order brain regions are easily understood, and our study provides additional supports to this viewpoint by examining cortical foldings.

The consistency of peaks across individuals is an important indicator. It reflects the similarity in cortical folding morphology among individuals. When we compared the consistency of shared and unique peaks, we discovered that shared peaks exhibit greater stability within the same species across different individuals. This observation was as we expected because the similarities in folding patterns could be related to preferences for neurons to migrate in cortical areas (Kriegstein et al. (2006); Friedrich et al. (2021)) and genetically coded Friedrich et al. (2021). These genes, which regulate the cortical structural morphology, are likely conserved in both human and macaque brains over time. Therefore, this has led to the stable presence of cortical folding patterns in both human and macaque species. Moreover, we found that the overall consistency of peaks in the macaque brain is much higher than that in the human brain, indicating that the spatial distribution of gyral peaks in the macaque brain is more concentrated across individuals compared to humans. This is possibly due to the simpler folding patterns that are more easily retained between individuals in macaque and human brains. Next, we computed the correlation between shared peaks consistency across species and found a positive correlation between the consistency of human and macaque (3). This implies that peaks that are widespread in the human brain are also widespread in corresponding regions in the macaque brain. These findings further supports the homology of human and macaque brain structures (Sereno and Tootell (2005); Modha and Singh (2010)).

The spatial distribution of shared/unique gyral peaks across species, as defined by our study, is not random but shows discernable patterns, which can be verified through statistical analysis of anatomical features (2). The shared peaks, in comparison to the unique peaks, exhibit larger sulc and local surface area, but smaller curvature. Furthermore, in human-specific anatomical features (not available in the macaque dataset), the shared gyral peaks exhibit thinner cortex and greater myelination. The associations among these anatomical features can validate the regularity of the distribution of shared and unique peaks. The associations among these anatomical properties of the brain have been extensively verified in previous studies, including the strong positive correlation between sulc and local surface area (Yang et al. (2016)), the negative correlation between myelin and curvature in most regions (Schmitt et al. (2021)), and the negative correlation between local surface area and cortical thickness (Maingault et al. (2016)). These findings confirm the validity of the anatomical characteristics of shared peaks and unique peaks. While many studies have confirmed the positive correlation of sulc and curvature throughout the whole brain (Yang et al. (2016)), the sulc and curvature in our conclusion displayed opposite trends in both shared and unique clusters. Possible explanations to this are many folds: Firstly, all of our gyral peaks are defined within the gyri, and the correlation between sulc and curvature within the gyri is much weaker than that in the whole brain. In addition, the correlation between sulc and curvature in some areas is very low, such as the anterior cingulate (most are shared peaks), dorsolateral frontal cortex(most are unique peaks), and middle temporal gyrus (most are unique peaks) (Schmitt et al. (2021)). This non-uniform spatial distribution leads to the disappearance of the correlation between sulc and curvature. Therefore, the anatomical patterns within the peaks and the global patterns of the entire brain are not in conflict.

Through evaluating the structural and functional connectivity properties of shared and unique peaks, we observed that shared peaks exhibit larger connectivity attributes, such as degree, strength, clustering coefficient, betweeness and efficiency, compared to unique peaks. Higher degree and strength values suggest that shared peaks are connected to more vertices in the brain network. Additionally, we found that clustering coefficient and efficiency, which measure local information transmission capacity and resilience to random attacks in a network, were higher in shared peaks. Betweeness, a centrality measure that quantifies the importance of a node in the network, also showed higher values for shared peaks, indicating greater importance of these peaks in the brain network. These results suggest that shared peaks may play a role as network hubs in contrast to unique peaks. Gyral peaks exhibit a high degree of connectivity within local neighborhoods, creating a “small world” structure within the network, and may behave as hubs in the structural/functional network, as suggested by previous studies (Sporns and Zwi (2004); Bassett and Bullmore (2006); Bullmore and Sporns (2009); He and Evans (2010)). In many studies, higher-order brain regions like the DMN are recognized as the global network hubs and the communication centers of the brain’s global network. These regions typically exhibit higher node degree and strength. However, there is an interesting finding in our study. That is, most of the shared peaks, mainly distributed in lower-order brain regions, have larger network properties compared to the unique peaks, mainly distributed in higher-order brain regions. There are two possible explanations for this. Firstly, peaks is defined at a much more local scale, in contrast to the definition of brain functional regions, such as DMN. This seemingly contradictory findings could be reconciled by their definitions of “network hubs” at respective coarse and fine scales. Specifically, while higher-order brain regions such as DMN serve as the information exchange centers for large-scale brain network, the information transfer within each region at a finer scale could be primarily facilitated by loci, such as the shared peak. These findings suggest that, peaks that are in larger-scale DMN while exhibiting lower hub-like attributes at a vertex-level, could be referred to as provincial hubs (Guimera and Nunes Amaral (2005); Hwang et al. (2017)). This can be understood as the preservation of the most fundamental and mainstream topological structure and communication patterns during the evolutionary process of species, while species-specific peaks that appear later in the evolutionary process may serve higher-order and more specific functions (Goulas et al. (2014); Rilling (2006)). Another issue worth discussing is the relationship between degree and clustering coefficient. Some studies focusing on social networks and random intersection graph models have found that clustering coefficient correlates negatively with degree Foudalis et al. (2011); Bloznelis (2013). While in our study, when comparing the functional network characteristics of shared and unique peaks, we found that the patterns of degree and clustering coefficient were similar (3). The differences in network characteristics between brain networks and social networks or random networks may reflect distinct organizational patterns in the brain compared to other networks. Furthermore, due to our focus on the internal properties of peaks in our study, the patterns observed may not align entirely with the principles followed by the entire brain network.

Through comparisons with multiple brain atlases, we found that the surrounding brain regions of shared peaks are more extensive compared to those of unique peaks. This finding may suggest a higher diversity of brain functions associated with shared peaks. From a microscopic perspective, brain function is determined by the structure and functional characteristics of cells. The brain is composed of various types of cells, and each type of cell contributes to different aspects of brain function. The differential expansion of cortical regions and the introduction of new functional modules during the process of evolution may be the result of changes in progenitor cells (Clowry et al. (2018)). In our experiment, the shared peaks represent regions with less cortical expansion, indicating a smaller proportion of ancestral cells. This may allow them to participate in a greater variety of brain functions and be surrounded by more diverse brain regions. From a macroscopic perspective, in the analysis of brain folding, a traditional approach is to partition the brain into a set of distinct regions, known as parcellation, based on functional, structural, or cytoarchitectural criteria. This parcellation serves as the most common unit of analysis in studying brain folds. This well-defined partitioning method provides an intuitive framework for analyzing the brain, leading to computational, statistical, and interpretational efficiencies(Eickhoff et al., 2018; Glasser et al., 2016b). Simply averaging all vertex characteristics within a region assumes the homogeneity within the region and only one dominant pattern (Haak and Beckmann, 2020). However, both functional and microstructural properties often highly variable within a region, and inconsistent across modalities. Additionally, adjacent vertices in different regions may also have similar characteristics. Therefore, boundaries vary depending on the chosen modality, and no clear boundaries are evident in all modalities or analysis approaches. The brain has no true “boundaries”. In our study, we observed that shared peaks in regions surrounded by a larger number of neighboring brain regions are more likely to be assigned to the “boundaries” of those regions across different classification approaches. Therefore, we speculate that these shared peaks might be involved in a more diverse range of brain functions.

Using lasso regression, we screened the genes on the cortex and identified 28 genes that made significant contributions to the classification of shared and unique peaks. Further applying Welch’s t-test, we found significant differential expression in 7 genes between the shared and unique peak regions. Among them, SNAP29 and KCNH5 are closely associated with neuronal activity and brain function, and these two genes show higher and lower expression levels in the shared peaks, respectively. While, low expression of SNAP29 protein levels disrupts neural circuits in a presynaptic manner, leading to behavioral dysfunctions Yan et al. (2021). Therefore, the majority of shared peaks located in lower-level brain regions exhibit higher SNAP29 expression, aiming to minimize the occurrence of low SNAP29 expression that could disrupt neural circuits and result in behavioral dysfunctions. Another differentially expressed gene was KCNH5. The voltage-gated Kv10.2 potassium channel, encoded by KCNH5, is broadly expressed in mammalian tissues, including the brain. According to previous studies, dysfunction of Kv10.2 may be associated with epileptic encephalopathies and autism spectrum disorder (ASD) (Hu et al. (2022)). And these two diseases happen to be more prevalent in humans, coinciding with the high expression of the KCNH5 gene in unique peaks.

Materials and Methods

Dataset Description

Human MRI

In our study, we utilized the Human Connectome Project (HCP) S900 Subjects MR imaging data from Q3 Release (https://www.humanconnectome.org/). The data was obtained from the Q3 Release and all participants involved provided written informed consent and the study was approved by the relevant institutional review boards. The MR images were acquired by a Siemens “Connectome Skyra” 3T scanner housed at Washington University in St Louis using a 32-channel head coil. For T1-weighted MRI: TR=2400 ms, TE=2.14 ms, flip angle=8 deg, FOV=224× 224 mm and resolution= 0.7×0.7×0.7 mm³. T2-weighted MRI: TR=3200 ms, TE=565 ms and resolution=0.7×0.7×0.7 mm³. Diffusion MRI (dMRI): TR=5520 ms, TE=89.5 ms, refocusing flip angle=160 deg, flip angle=78 deg, FOV=210×180 mm, matrix=168×144, resolution=1.25×1.25×1.25 mm³, 1.25 mm isotropic voxels, echo spacing=0.78 ms, BW=1488 Hz/P x. Resting state fMRI (rfMRI): TR=720 ms, TE=33.1 ms, flip angle=52 deg, FOV=208×180 mm, matrix=104×90, 1200 time points, 2.0 mm isotropic voxels, BW=2290 Hz/P x.

We applied the standard HCP MR structural pipelines (Glasser et al. (2013); Fischl (2012); Jenkinson et al. (2002, 2012)) for processing all structural MR images. This involved the following three main steps: 1) PreFreeSurfer pipeline (Jovicich et al. (2006); Van der Kouwe et al. (2008); Smith (2002)) which corrected for image distortion, aligned and averaged T1w and T2w images and registered the subject’s native structural volume space to MNI space. 2) FreeSurfer pipeline (Dale et al. (1999); Fischl et al. (1999, 2002); Ségonne et al. (2005)) including segmentation of brain volume, reconstruction of white matter and pial surfaces, and registering to fsaverage surface atlas; 3) PostFreeSurfer pipeline, including surface registration to the Conte69 surface template (Van Essen et al. (2012b)) by using MSM-All algorithm (Glasser et al. (2016); Robinson et al. (2014, 2018)). In this step, cortical folding, myelin maps, and resting state fMRI (rfMRI) correlations together for registration, which improved the cortical correspondences across different subjects. For our study, the white matter cortical surface with 64,984 vertices after MSM-All registration and the associated cortical folding features such as sulc, myelin, and cortical thickness, were adopted for cross-subjects analysis. For the diffusion MRI (dMRI) data, we performed fiber tractography using MRtrix3 (Tournier et al. (2019), https://www.mrtrix.org)). Each individual had 40,000 fiber tracts reconstructed (). Amaximum length limit of 150mm was defined to reduce the presence of false positives (Varriano et al. (2018)).

Gene Expression Data

The AHBA microarray gene expression data consists of 3702 samples from 6 typical adult human brains. Several hundred samples (mean ± standard deviation: 617±241) were collected from cortical, subcortical, brainstem and cerebellar regions in each brain to profile genome-wide gene expression. In the AHBA, each gene probe is associated with a numerical ID and a platform-specific label or name. If a probe is assigned to represent a unique gene it is also characterized with a range of gene-specific labels such as gene symbol and an Entrez Gene ID–a stable identifier for a gene generated by the Entrez Gene database at the National Center for Biotechnology Information (NCBI). The probe-level data offer high-resolution coverage of nearly the entire brain, providing expression measures for over 20,000 genes from 3702 spatially distinct tissue samples. The AHBA data is available at figshare https://figshare.com/s/441295fe494375aa0c13. The AHBA dataset has been preprocessed, and detailed information can be referred to (Arnatkevičiūtė et al. (2019)). The first six processing steps produce the region×gene matrix that can be used for the regional analyses.

Macaques MRI

We selected rhesus macaque monkeys’ structural and functional MR imaging data aging from 0.8-4.5 years from the non-human primate (NHP) consortium PRIME-DE from University of Wisconsin– Madison (http://fcon_1000.projects.nitrc.org/indi/indiPRIME.html). The full dataset consisted of 592 rhesus macaque monkeys (macaca mulatta) scanned on a 3T with a 4-channel coil. For T1-weighted MRI: TR=11.4 ms, TE=5.41 ms, flip angle=10 deg, image matrix=512×248×512 and resolution=0.27× 0.50×0.27 mm³. The rsfMRI data were preprocessed based on DPARSF, which included slice timing, realignment, covariant regression, band-pass filtering (0.01-0.1 Hz), and smoothing (FWHM=4 mm). We fed T1w images to CIVET, registering it into the NMT-standardized space (Seidlitz et al. (2018)) using an affine transformation, followed by image resampling and tissue segmentation. The recon-structed white matter cortical surface was obtained using Freesurfer. We resampled the surfaces to 40k vertices to ensure vertex-to-vertex correspondence across subjects by spherical registration. After linear registration between fMRI and T1w MRI via FLIRT, we mapped the volume time-series to surface vertices for further analysis.

Peak Cluster Extraction

Based on our previous work, gyral peaks are defined as the highest point of the gyri (Zhang et al. (2022)). Gyral height was measured by “Sulc” (https://surfer.nmr.mgh.harvard.edu/, Fischl (2012)), which was defined as the displacement from a vertex on the surface to a hypothetical mid-surface, which is between the gyri and sulci, and the “mean” of displacements of all vertices is zero (Fischl et al. (1999)). Thus, gyral peaks on individuals were identified by locating the vertex of the minimum sulc value within the x-ring (4-ring for humans, 3-ring for macaques) neighborhood on the grid (Zhang et al. (2022, 2023)). To obtain group-wise peak clusters, all gyral peaks in individual spaces of the two species were projected onto the respective template white matter surface, which produced a count map of peaks for each species. Of note, vertex-to-vertex correspondences were established across all surfaces within each species. Next, peak count maps of two species were processed by anisotropic smoothing, with n iterations within an k-ring neighborhood, as described in (Meng et al. (2014); Zhang et al. (2023)). Finally, we applied the watershed clustering algorithm, as detailed in (Meng et al. (2014); Rettmann et al. (2002); Yang and Kruggel (2008); Zhang et al. (2023)) to the smoothed count map to automatically generate group-wise peak clusters for each species. Notably, the selection of parameters for anisotropic smoothing and watershed clustering algorithm were based on the previous work (Zhang et al. (2022, 2023)). Parameters of these three steps (individual peak extraction, anisotropic smoothing and watershed clustering algorithm) on two species are reported in the supplementary materials. In total, 192 (LH: 96, RH: 96) and 85 (LH: 42, RH: 43) peak clusters were detected on Humans and macaques, respectively (Figure.1(a)).

Cross-species Registration

To elucidate the inter-species relationship of group gyral peaks between humans and macaques, a functional joint alignment technique (Xu et al. (2020)) was employed to project macaque peak clusters onto the human cortical surface. They first constructed a joint similarity matrix by concatenating within- and cross-species similarities of connectivity patterns. Next, the diffusion embedding algorithm applied on the similarity matrix. Finally, we use gradients as surface features and align the cortical surfaces of humans and macaques using Multimodal Surface Matching (MSM) (Robin-son et al. (2014)). This technique builds upon recent advances in high-dimensional common space representations of functional organization and offers a transformational framework between human and macaque cortices.

Definition of Shared and Unique Peak Clusters

After the cross-species registration mentioned above, the group-wise gyral peak clusters of the two species were placed on the same template surface. The determination of peak clusters that are shared between species involves two criteria: 1) the Dice of clusters>0; and 2) the geodesic distance between the centers of the two clusters is less than 7 mm. If a pair of clusters satisfies either one of these two criteria, they can be identified as peak clusters that are shared between species. The difference set between all peaks of the two species and the shared peaks is the set of unique peaks for each species.

Anatomical Features of Gyral Peaks

We analyzed the anatomical characteristics of shared and unique gyral peaks, including sulc, curvature (the amount of bending at a point on a convoluted surface), thickness (the distance from the point on the pial surface to the nearest point on the white surface (Fischl and Dale (2000)) myelin, and local surface area (We calculate that the average area of all triangles in the neighborhood of vertex i is S_i. The local surface area of vertex i is the mean neighborhood area S_i divided by the mean of all vertices in the whole brain ).

Functional and Structural connectivity

We parcellated the white matter surface (excluded the regions between two hemispheres) into 1400 patches for human and 1700 patches for macaque (He et al. (2022)) due to the number of vertices on the surface. We constructed a structural connective graph G_s = {V, E_s, A_s} and a functional connective graph G_f = {V, E_f, A_f } for each subject. Graph nodes v_s and v_f were defined as cortical patches of the same area. For human individual structural connectivity matrix represents the fiber count connecting the two nodes. For human and macaque individual functional connectivity matrix, we calculate the Pearson correlation coefficient (PCC) between the average time-series between two nodes and , followed by Fisher’s z-transformation. Due to the vertex-to-vertex correspondences across individual surfaces of each species, the patches (or nodes) had cross-subject correspondences as well. On this basis, we averaged the structural and functional connectivity matrix of each subject to obtain a group-average structural and functional connectivity matrix and . Then, for each row in the group-average functional connectivity matrix, the values of the top 10% of connections were retained, whereas all others were zeroed. On this group-average graph, we computed nodal graph metrics, including degree, strength, clustering coefficient, betweeness, and efficiency via Brain Connectome Toolkit (https://sites.google.com/site/bctnet/). The definitions of these network properties are detailed in the supplementary material.

Feature selection of genes

Since we divided human gyral peaks into peaks shared with macaque and peaks unique to human, we aimed to investigate the genes that are significantly different expressed between two types of gyral peak. The preprocessed AHBA gene data is in the form of region×gene and the region above referred to the parcellation of a certain atlas, such as Aparc, Schaefer100, Schaefer500, Schaefer1000, etc. We finally selected Schaefer500 atlas for this study because high resolution may result in some areas with no gene data (more details refer to Arnatkevičiūtė et al. (2019)), while low resolution may result in multiple categories of clusters being located in the same region. Therefore, Schaefer500 was chosen as the most suitable atlas for our analysis. We first labeled all regions of Schaefer500 as shared, unique, or other based on the positions of group-wise gyral peaks. Then, Lasso (a linear regression method that uses L1 regularization for gene selection) was applied on this labeled gene data. The cost function of Lasso regression is as follows: . An important parameter of Lasso is lambda, which affects the sparsity of feature selection. We used 10-fold cross-validation to select the optimal lambda. By considering the maximization of accuracy (acc) and minimization of mean squared error (MSE) simultaneously, the lambda value was ultimately determined to be 0.027 (Figure. 5 (b)). The accuracy of training set was 0.84, and the MSE was 0.64; The accuracy of test set was 0.75, and the MSE was 1.00.

Data availability

All human data analyzed in this manuscript were obtained from the open-access HCP adult sample (https://www.humanconnectome.org/). Macaque data came from PRIME-DE (http://fcon_1000.projects.nitrc.org/indi/indiPRIME.html). Fiber tracking based on MRtrix3 (https://www.mrtrix.org).

Acknowledgements

We would like to thank the various contributors to the open access databases that our data was downloaded from. HCP data were provided by the Human Connectome Project, WU-Minn Consortium (Principal Investigators: David Van Essen and Kamil Ugurbil; 1U54MH091657) funded by the 16 NIH Institutes and Centers that support the NIH Blueprint for Neuroscience Research; and by the McDonnell Center for Systems Neuroscience at Washington University. Macaque data were provided by the PRIME-DE. Primary support for the work by Michael P. Milham and the INDI team was provided by gifts from Joseph P. Healy to the Child Mind Institute, as well as by the BRAIN Initiative (R01MH111439). MPM is a Randolph Cowen and Phyllis Green Scholar. Primary support for the work by Charles Schroeder is provided by the BRAIN Initiative (R01MH111439) and the Sylvio O. Conte Center “Neurobiology and Dynamics of Active Sensing” (P50MH109429). Primary support for the work by Daniel Margulies is provided by the Max Planck Society.

Additional information

Funding

Appendix

Parameter selection

Table A1. presents all the parameters used in the three algorithms for detecting individual and group peaks in the two species, along with their corresponding meanings. We used x-order ring neighbor to detect peaks on individual surfaces and chose k-order ring plus n-iterations for anisotropic smooth algorithm for the count map. Parameters for watershed clustering algorithm are related to the value of the count map. fg and bg respectively determine the minimum and maximum count of the segmented area. A smaller value of parameter merge results in more clusters. All parameters were determined based on previous studies (Zhang et al. (2022, 2023)).

Table A1.

The location of the Human and Macaque Peak Clusters

Table A2. and Table A3. displays the locations of all peaks in the human brain. The region names are derived from the aparc2009 parcellation in human and BA05 parcellation in macaque.

Shared/Unique Peak cluster centers’ location

The main text presents the counts of peak cluster centers occurrence in different regions (Cole-Anticevic). In order to eliminate the influence of different brain area sizes on the count of cluster, we normalized the count by the regional surface area and reported in Figure A1. The results were similar to those on the original counts. Our findings indicated that most of the shared peaks are located in low-order sensory and motor areas, while most of the unique peaks are located in higher-order cortices. The regions with the highest density of shared peak cluster centers are V1, Auditory (Aud), and Ventral-Multimodal (VMN), while the region with the highest density of unique peak cluster centers is Language (Lan), DMN, Orbito-Affective (OAN), and FPN.

Confidence of Shared Peaks

FigureA2 (a) illustrates the locations of all shared peaks. There are two definitions of shared peaks: 1) the Dice of clusters>0; and 2) the geodesic distance between the centers of the two clusters is less than 7mm. The credibility of shared peak clusters defined by the coincidence rate of clusters is measured using the overlap ratio. The higher the overlap rate of clusters, the higher the confidence of shared clusters between species FigureA2 (b). The credibility of shared peak clusters defined by the distance of cluster centers is measured using the ranking of center distances. The distance between the two closest clusters in the whole brain is set as 100 points, and the distances of other clusters are proportionally reduced accordingly FigureA2 (c).

Table A2.

Table A3.

Network properties for graph analysis

Network Properties

Degree

Degree is the most important description of the statistical characteristics of node connection. Degree K_i is defined as the number of edges directly connected to a node i. The greater the degree of the node, the more connections the node has, and the more important the status of the node in the network. G is an undirected weighted network in our work, the degree of node i is defined as:

Strength

For a N-node weighted network G which weight matrix is W, the strength of node i is defined as:

Figure A1.

Cluster Coefficient

The clustering coefficient of a vertex i is the probability that the neighbours of this vertex (all other vertices to which it is connected by an edge) are also connected to each other. The clustering coefficient of a vertex ranges between 0 and 1.

Betweenness Centrality

Betweenness Centrality indicates the times of a node appears on all shortest paths in a graph. σ_st is the number of shortest paths from node s to node t, and σ_st (v_i) is the number of times those paths pass through (v_i).

Efficiency

The average communication efficiency of the network G is then defined as the average over the pairwise efficiencies:

Spatial Relationship Between Peaks and Functional Regions

Figure A2.

Table A4.