Genetic neurocognitive profile of autism unveiled with gene transcription

Genetic neurocognitive profile of autism unveiled with gene transcription | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article Genetic neurocognitive profile of autism unveiled with gene transcription Sheng Hu, Yingxing Zhang, Fangfang Li, Hongli Wu, Wei Du, Jianhua Shu, and 4 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-5425486/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract The neurobiological basis for elaborating phenotypic heterogeneity within autism spectrum disorder (ASD) remains unknown. Applying the principal component analysis to the Neurosynth database, we established neurocognitive profiles to characterize the phenotypic heterogeneity of ASD, revealing a cortical hierarchical axis that separates the temporal cortex from other networks. By integrating neurocognitive profiles with transcriptomic data, we found that gene sets shaping the patterns of neurocognitive profiles are enriched in ASD-related biological processes and ASD pathogenic risk. Using a data-driven approach, we identified a topographic network for ASD, comprising the temporal, frontal, somatosensory, and visual cortices, with its transcriptomic signatures differentiating between regions over neurodevelopment. Additionally, functional reorganization in ASD within the topographic network has occurred with the temporal cortex as the central node. Collectively, our results reveal spatially covarying transcriptomic and neurocognitive profiles, emphasizing the influence of functional reorganization and its underlying genetic mechanism on phenotypic heterogeneity in ASD. Health sciences/Diseases/Psychiatric disorders/Autism spectrum disorders Biological sciences/Neuroscience/Genetics of the nervous system Figures Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Main Autism spectrum disorder (ASD) is a highly heritable and heterogeneous neurodevelopmental disorder characterized by impairments in social communication and interaction, anomalies in sensory responses, repetitive behaviors, and restricted interests 1,2 . The heterogeneity is also complicated by the manifestation of comorbidities, highlighting the mental burden in ASD 3 . Distinct conditions within ASD may exhibit varying pathophysiological processes 4 , as identified by genetic and magnetic resonance imaging (MRI) analyses 5,6 . The co-occurring ASD conditions complicate the disease course and lead to functional deterioration and poorer clinical outcomes 7 . Thus, clarifying the neural mechanisms of comorbidity within ASD benefits from discovering new therapeutic approaches for different types of ASD. Functional MRI (fMRI) studies have uncovered that atypical activity in the visual cortex and salient network is involved in dysregulation with social cognition and language processing in ASD 8-10 , and that aberrant functional connectivity in the corticostriatal and frontoparietal network can characterize different types of ASD 6,11 . Cortex-wide gradient mapping techniques described a disruption in macroscale hierarchy affecting integration and segregation of unimodal and transmodal networks in ASD, especially involving sensory and higher-order default mode regions 12 . Moreover, the neuroimaging basis of phenotypic heterogeneity has been demonstrated in ASD, exhibiting that the subgroups defined by anatomical features can enhance the prediction of symptom severity in subtyping ASD 13 , that the abnormal brain connections distinguish individuals with ASD from healthy controls 14 and that atypical connectivity patterns underlie different forms of ASD 6 . However, no studies attempt to uncover the neurocognitive architecture of phenotypic heterogeneity for ASD. In other words, the neurobiological basis of comorbidities in ASD remains poorly understood. Modern advancements in neuroimaging, coupled with the promotion of global data-sharing initiatives, have fundamentally enhanced our ability to investigate neurocognitive processes and their underlying molecular mechanisms. Meta-analyses by synthesizing human neuroimaging data have informed how brain regions respond to a spectrum of cognitive, perceptual, and emotional tasks 15-17 . Moreover, advanced high-throughput transcriptomes have provided precise gene expression maps distributed over the whole brain, highly linking the spatial distribution of molecular dynamics with macroscale cortical organization of anatomical properties 18-20 . Therefore, integrating brain-mapping initiatives and high-throughput transcriptomes into neuroimaging analyses offers an unheard-of opportunity to unravel spatial associations between the brain’s gene expression profiles and the neurocognitive architecture of phenotypic heterogeneity for ASD. Here, we obtain probabilistic functional association maps linked to the phenotypic heterogeneity of ASD from Neurosynth 15 to construct the neurocognitive architecture. Then, we test whether the neurocognitive architecture is related to the properties of macroscale hierarchy. In parallel, we apply partial least-squares (PLS) analysis 21 to identify the spatial associations between the transcriptional data (Allen Human Brain Atlas (AHBA) 20 ) and the neurocognitive architecture. This procedure can generate the weighted gene expression maps for these neurocognitive processes. Gene enrichment analysis was further employed to investigate transcriptomic associations with known pathogenic variants linked to ASD and explore transcriptomic associations with other psychiatric risks 22 . Considering that ASD is a neurodevelopmental disorder, we evaluated spatiotemporal gene expression dynamics from prenatal to adulthood stages using a developmental genetic dataset of BrainSpan 23,24 . Finally, we identify the brain–behavior, and brain organization (see methods) for explaining the macroscale hierarchy of phenotypic heterogeneity in ASD using a large-scale resting-state fMRI dataset (Autism Brain Imaging Data Exchange (ABIDE) I 11 ). Results Neurocognitive profiles of ASD To establish the neurocognitive architecture, we acquired the functional association maps, which represent probabilistic measures of whether cognitive terms (such as restricted, social interactions, and anxiety) from Neurosynth are functionally related to specific brain regions 15 . The terms were chosen based on our literature reviews concerning heterogeneous phenotypes in ASD 1,2,5,7 ( Supplementary Table 1 ). The functional association maps were subsequently projected on the cerebral cortex using HCP_MMP1.0 with 360 parcels 25 ( Fig. 1a ). Then, we applied the principal component analysis (PCA) to determine the dominant component of spatial variation across these functional association maps, thereby defining the dominant components as neurocognitive profiles for ASD ( Supplementary Fig. 1 ). The principal neurocognitive profile exhibits higher PCA scores in the superior medial frontal cortex, middle frontal cortex, insular cortex, and parietal cortex while showing lower PCA scores in the inferior temporal cortex and inferior medial frontal cortex ( Fig. 1b ). Then, we assigned parcels of HCP_MMP1.0 to six networks, followed by projecting the PCA scores into the networks 25 . At the network level, the temporal cortex (TPC) shows a significantly lower PCA score than other networks ( Fig. 1c ), including the frontal cortex (FTC), auditory cortex (ADC), sensorimotor cortex (SMC), posterior cortex (POC) and visual cortex (VSC). The second neurocognitive profile demonstrates higher PCA scores in the FTC, TPC, and posterior cingulate cortex while exhibiting lower PCA scores in the SMC, POC, and VSC ( Supplementary Fig. 2 ). Moreover, we also employed the Laplacian eigenmaps to reproduce the neurocognitive profiles ( Fig. 1d ). The Laplacian eigen score is significantly correlated with the PCA score of the dominant component ( R = 0.8687, P Perm < 0.0001, Fig. 1d ). Neurocognitive profiles reflect cortical hierarchies Having identified neurocognitive profiles for phenotypic heterogeneity in ASD, we next investigated whether these topographic patterns relate to cortical hierarchical properties 26 . Here, we first calculated the spatial correlation between the principal neurocognitive profiles and the map of geodesic distance from the early visual cortex that defines a posterior-anterior gradient, exhibiting a strong negative correlation ( R = -0.2388, P Perm = 0.0019; Fig. 2 a ). We then averaged a measurement of the T1w/T2w ratio (a widely used proxy for intracortical myelin) from the left hemisphere cortex across 1113 subjects from the Human Connectome Project (HCP). The principal neurocognitive profile shows a significant positive spatial correlation with the T1w/T2w ratio map ( R = 0.3291, P Perm < 0.0001; Fig. 2b ). Finally, we applied the diffusion map embedding on a group-averaged functional connectivity (FC) matrix computed from the 1,084 HCP participants to generate the principal functional gradient. The principal neurocognitive profile has a strong negative spatial correlation with the principal functional gradient ( R = -0.3415, P Perm < 0.0001; Fig. 2c ). In addition, the second neurocognitive profile also shows strong spatial correlations with cortical hierarchical properties ( Supplementary Fig. 3 ). Transcriptomic correlates of neurocognitive profiles To further explore the transcriptomic signatures of the neurocognitive profiles, we mapped normative regional gene expression profiles for 10027 microarray probes in the AHBA, including gene expression data for 1290 samples from 6 healthy donors (N = 5 males, N = 1 female, aged 24~57 years) 20 , to the HCP_MMP1.0 atlas using a standard processing pipeline 27 . Next, we used the PLS regression analysis to test the associations of weighted gene expression with the spatial distribution of the principal neurocognitive profile. Two PLS components explained 17.7% (PLS-1: P Perm = 0.004) and 19% (PLS-2: P Perm < 0.001) of the covariance between the principal neurocognitive profile and AHBA gene expression ( Fig. 3a and Supplementary Fig. 4 ). Also, both PLS-1 ( R = 0.37, P < 0.0001) and PLS-2 ( R = 0.39, P < 0.0001) are spatially correlated with the principal neurocognitive profile ( Fig. 3b ). PLS-1 represents genes with high-weighted gene expression in the somatosensory and visual cortex but showing low expression in the medial frontal, insular, and temporal cortex ( Fig. 3a ). PLS-2 is characterized by high-weighted gene expression in the middle frontal cortex but shows low expression in the visual cortex ( Fig. 3a ). Gene enrichment analysis results exhibit that PLS-1 is mainly enriched in ADHD and autism ASD-linked metabolic pathways and SNP, head development, regulation of metal ion transport, and regulation of trans-synaptic signaling, indicating that the neurocognitive profile reflects the neurobiological basis of phenotypic heterogeneity in ASD 28 ; PSL-2 is primarily involved in response to growth factor, neuron projection development, cell junction organization and extracellular matrix organization, foreboding the neurocognitive profile may be linked to neural connections formation during neuron development 29 ( Fig. 3c ). Specific analysis reveals that PLS-1 is enriched in autism pathogenic risk and other neuropsychiatric risks, such as depression and epilepsy ( Fig. 3d ), indicating that the weighted gene expression shapes the neurocognitive processes related to phenotypic heterogeneity of ASD. Furthermore, the pattern of PLS-1 covarying with specific terms, such as restricted, repetition, attention task, sensory, anxiety, and depression, is strongly correlated with heterogenous phenotypes or clinical symptoms comorbidity in ASD. The PLS regression analysis was also performed to test the associations of weighted gene expression with the spatial distribution of the second neurocognitive profile, displaying weighted gene expression maps of two PLS components ( Supplementary Fig. 5a ). PLS-1 is enriched in the regulation of neuron projection development, regulation of monatomic ion transport, and head development, while PLS-2 is mainly involved in the regulation of cell junction assembly and inorganic ion transmembrane transport ( Supplementary Fig. 5b ). Importantly, both PLS-1 and PLS-2 are enriched in autism pathogenic risk and other neuropsychiatric risk such as depression ( Supplementary Fig. 5c ). Furthermore, the pattern of PLS is also covarying with terms related to heterogenous phenotypes or clinical symptoms of comorbidity in ASD ( Supplementary Fig. 5d ). Identify topographic networks of phenotypic heterogeneity in ASD Having identified the gene expression signatures for the neurocognitive profiles, we next test whether the weighted gene expressions distributed in the cerebral cortex contribute to these neurocognitive processes. Accordingly, we set a range of thresholds (from 0 to |0.1|) with a step of 0.01 for the PLS score to identify the topographic networks of phenotypic heterogeneity in ASD. Specifically, we began by applying a threshold to the PLS scores, which allowed us to generate a topographic network. We next selected functional features that are regional homogeneity (ReHo) calculated from a large-scare resting-state fMRI dataset of ASD (see methods) based on the topographic network. Finally, these features feed into the support vector machine (SVM) classification model for distinguishing ASD patients from healthy controls. The topographic network can reflect neurocognitive processes of the phenotypic heterogeneity in ASD if the model can accurately classify ASD patients from healthy controls. We found that the PLS-1 score of the principal neurocognitive profile threshed at 0.07 can better explain the neurocognitive processes with a significant classifying accuracy of 63.67% ( P = 0.002, 1000 times permutation; Fig. 4a and b ). The topographic network features regions with higher weighted gene expression in the somatosensory and visual cortices, while regions with lower weighted gene expression are found in the medial frontal and temporal cortices ( Fig. 4b ). Furthermore, we showed the regions with higher gene expression are significantly related to social behaviors ( R = -0.2309, P = 0.0407) and age ( R = -0.3252, P = 0.0035), and regions with lower gene expression are strongly related to stereotyped behaviors and restricted interests ( R = 0.2262, P = 0.045, Fig. 4c ), indicating the topographic network involving in ASD’s neurocognitive processes. We also found that the PLS-2 score of the second neurocognitive profile threshed at 0.01 ( Supplementary Fig. 6 ), which gets a significant classifying accuracy of 59.21% ( P = 0.028, 1000 times permutation). However, no clinical behaviors of ASD are correlated with the topographic network, indicating that the topographic network cannot well explain the neurocognitive processes for phenotypic heterogeneity of ASD. Transcriptomic signatures in neurocognitive profiles over development Given the continued development of neurocognitive processes across the lifespan, we sought to investigate the spatiotemporal trajectories of transcriptomic signatures through human development. Here, we analyzed the spatiotemporal trajectory of the identified topographic network, including the frontal, temporal, somatosensory, and visual cortex, over brain development. We employed the BrainSpan 23,24 in the following analyses, a gene expression database of brain tissue across development covering the period from 8 post-conception weeks to 40 years across 16 cortical regions. The ages were binned into nine-time windows (W1 to W9), encompassing the embryonic period, fetal development, infancy, childhood, adolescence, and adulthood. The spatiotemporal transcriptomic signatures were defined as the projection of gene expression of BrainSpan onto the PLS analysis-defined gene expression (such as PLS-1 and PLS-2), generating an estimated gene expression per region and time window (see methods). For PLS-1 of the principal neurocognitive profile, gene expression decreased during embryonic and early to middle fetal stages (W1-W3), followed by an increase throughout the remaining life stages. In addition, the transcriptomic signature has gradually differentiated between brain regions throughout neurodevelopment ( Fig. 5a ). In contrast, the spatiotemporal trajectory of PLS-2 ( Fig. 5b ) mirrors that of PLS-1 from embryonic to late childhood stages (W1-W7) but shows a decline from late adolescence to adulthood (W8-W9). Interestingly, the estimated gene expression of the second neurocognitive profile has an opposite spatiotemporal trajectory to that of the principal neurocognitive profile ( Supplementary Fig. 7 ). Topographic network organization in ASD Having identified the transcriptomic signatures of the topographic network for the neurocognitive processes associated with the phenotypic heterogeneity of ASD, we next hypothesized that the functional organization occurs in the topographic network in individuals with ASD. To solve this issue, we tested the network similarity using a data-driven approach in which we used the blood-oxygen-level-dependent (BOLD) time series from each parcel to train an SVM model for achieving six network classification tasks ( Fig. 6a ). This was motivated by the observation that brain regions with similar features of the BOLD time series are more likely to support similar functions. For instance, the BOLD time series within the default mode network has a similar feature, such as the longest temporal autocorrelation delay, while those in the early visual network show the shortest temporal autocorrelation delay 30 . Therefore, the function of a given network can shift to being closer to or farther from that of another network if functional reorganization occurs between the two networks in the brains of individuals with ASD. This is proved by our classification analysis, which shows that the classification accuracy of the networks varies between ASD patients and healthy controls. We found that the classification accuracy of a given network has no difference between ASD patients and healthy controls (frontal cortex: [T = 1.29, P = 0.196], Fig. 6b ; temporal cortex: [T = 1.64, P = 0.104], Fig. 6c ; auditory cortex: [T = 1.316, P = 0.189], Fig. 6d ; somatosensory cortex: [T = -0.203, P = 0.839], Fig. 6 e ; posterior cortex: [T = -0.233, P = 0.816], Fig. 6f ; visual cortex: [T = -1.39, P = 0.167], Fig. 6g ). We further analyzed whether the given networks are more likely to classify as other networks in individuals with ASD and healthy controls. Importantly, we exhibited that the TPC is more likely to classify as FTC (T = -2.059, P = 0.0409) and POC (T = -2.325, P = 0.0212) in individuals with healthy controls ( Fig. 6c ), indicating that functional similarities between the TPC and both the FTC and POC are closer in healthy controls than that in ASD. The TPC is more likely to classify as SMC (T = 2.064, P = 0.0404) in individuals with ASD ( Fig. 6c ), indicating that the functional similarity between TPC and SMC is closer in ASD than that in healthy controls. These findings suggest that there is dysregulation in the functional interactions between the TPC and both the FTC and POC in individuals with ASD, while interactions between the TPC and SMC are enhanced in individuals with ASD. Discussion Here, we identified a neurocognitive profile for phenotypic heterogeneity of ASD and further analyzed its spatial association with gene expression. Collectively, this pattern delineates a strong spatial distribution difference among networks, especially separating TPC from other networks. The neurocognitive profiles follow cortical hierarchical organization and are significantly related to multiple cortical properties. By integrating a large-scale fMRI dataset with a weighted gene expression map, we identified a topographic network associated with phenotypic heterogeneity in ASD. Further, we uncovered how the transcriptomic signatures of this network evolve throughout neurodevelopment. Finally, we found that functional reorganization has occurred within the topographic network with the TPC as the central node in ASD. Our results directly bridge macroscale neurocognitive processes with microscale gene expression, highlighting the influence that functional reorganization underlying neurobiological mechanisms have on phenotypic heterogeneity in ASD. The present study was motivated by previous reports that link gene expression to structural and functional architectures in both healthy participants and ASD brains. Gene expression profiles have strong associations with cortical folding 31 , functional hierarchy 32 , and T1w/T1w ratio 33 in healthy human brains and have been linked to multiple microstructural features in brains with ASD 34,35 , potentially reflecting a hierarchical axis of microstructural and macroscale functional properties. In particular, disruptions in macroscale hierarchy affect the integration and segregation of unimodal and transmodal networks in ASD 12 , and further, these atypical functional connectomes can be explained by regional differences in the expression of distinct ASD-related gene sets 6 . While these reports exhibit a link between transcriptomic profiles and multiple functional and structural features, the neurocognitive processes reflected phenotypic heterogeneity of ASD of such trends are not well understood. Our results fill in this huge gap, showing statistically spatial associations of neurocognitive processes with cortical hierarchical properties and transcription. Collectively, spatial alignment of transcriptomic profiles and neurocognitive processes may appear to show a dominant underlying hierarchical axis, shaping phenotypic heterogeneity in ASD. Of note, the neurocognitive profile reveals a marked divergence in network distribution, with the TPC notably separated from other networks. In correspondence with our result, a previous study reports the inter-individual variability increased in ASD relative to healthy controls in default model, somatomotor, and attention networks but showed reduced variability in the lateral temporal cortex, highlighting the atypical pattern reflecting inter-individual heterogeneity in ASD 35 . The TPC may function as a node hub, forming dysfunctional connections with brain regions such as the prefrontal cortex, precuneus, and cuneus, thereby influencing emotion-related decision-making in ASD 36 . This evidence is supported by our data-driven results, which demonstrate the TPC as a central hub facilitating functional reorganization within the FTC, SMC, and POC networks in ASD. Moreover, gray matter volume and cortical thickness decreased in the TPC may bring about its dysregulation with other brain areas 37-39 , ultimately resulting in failed language development and atypical object perception in ASD 40,41 . Thus, these findings suggest that the TPC may act as a central node to be involved in complex dysfunction in ASD. In the present study, we found that the neurocognitive profile has spatial correspondence with multiple cortical properties, demonstrating a cortical hierarchical axis that separates TPC from other brain networks. Patterns of gene expression spatially shape the cortical hierarchical properties, including microstructural distribution and functional organization 32,33,42 . In the AHBA dataset, we found that the neurocognitive profile can be explained by the transcription, showing that the weighted gene expressions (both PLS-1 and PLS-2 scores) are spatially correlated with the neurocognitive profile. Gene enrichment analysis revealed that the top 10 th gene sets that shape the neurocognitive profile are enriched in ASD-related biological processes (PLS-1) and molecular processes in the formation of neural connections during neurodevelopment (PLS-2). In particular, the weighted gene expression of PLS-1 is enriched in ASD pathogenic risk genes, unraveling that ASD risk gene sets shape the neurocognitive profile in the phenotypic heterogeneity in ASD. We mapped the whole-genome transcription patterns to the spectrum of neurocognitive function across multiple brain areas, showing that the neurocognitive terms related to ASD behaviors, such as restricted, attention, and reaction 1 , contribute to the weighted gene expression of PLS-1. At the same time, the association between gene expression and behaviors has been approached from another study, delineating a ventromedial–dorsolateral axis and separating gene sets related to perceptual versus affective function 43 . Thus, these findings uncovered a cortical hierarchical axis, separating TPC from other brain networks, which is spatial covarying with transcription to characterize a gene-neurocognition signature of phenotypic heterogeneity in ASD. Indeed, the gene sets in brain regions with highly weighted gene expression shape the neurocognitive profile, but whether these brain regions can characterize phenotypic heterogeneity in ASD. Using a data-driven approach combined with a large-scale fMRI dataset, we identified a topographic network characterized by higher weighted gene expression in the somatosensory and visual cortices and lower weighted gene expression in the medial frontal and temporal cortices. All of these brain regions participate in brain pathological processes of ASD 44,45 . Moreover, brain regions within the topographic network can classify ASD from healthy controls and relate to ASD behaviors such as social cognition, stereotyped behaviors, and restricted interests, suggesting that the identified topographic network is involved in phenotypic heterogeneity in ASD. In line with our results, a review of the literature has reported the predictive models that phenotypic features within the brain regions that are spatially similar to the topographic network, differing across subtypes, may help triage individuals for better care management in ASD 46 . Therefore, how transcriptomic–neurocognitive links vary across subtypes is a key question for future studies. Gene expression profiles are involved in shaping cortical reorganization throughout neurodevelopment, comprising folding 31 and formation of corticocortical connection 47 . Using the BrainSpan dataset, we found that gene-neurocognition signature within the topographic network exhibits a spatiotemporal trajectory over neurodevelopment, gradually becoming most remarkable and differentiation between brain regions in adulthood ( Fig. 5a ). This suggests that continued differentiation of neurocognitive processes during maturation gradually forms the phenotypic heterogeneity in ASD. Furthermore, we observed that functional reorganization has occurred within the topographic network with the TPC as the central node, suggesting that functional dysregulation within this network contributes to the phenotypic heterogeneity in ASD. Consequently, these findings suggest that gene expression shapes functional reorganization and ultimately forms the neurocognitive profile reflecting phenotypic heterogeneity in ASD. Several methodological factors may limit the scope of our findings. First, the main analysis is based on small samples of postmortem brains, potentially limiting the generalizability of the results. Thus, more comprehensive microarray gene expression datasets are necessary for future studies to ensure a more reliable understanding of the observed phenomena. Further, spatiotemporal trajectories of gene-neurocognition signature were analyzed based on BrainSpan data, which are limited by small cortical genetic samples. Finally, subcortical topographic changes were well documented in previous studies 48,49 . How gene-neurocognition signature shapes subcortical profiles of ASD should be investigated in future work. In summary, we demonstrated a neurocognitive profile related to phenotypic heterogeneity in ASD, showing a cortical hierarchical axis that separates the TPC from other networks. Alignment the patterns of gene expression with neurocognitive profile, we showed that gene-neurocognition signature shapes the neurocognitive processes and the ASD pathogenic risk, ultimately manifesting as a topographic network that gradually differentiates during neurodevelopment. Finally, we observed that functional reorganization has occurred within the topographic network with the TPC as the central node. Collectively, these results highlight that genomic-neurocognition shapes the abnormal functional reorganization to form a neurocognitive profile related to phenotypic heterogeneity in ASD. Methods Neurocognitive profiles The meta-analytical functional association maps, of which probabilistic measures reflect the activation of specifical brain function, were obtained from Neurosynth 15 (https://neurosynth.org/analyses/terms/), a tool that synthesizes fMRI results from more than 15000 studies. Although more than 1000 terms are reported in Neurosynth, we focus primarily on neurocognitive processes related to phenotypic heterogeneity of ASD based on our comprehensive literature review 1,2,5,7 and therefore limit the terms of interests cognitive and behavioral terms. We used 40 terms, ranging from specific behaviors (reaction, repetition, and restricted) to specific cognitive processes (attention and cognitive control) and emotional states (anxiety and negative emotions). All functional association maps were parcellated into 360 bilateral hemisphere cortical regions using HCP_MMP1.0 25 . Then, we applied the PCA 50 to determine the dominant component of spatial variation across these functional association maps, thereby defining the dominant components as neurocognitive profiles for ASD. Moreover, we also employed the Laplacian eigenmaps 51 to reproduce the neurocognitive profiles for validation analysis. HCP dataset This study included data from 1206 healthy young adults (age range, 22 to 37 years) from the publicly released HCP Young Adult dataset 52 . The subjects who met the following criteria, such as a history of psychiatric disorder, substance abuse, cardiovascular disease, or any other severe health conditions, were excluded from the present study. Recruitment, data acquisition, and written consent are described in detail at https://www.humanconnectome.org/study/hcp-young-adult and in other publications 52 . The subjects who did not include high-resolution cortical T1w/T2w maps (0.7 mm 3 isotropic) were removed from the current study. Finally, 1113 subjects who contained T1w/T2w maps remained in the following analyses. Subsequently, T1w/T2w maps were averaged across subjects, and the averaged T1w/T2w map was projected on the left hemisphere with 180 areas using HCP multimodal parcellation. The minimally processed resting-state fMRI data of 1084 participants (age range, 22 to 37 years) from HCP were included for functional gradient analysis. The images were further processed to regress out nuisance variables from head movements and physiological noise using Analysis of Functional Neuroimages software (AFNI, https://afni.nimh.nih.gov/) and then nonlinearly registered to the Montreal MNI_ICBM152 standard space. The time series of fMRI data were averaged within each parcel using the HCP_MMP1.0 template, and Pearson correlation analysis was performed between each parcel to construct an FC matrix for each participant. Finally, the FC matrices were averaged across all participants to generate the averaged FC matrix. Resting-state fMRI dataset of ASD and data preprocessing Original fMRI and demographic data were obtained from the open-access data set, Autism Brain Image Data Exchange I (https://fcon_1000.projects.nitrc.org/indi/abide/abide_I.html). A total of 79 ASD patients (age range, 7 to 39 years) and 105 healthy controls (age range, 7 to 31 years) from NYU Langone Medical Center were included in the present study. The clinical assessment scores for the Autism Diagnostic Observation Schedule (ADOS), Social Communication Questionnaire (SCQ), and Social Responsiveness Scale (SRS) were collected. The detailed MRI scanning parameters and demographic information are available at https://fcon_1000.projects.nitrc.org/indi/abide/abide_I.html. Preprocessing of the functional images involved the following steps: (1) discarding the first 10 time points of functional images to account for magnetic stabilization; (2) slice timing correction to compensate for temporal shifts of different slices; (3) within-subject fMRI image realignment to estimate and spatially correct for head motions of different volumes; (4) rigid-body T1 image registration to the functional mean image, then the T1 image normalized to the Montreal MNI_ICBM152 standard space to generate a transformation matrix, and finally aligned the fMRI image to the MNI_ICBM152 standard space using the transformation matrix; (5) functional image resampling to 3×3×3 mm 3 voxel size followed by regressing out confounding signals, including linear trends, white matter (WM), cerebrospinal fluid (CSF), and head motion parameters. All data preprocessing was performed on Analysis of Functional Neuroimages software (AFNI, https://afni.nimh.nih.gov/). Functional gradient analysis Functional gradient analysis was performed using the Brainspace toolbox (https://github.com/MICA-MNI/BrainSpace). The averaged FC matrix from HCP was threshed to retain the top 10% connections of each node, and the cosine similarity between each pair of nodes was computed. Furthermore, the similarity matrix was scaled into a normalized angle matrix to avoid negative values 22 . The diffusion map embedding approach was finally applied to identify gradient components that explain most functional connectome variances. Following the previous recommendation, we set the manifold learning parameter α = 0.5 in the diffusion process 53 . Finally, the gradient map was projected to the 360 bilateral hemisphere cortical regions using HCP_MMP1.0. Geodesic distance of cortex We calculated the Euclidean distance between each cortical parcel from the HCP_MMP1.0 atlas and the early visual cortex. The coordinates of each cortical parcel were acquired from the website at https://neuroimaging-core-docs.readthedocs.io/en/latest/pages/atlases.html. Then, the geodesic distance of each parcel was acquired to form a geodesic distance map. Gene expression data and preprocessing We utilized left hemisphere microarray-based gene expression data from the Allen Human Brain Atlas 19,20 (AHBA) (http://human.brain-map.org, RRID: SCR_007416). The microarray gene expression data were obtained from six donors (mean age: 42.5 years, five males and one female), including two complete brains and four left hemispheres. None of the donors had a known history of neuropsychiatric or neurological conditions. Exclusion criteria included brain injury or disease, epilepsy, drug/alcohol dependency, > 1 hour on the ventilator, positive for infectious disease, prion disease, chronic renal failure, cancer deaths, brain cancer, and time since death > 30 hours. The gene expression of each sample from all donors was quantified across 58692 probes, resulting in 20,737 gene expression levels per sample. The tissue samples were also spatially registered to the Montreal Neurological Institute (MNI) coordinate space, and the locations of each sample were recorded with MNI coordinates. The gene expression data of the brain samples were preprocessed using an AHBA processing pipeline 27 (https://github.com/BMHLab/AHBAprocessing) with the recommended default setting. Specifically, probe-to-gene reannotation was performed using the latest sequencing database, and probes with values that did not exceed the background noise were filtered. When multiple probes for a gene were available, the probe with the highest correlation with the RNA-seq expression data was selected. Next, each sample was assigned to its nearest cortical parcel of HCP_MMP1.0 parcellation (left hemisphere) with 180 parcels. These procedures yielded 1290 brain tissue samples covering 176 regions within the left cortex, with each sample containing the expression data of 10,027 genes. PLS analysis PLS regression analysis 21 was applied to investigate the spatial relationship between gene expression and neurocognitive profiles. PLS analysis, an unsupervised multivariate statistical technique, decomposes relationships between predictor variables and response variables into orthogonal sets of latent variables with maximum covariance. These latent variables are a linear combination of orthogonal variables. We first aligned the gene expression data (10027 genes) and neurocognitive profile map to the HCP_MMP1.0 atlas 25 . The gene expression data and neurocognitive profile map were used as the predictor variables and the response variables, respectively 54 . The rows (brain regions) of the neurocognitive profile matrix were randomly selected and replaced 10000 times 43 . PLS analysis was reperformed using a new bootstrapped neurocognitive profile matrix to generate a null distribution of the ratio of variance explained, ensuring that the PLS component was significantly greater than expected by chance. For each significant component, a bootstrapping method was employed to evaluate the estimation error associated with the weight of each gene. The weight of each gene was then divided by the estimated error to derive the adjusted weight 54 . Genes were ranked based on their corrected weights, reflecting their contributions to the PLS regression components. Gene enrichment analysis The top 10th percentile of 10027 ranked genes was applied to a Metascape analysis tool (https://metascape.org/gp/index.html#/main/step1) to uncover biological processes enriched in the list of genes 55 . The top 10 percent of PLS genes were input to the Metascape website, and the obtained enrichment pathways were thresholded for significance at 5%, corrected by the false discovery rate (FDR) approach. Specificity analysis Specificity analysis was used to assess whether known psychiatric risk genes, such as bipolar disorder, depression, autism, schizophrenia, and intellectual disability genes, were enriched in the PLS components 56 . The disorder-related risk genes provided by the AHBA (https://help.brain-map.org/display/humanbrain/Documentation). We calculated the enrichment ratio (ER) for each PLS component. The ER is defined as the difference between the mean bootstrap weight of the candidate gene and the mean bootstrap weight of the same number of randomly permuted genes, which was further divided by the standard deviation weight of the permuted genes. Significance was determined by the percentile of the bootstrap weight of the candidate genes relative to the bootstrap weights of randomly selected genes from 10,000 permutations 56 . A positive/negative ER of a given condition indicates that the risk genes are expressed to a higher/lower degree relative to the baseline expression level. ReHo analysis The ReHo was calculated using the DPABI toolbox 57 (https://rfmri.org/DPABI). The ReHo analysis was performed on the preprocessed images of ASD. Individual ReHo map was generated by calculating the Kendall coefficient concordance to measure the similarity of the BOLD time series of a given voxel and its 26 nearest neighbors in a voxel-wise way 58 . Then, a z-transformation was applied to the individual ReHo map to generate a normally distributed ReHo map. Finally, the normalized individual ReHo map was projected to the 360 bilateral hemisphere cortical regions using HCP_MMP1.0. SVM classification analysis The SVM classification analysis used a MATLAB toolbox (https://www.csie.ntu.edu.tw/~cjlin/libsvm/). Here, we opted for a linear function kernel for the SVM model and employed a grid search function to determine the optimal cost parameter (C) for the SVM hyperparameters. The performance of the SVM model was evaluated using a 10-fold cross-validation strategy, where the data were divided into ten partitions. The SVM model was trained using data from 9 partitions and tested on the remaining partition, and this process was repeated 10 times. The final performance assessment was determined by combining the results from these ten models, and accuracy was calculated based on the correct labeling assessments. Identify network topographic pattern Here, we set a range of thresholds (from 0 to |0.1|) with a step of 0.01 for the PLS score to identify the networks of phenotypic heterogeneity in ASD. Specifically, we began by applying a threshold to the PLS scores, which allowed us to generate a topographic network. We next selected ReHo within the topographic network as a training feature. Finally, these features feed into the SVM classification model for classifying ASD patients from healthy controls. Additionally, we randomly selected training features with the same dimensions as the topographic network to retrain the SVM model. This procedure was repeated 1000 times to generate a null distribution of accuracy. The observed accuracy was then compared with the null distribution to ascertain whether the topographic network was influenced by random effects and/or confounding factors. The topographic network can reflect neurocognitive processes of the phenotypic heterogeneity in ASD if the model can accurately distinguish ASD patients from healthy controls. Spatiotemporal gene expression over development We used the PsychENCODE BrainSpan dataset 23,24 to calculate the spatiotemporal trajectories for each PLS component obtained in the PLS regression analysis. BrainSpan is a gene expression database of brain tissue across development (https://www.brainspan.org/static/download.html) covering the period from 8 postconception weeks to 40 years of age. The ages were binned into nine-time windows (W1 to W9), encompassing the embryonic period, fetal development, infancy, childhood, adolescence, and adulthood. Detailed information is available in a previous study 23 . In the current analysis, we used the gene expression data from 16 cortical regions across all nine-time windows to calculate the spatiotemporal profile. This profile is defined as the regional average of each BrainSpan gene expression level, weighted by its PLS analysis-defined weights 59 . Network classification analysis To identify the topographic network organization in individuals with ASD, we used the BOLD time series from each parcel to train an SVM model for achieving six network classification tasks. Specifically, we began by averaging the BOLD time series within each parcel associated with a given network. The resulting averaged time series for each parcel were then selected as features for training the SVM classification model ( Fig. 6a ). We used a one-vs-one approach to classify the given network from other networks. Then, we constructed the confusion matrix to summarize the prediction results for each network. In this matrix, each row represents an instance of the actual class (i.e., an actual network), and each column represents an instance of the predicted class 60 (i.e., the predicted network). The diagonal elements indicate the number of points where the predicted label matches the true label, while off-diagonal elements represent mislabeled instances by the classifier ( Fig. 6a ). Importantly, the given network is more likely classified as a specific network (i.e., the higher mislabeled instances in off-diagonal elements), demonstrating that the functional similarity of the given network shows closer with the specific network. Statistical analyses The differences in PCA score distribution between the TPC and other networks, such as FTC, ADC, SMC, POC, and VSC, were analyzed using a two-sample t-test. The spatial correlation between two spatial patterns using Pearson correlation analysis and the statistical significance of spatial correlation was calculated using a permutation test with 10000 times. The contribution of each neurocognitive term to weighted gene expression was calculated as the Pearson correlation between the term’s functional association map and PLS-1 or PLS-2. The classification accuracy of a given network between ASD and healthy controls was analyzed using a two-sample t-test. The relationships between ReHo and ASD behaviors were calculated using the Pearson correlation approach. Declarations Data availability The meta-analytical terms used were obtained from Neurosynth (https://neurosynth.org/analyses/terms/). T1w/T2w maps and resting-state fMRI data for functional gradient analysis can be obtained online at https://www.humanconnectome.org/study/hcp-young-adult. The resting-state fMRI of ASD can be obtained from Autism Brain Image Data Exchange I (https://fcon_1000.projects.nitrc.org/indi/abide/abide_I.html). The human gene expression data used in the present study are available in the Allen Human Brain Atlas (‘Complete normalized microarray datasets, http://human.brain-map.org). The disorder-related risk genes were obtained from the Allen Human Brain Atlas and are available at https://help.brain-map.org/display/humanbrain/Documentation. BrainSpan dataset can be found at https://www.brainspan.org/static/download.html. Code availability The resting-state fMRI data from the HCP were preprocessed using the HCP minimal processing pipeline (available at https://github.com/Washington-University/HCPpipelines). The resting-state fMRI data of ASD were preprocessed using Analysis of Functional Neuroimages software (AFNI, https://afni.nimh.nih.gov/). Functional gradient analysis was performed using the Brainspace toolbox (https://github.com/MICA-MNI/BrainSpace). The gene expression data of the brain samples were preprocessed using an AHBA processing pipeline (https://github.com/BMHLab/AHBAprocessing). Partial least squares regression analysis was performed using a standard pipeline (https://github.com/KirstieJane/NSPN_WhitakerVertes_PNAS2016). Gene enrichment analysis was conducted in a Metascape analysis tool (https://metascape.org/gp/index.html#/main/step1). The ReHo was calculated using the DPABI toolbox (https://rfmri.org/DPABI). The SVM classification analysis used a MATLAB toolbox (https://www.csie.ntu.edu.tw/~cjlin/libsvm/). Additional analyses were carried out using custom scripts written in MATLAB R2023b (available at https://github.com/DevlinHu/Autism-Profile). Acknowledgements This research was supported by the Natural Science Research Projects of Anhui Provincial Department of Education (grant no. KJ2021A0580 to S.H., 2022AH050475 to F.L.), and Major Natural Science Research Projects of Anhui Provincial Department of Education (2024AH040143 to J.S.). References Lord, C. , et al. Autism spectrum disorder. Nature Reviews Disease Primers 6 (2020). Lord, C., Elsabbagh, M., Baird, G. & Veenstra-Vanderweele, J. Autism spectrum disorder. The Lancet 392 , 508-520 (2018). Kohane, I.S. , et al. The co-morbidity burden of children and young adults with autism spectrum disorders. PloS one 7 , e33224 (2012). Lombardo, M.V., Lai, M.-C. & Baron-Cohen, S. Big data approaches to decomposing heterogeneity across the autism spectrum. Molecular psychiatry 24 , 1435-1450 (2019). Jeste, S.S. & Geschwind, D.H. Disentangling the heterogeneity of autism spectrum disorder through genetic findings. Nature Reviews Neurology 10 , 74-81 (2014). Buch, A.M. , et al. Molecular and network-level mechanisms explaining individual differences in autism spectrum disorder. Nature Neuroscience 26 , 650-663 (2023). Masi, A., DeMayo, M.M., Glozier, N. & Guastella, A.J. An Overview of Autism Spectrum Disorder, Heterogeneity and Treatment Options. Neuroscience Bulletin 33 , 183-193 (2017). Kana, R.K. Sentence comprehension in autism: thinking in pictures with decreased functional connectivity. Brain 129 , 2484-2493 (2006). Koyama, M.S. , et al. Resting-State Functional Connectivity Indexes Reading Competence in Children and Adults. Journal of Neuroscience 31 , 8617-8624 (2011). Green, S.A., Hernandez, L., Bookheimer, S.Y. & Dapretto, M. Salience network connectivity in autism is related to brain and behavioral markers of sensory overresponsivity. Journal of the American Academy of Child & Adolescent Psychiatry 55 , 618-626. e611 (2016). Di Martino, A. , et al. The autism brain imaging data exchange: towards a large-scale evaluation of the intrinsic brain architecture in autism. Molecular psychiatry 19 , 659-667 (2014). Hong, S.-J. , et al. Atypical functional connectome hierarchy in autism. Nature Communications 10 (2019). Bernhardt, B.C., Milham, M.P., Di Martino, A., Valk, S.L. & Hong, S.-J. Multidimensional Neuroanatomical Subtyping of Autism Spectrum Disorder. Cerebral Cortex 28 , 3578-3588 (2018). Kim, J. , et al. Low-dielectric-constant polyimide aerogel composite films with low water uptake. Polymer Journal 48 , 829-834 (2016). Yarkoni, T., Poldrack, R.A., Nichols, T.E., Van Essen, D.C. & Wager, T.D. Large-scale automated synthesis of human functional neuroimaging data. Nature Methods 8 , 665-670 (2011). Dockès, J. , et al. NeuroQuery, comprehensive meta-analysis of human brain mapping. elife 9 , e53385 (2020). Gorgolewski, K.J. , et al. NeuroVault. org: a web-based repository for collecting and sharing unthresholded statistical maps of the human brain. Frontiers in neuroinformatics 9 , 8 (2015). Burt, J.B. , et al. Hierarchy of transcriptomic specialization across human cortex captured by structural neuroimaging topography. Nature neuroscience 21 , 1251-1259 (2018). Hawrylycz, M. , et al. Canonical genetic signatures of the adult human brain. Nature neuroscience 18 , 1832-1844 (2015). Hawrylycz, M.J. , et al. An anatomically comprehensive atlas of the adult human brain transcriptome. Nature 489 , 391-399 (2012). Abdi, H. Partial least squares regression and projection on latent structure regression (PLS Regression). Wiley interdisciplinary reviews: computational statistics 2 , 97-106 (2010). Hu, S. , et al. Structural lesions and transcriptomic specializations shape gradient perturbations in Wilson disease. Brain Communications 6 (2024). Li, M. , et al. Integrative functional genomic analysis of human brain development and neuropsychiatric risks. Science 362 , eaat7615 (2018). Zhu, Y. , et al. Spatiotemporal transcriptomic divergence across human and macaque brain development. Science 362 , eaat8077 (2018). Glasser, M.F. , et al. A multi-modal parcellation of human cerebral cortex. Nature 536 , 171-178 (2016). Huntenburg, J.M., Bazin, P.L. & Margulies, D.S. Large-Scale Gradients in Human Cortical Organization. Trends Cogn Sci 22 , 21-31 (2018). Arnatkeviciute, A., Fulcher, B.D. & Fornito, A. A practical guide to linking brain-wide gene expression and neuroimaging data. Neuroimage 189 , 353-367 (2019). Martin, J. , et al. Biological overlap of attention-deficit/hyperactivity disorder and autism spectrum disorder: evidence from copy number variants. Journal of the American Academy of Child & Adolescent Psychiatry 53 , 761-770. e726 (2014). Tseng, Q. , et al. Spatial organization of the extracellular matrix regulates cell–cell junction positioning. Proceedings of the National Academy of Sciences 109 , 1506-1511 (2012). Raut, R.V., Snyder, A.Z. & Raichle, M.E. Hierarchical dynamics as a macroscopic organizing principle of the human brain. Proceedings of the National Academy of Sciences 117 , 20890-20897 (2020). Alexander-Bloch, A.F. , et al. Imaging local genetic influences on cortical folding. Proceedings of the National Academy of Sciences 117 , 7430-7436 (2020). Xia, Y. , et al. Development of functional connectome gradients during childhood and adolescence. Science Bulletin 67 , 1049-1061 (2022). Burt, J.B. , et al. Hierarchy of transcriptomic specialization across human cortex captured by structural neuroimaging topography. Nat Neurosci 21 , 1251-1259 (2018). Dear, R. , et al. Cortical gene expression architecture links healthy neurodevelopment to the imaging, transcriptomics and genetics of autism and schizophrenia. Nature Neuroscience 27 , 1075-1086 (2024). Benkarim, O. , et al. Connectivity alterations in autism reflect functional idiosyncrasy. Communications Biology 4 (2021). Rolls, E.T. , et al. Effective connectivity in autism. Autism Research 13 , 32-44 (2019). Wallace, G.L., Dankner, N., Kenworthy, L., Giedd, J.N. & Martin, A. Age-related temporal and parietal cortical thinning in autism spectrum disorders. Brain 133 , 3745-3754 (2010). Zilbovicius, M. , et al. Temporal lobe dysfunction in childhood autism: a PET study. American Journal of Psychiatry 157 , 1988-1993 (2000). McAlonan, G.M. Mapping the brain in autism. A voxel-based MRI study of volumetric differences and intercorrelations in autism. Brain 128 , 268-276 (2004). Schultz, R.T. , et al. Abnormal ventral temporal cortical activity during face discrimination among individuals with autism and Asperger syndrome. Archives of general Psychiatry 57 , 331-340 (2000). Eyler, L.T., Pierce, K. & Courchesne, E. A failure of left temporal cortex to specialize for language is an early emerging and fundamental property of autism. Brain 135 , 949-960 (2012). Tong, C. , et al. Multimodal analysis demonstrating the shaping of functional gradients in the marmoset brain. Nature Communications 13 (2022). Hansen, J.Y. , et al. Mapping gene transcription and neurocognition across human neocortex. Nat Hum Behav 5 , 1240-1250 (2021). Clements, C.C. , et al. Evaluation of the social motivation hypothesis of autism: a systematic review and meta-analysis. JAMA psychiatry 75 , 797-808 (2018). Philip, R.C. , et al. A systematic review and meta-analysis of the fMRI investigation of autism spectrum disorders. Neuroscience & Biobehavioral Reviews 36 , 901-942 (2012). Horien, C. , et al. Functional Connectome–Based Predictive Modeling in Autism. Biological Psychiatry 92 , 626-642 (2022). Váša, F. , et al. Conservative and disruptive modes of adolescent change in human brain functional connectivity. Proceedings of the National Academy of Sciences 117 , 3248-3253 (2020). Evans, M.M., Kim, J., Abel, T., Nickl-Jockschat, T. & Stevens, H.E. Developmental Disruptions of the Dorsal Striatum in Autism Spectrum Disorder. Biological Psychiatry 95 , 102-111 (2024). Murayama, C. , et al. Extrastriatal dopamine D2/3 receptor binding, functional connectivity, and autism socio-communicational deficits: a PET and fMRI study. Molecular Psychiatry 27 , 2106-2113 (2022). Wold, S., Esbensen, K. & Geladi, P. Principal component analysis. Chemometrics and intelligent laboratory systems 2 , 37-52 (1987). Belkin, M. & Niyogi, P. Laplacian eigenmaps for dimensionality reduction and data representation. Neural computation 15 , 1373-1396 (2003). Glasser, M.F. , et al. The minimal preprocessing pipelines for the Human Connectome Project. Neuroimage 80 , 105-124 (2013). Vos de Wael, R. , et al. BrainSpace: a toolbox for the analysis of macroscale gradients in neuroimaging and connectomics datasets. Communications Biology 3 (2020). Whitaker, K.J. , et al. Adolescence is associated with genomically patterned consolidation of the hubs of the human brain connectome. Proc Natl Acad Sci U S A 113 , 9105-9110 (2016). Zhou, Y. , et al. Metascape provides a biologist-oriented resource for the analysis of systems-level datasets. Nature communications 10 , 1523 (2019). Lee, H.M. , et al. Multimodal mapping of regional brain vulnerability to focal cortical dysplasia. Brain 146 , 3404-3415 (2023). Yan, C.-G., Wang, X.-D., Zuo, X.-N. & Zang, Y.-F. DPABI: data processing & analysis for (resting-state) brain imaging. Neuroinformatics 14 , 339-351 (2016). Zang, Y., Jiang, T., Lu, Y., He, Y. & Tian, L. Regional homogeneity approach to fMRI data analysis. Neuroimage 22 , 394-400 (2004). Qiu, B. , et al. Transcriptomic profiles link corticostriatal microarchitecture to genetics of neurodevelopment and neuropsychiatric risks. (2024). Wang, X. , et al. The Brain’s Topographical Organization Shapes Dynamic Interaction Patterns That Support Flexible Behavior Based on Rules and Long-Term Knowledge. The Journal of Neuroscience 44 (2024). Additional Declarations There is NO Competing Interest. Supplementary Files SupplementaryMaterial.docx Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-5425486","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":382030076,"identity":"88a35840-07b3-45cb-a181-6df7e19d48a8","order_by":0,"name":"Sheng Hu","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA2ElEQVRIiWNgGAWjYBACAwYGNhib8UFCRQ0pWtgYmA0enDlGmhY2yYctzIS1mEukP3vwcUdt4vz5zc8qEhvYGPjbuxPwarGckWNuOPPM8cQNx9jMbiTukGGQOHN2A36H3chhk+ZtO5a4gY2H7UbiGTYGA4lcQlrSn0n/BWqZ38bDVpDYxkyMlgQzaca2msSGYzxsDMRpOfPGTLK37YDxhmNpxhIJZ47xEPbL8fRnEj/b6mTnNx9++PFHRY0cf3svfi1QcBjO4iFGOQjUEatwFIyCUTAKRiIAABzoS2H3QVdGAAAAAElFTkSuQmCC","orcid":"","institution":"University of science and technology of China","correspondingAuthor":true,"prefix":"","firstName":"Sheng","middleName":"","lastName":"Hu","suffix":""},{"id":382030077,"identity":"322c1623-21cd-44d3-a485-faee0f891c0f","order_by":1,"name":"Yingxing Zhang","email":"","orcid":"","institution":"Child Healthcare Department, Anhui Provincial Children’s Hospital","correspondingAuthor":false,"prefix":"","firstName":"Yingxing","middleName":"","lastName":"Zhang","suffix":""},{"id":382030078,"identity":"c64ab13d-6b58-4712-9cfe-d5814ebb4247","order_by":2,"name":"Fangfang Li","email":"","orcid":"","institution":"School of Medical Information Engineering, Anhui University of Chinese Medicine","correspondingAuthor":false,"prefix":"","firstName":"Fangfang","middleName":"","lastName":"Li","suffix":""},{"id":382030079,"identity":"a84a5069-0169-4e39-8fbf-9eb44423a048","order_by":3,"name":"Hongli Wu","email":"","orcid":"","institution":"School of Medical Information Engineering, Anhui University of Chinese Medicine","correspondingAuthor":false,"prefix":"","firstName":"Hongli","middleName":"","lastName":"Wu","suffix":""},{"id":382030080,"identity":"9c9bbb06-ffb4-4fd4-b73a-61015709fc46","order_by":4,"name":"Wei Du","email":"","orcid":"","institution":"School of Medical Information Engineering, Anhui University of Chinese Medicine","correspondingAuthor":false,"prefix":"","firstName":"Wei","middleName":"","lastName":"Du","suffix":""},{"id":382030081,"identity":"fd85167a-cf1a-41c5-ae93-1e0ebdde43ae","order_by":5,"name":"Jianhua Shu","email":"","orcid":"","institution":"School of Medical Information Engineering, Anhui University of Chinese Medicine","correspondingAuthor":false,"prefix":"","firstName":"Jianhua","middleName":"","lastName":"Shu","suffix":""},{"id":382030082,"identity":"422ea28c-ce3c-45d2-8f6f-43374b37354b","order_by":6,"name":"Anqing Wang","email":"","orcid":"","institution":"Medical Imaging Center, First Affiliated Hospital of Anhui University of Chinese Medicine","correspondingAuthor":false,"prefix":"","firstName":"Anqing","middleName":"","lastName":"Wang","suffix":""},{"id":382030083,"identity":"48b7c482-c533-4337-971f-58e293a84c83","order_by":7,"name":"Chunsheng Xu","email":"","orcid":"","institution":"Medical Imaging Center, First Affiliated Hospital of Anhui University of Chinese Medicine","correspondingAuthor":false,"prefix":"","firstName":"Chunsheng","middleName":"","lastName":"Xu","suffix":""},{"id":382030084,"identity":"0a9af281-d9d3-4df4-abdc-5d5c64ef5409","order_by":8,"name":"Chuanfu Li","email":"","orcid":"","institution":"Medical Imaging Center, First Affiliated Hospital of Anhui University of Chinese Medicine","correspondingAuthor":false,"prefix":"","firstName":"Chuanfu","middleName":"","lastName":"Li","suffix":""},{"id":382030085,"identity":"dc000d2b-9fbb-47f2-8b97-8f375aa6401a","order_by":9,"name":"Ya Wang","email":"","orcid":"","institution":"Child Healthcare Department, Anhui Provincial Children’s Hospital","correspondingAuthor":false,"prefix":"","firstName":"Ya","middleName":"","lastName":"Wang","suffix":""}],"badges":[],"createdAt":"2024-11-10 10:40:14","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-5425486/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-5425486/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":69861465,"identity":"0c9cc9b3-8dd6-4d6f-8b6b-2fc1e47d3cea","added_by":"auto","created_at":"2024-11-26 05:40:00","extension":"jpg","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":491826,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eThe principal neurocognitive profile reflects cortical hierarchical properties.\u003c/strong\u003e \u003cstrong\u003ea\u003c/strong\u003e, The neurocognitive profile reflects the posterior-anterior axis. The spatial correlation was calculated between the neurocognitive profile and geodesic distance (\u003cem\u003eR\u003c/em\u003e = -0.2388, \u003cem\u003eP\u003c/em\u003e\u003csub\u003e\u003cem\u003ePerm\u003c/em\u003e\u003c/sub\u003e = 0.0019). The geodesic distance was defined as the Euclidean distance between each cortical parcel from the HCP_MMP1.0 atlas and the early visual cortex. \u003cstrong\u003eb\u003c/strong\u003e, The neurocognitive profile reflects structural hierarchy. The spatial correlation was calculated between the neurocognitive profile and the T1w/T2w ratio (\u003cem\u003eR\u003c/em\u003e = 0.3291, \u003cem\u003eP\u003c/em\u003e\u003csub\u003e\u003cem\u003ePerm\u003c/em\u003e\u003c/sub\u003e \u0026lt; 0.0001). The averaged T1w/T2w ratio map of the left hemisphere cortex was averaged from the individual T1w/T2w ratio maps across 1113 subjects from the HCP. \u003cstrong\u003ec\u003c/strong\u003e, The neurocognitive profile reflects functional hierarchy. The spatial correlation was calculated between the neurocognitive profile and functional gradient (\u003cem\u003eR\u003c/em\u003e = -0.3415, \u003cem\u003eP\u003c/em\u003e\u003csub\u003e\u003cem\u003ePerm\u003c/em\u003e\u003c/sub\u003e \u0026lt; 0.0001). We applied the diffusion map embedding on a group-averaged FC matrix computed from the 1,084 HCP participants to generate the functional gradient. The statistical significance of spatial correlation was evaluated using a permutation test with 10000 times. HCP, human connectome project; FC, functional connectivity.\u003c/p\u003e","description":"","filename":"Figure2.jpg","url":"https://assets-eu.researchsquare.com/files/rs-5425486/v1/71a245a73bda450f208dc864.jpg"},{"id":69862449,"identity":"4943cfc0-d12c-48db-bd61-b129d3400980","added_by":"auto","created_at":"2024-11-26 06:04:01","extension":"jpg","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":1496672,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eAssociations between principal neurocognitive profile and gene expression.\u003c/strong\u003e \u003cstrong\u003ea\u003c/strong\u003e, Gene expression map derived based on the local maximum correlation. The color scale indicates the score for PLS-1 and PLS-2, namely, the weighted average expression level of 10027. \u003cstrong\u003eb\u003c/strong\u003e, Correlation between PLS and neurocognitive profile. Both PLS-1 (\u003cem\u003eR\u003c/em\u003e = 0.37, \u003cem\u003eP\u003c/em\u003e\u003csub\u003e\u003cem\u003ePerm\u003c/em\u003e\u003c/sub\u003e \u0026lt; 0.0001) and PLS-2 (\u003cem\u003eR\u003c/em\u003e = 0.39, \u003cem\u003eP\u003c/em\u003e\u003csub\u003e\u003cem\u003ePerm\u003c/em\u003e\u003c/sub\u003e \u0026lt; 0.0001) show significant correlation with the neurocognitive profile. The statistical significance of spatial correlation was evaluated using a permutation test with 10000 times. \u003cstrong\u003ec\u003c/strong\u003e, Enrichment network showing the intra- and intercluster similarities of enriched annotations. Each term is represented by a node, where its size represents the number of input genes included in that term, and its color represents its cluster identity. \u003cstrong\u003ed\u003c/strong\u003e, Specific analysis. Left, PLS-1 is enriched in ASD pathogenic genes (\u003cem\u003eP\u003c/em\u003e = 0.0095, \u003cem\u003eER\u003c/em\u003e = -3.353). Right, PLS-1 is also enriched for risk genes of depression, while PLS-2 was additionally enriched for that of epilepsy. A bootstrapped permutation test (N = 10000) was used to evaluate the significance of the observed ER, followed by FDR correction for multiple comparisons. The dotted line indicates FDR = 0.05. \u003cstrong\u003ee\u003c/strong\u003e, Neurocognitive terms contribute to the weighted gene expression of PLS-1. The contribution of each neurocognitive term to weighted gene expression of PLS-1 was calculated as the Pearson correlation between the term’s functional association map and PLS-1 score. The font size of a given cognitive term corresponds to the correlation coefficient between the term’s functional association map and PLS-1 score. ER, enrichment ratio, FDR, false discovery rate.\u003c/p\u003e","description":"","filename":"Figure3.jpg","url":"https://assets-eu.researchsquare.com/files/rs-5425486/v1/e9a46e4b8db1030abd209178.jpg"},{"id":69862448,"identity":"1301ac3f-05d6-4dfe-87ef-9efb056c93b3","added_by":"auto","created_at":"2024-11-26 06:04:01","extension":"jpg","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":427875,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eIdentifying the topographic network for neurocognitive processes of phenotypic heterogeneity in ASD. a\u003c/strong\u003e, The SVM classification accuracy. We set a range of thresholds (from 0 to |0.1|) with a step of 0.01 for the PLS score to identify the topographic networks of phenotypic heterogeneity in ASD. We began by applying a threshold to the PLS scores, which allowed us to generate a topographic network. Based on the topographic network, the functional features that are ReHo calculated from a large-scare fMRI dataset of ASD. Finally, these features feed into the SVM classification model for distinguishing ASD patients from healthy controls. *, \u003cem\u003eP\u003c/em\u003e \u0026lt; 0.05; **, \u003cem\u003eP\u003c/em\u003e \u0026lt; 0.01. \u003cstrong\u003eb\u003c/strong\u003e, The topographic network. Left, the topographic network was identified based on the highest SVM classification accuracy. Right, the observed classification accuracy was significant when compared with the null distribution of classification accuracies (Accuracy = 63.67%, \u003cem\u003eP\u003c/em\u003e = 0.002). \u003cstrong\u003ec\u003c/strong\u003e, Correlations between ReHo and clinical symptoms. The averaged ReHo within the regions with higher or lower gene expression were extracted to calculate Pearson’s correlation with clinical symptoms of ASD, such as ADOS SOCIAL, ADOS STEREO, and age. ReHo, regional homogeneity; SVM, support vector model; ADOS SOCIAL, social total subscore of the classic ADOS. ADOS STEREO, stereotyped behaviors, and restricted interests total subscore of classic ADOS.\u003c/p\u003e","description":"","filename":"Figure4.jpg","url":"https://assets-eu.researchsquare.com/files/rs-5425486/v1/d8d2e1bae55feea8d5397550.jpg"},{"id":69861470,"identity":"b951633f-fb4b-49f4-b461-174f9cbfee1e","added_by":"auto","created_at":"2024-11-26 05:40:00","extension":"jpg","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":244918,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eDevelopmental spatiotemporal trajectory of gene-neurocognition for the principal neurocognitive profile. a\u003c/strong\u003e, The spatiotemporal trajectory of PLS-1. \u003cstrong\u003eb\u003c/strong\u003e, The spatiotemporal trajectory of PLS-2. Dots represent cortical samples at a given time point color-coded by lobes. The colored lines are the third-order polynomial regression across the time window, thereby showing the overall trajectory.\u003c/p\u003e","description":"","filename":"Figure5.jpg","url":"https://assets-eu.researchsquare.com/files/rs-5425486/v1/e61463bed487d0b0a4659bce.jpg"},{"id":69861471,"identity":"86cabc74-004b-4f11-ad00-a202ca9b8a45","added_by":"auto","created_at":"2024-11-26 05:40:00","extension":"jpg","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":736866,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eFunctional reorganization within the topographic network in ASD.\u003c/strong\u003e \u003cstrong\u003ea\u003c/strong\u003e, Overview of analytic approaches. We used the BOLD time series from each parcel to train an SVM model for achieving six network classification tasks. We began by averaging the BOLD time series within each parcel associated with a given network. The resulting averaged time series for each parcel were then selected as features for training the SVM classification model. We used a one-vs-one approach to classify the given network from other networks. Then, we constructed the confusion matrix to summarize the prediction results for each network. In this matrix, each row represents an instance of the actual class (i.e., an actual network), and each column represents an instance of the predicted class (i.e., the predicted network). \u003cstrong\u003eb\u003c/strong\u003e, The probabilities of classifying FTC as each network. \u003cstrong\u003ec\u003c/strong\u003e, The probabilities of classifying TPC as each network. \u003cstrong\u003ed\u003c/strong\u003e, The probabilities of classifying ADC as each network. \u003cstrong\u003ee\u003c/strong\u003e, The probabilities of classifying SMC as each network. \u003cstrong\u003ef\u003c/strong\u003e, The probabilities of classifying POC as each network. \u003cstrong\u003eg\u003c/strong\u003e, The probabilities of classifying VSC as each network. The differences in probabilities of each network between ASD and healthy controls were analyzed using a two-sample t-test. FTC, frontal cortex; TPC, temporal cortex; ADC, auditory cortex; SMC, sensorimotor cortex; POC, posterior cortex; VSC, visual cortex.\u003c/p\u003e","description":"","filename":"Figure6.jpg","url":"https://assets-eu.researchsquare.com/files/rs-5425486/v1/cba6745f0b0e36e45fc92228.jpg"},{"id":69923127,"identity":"17a8de59-6539-4202-9fc7-b5abe2d2334f","added_by":"auto","created_at":"2024-11-26 15:49:52","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":4345965,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-5425486/v1/7210f87f-fd17-4854-a451-1f23ca667f7c.pdf"},{"id":69861466,"identity":"04781ddb-4ba7-4071-8890-db83fe1a8f35","added_by":"auto","created_at":"2024-11-26 05:40:00","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":1380843,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cbr\u003e\u003c/p\u003e","description":"","filename":"SupplementaryMaterial.docx","url":"https://assets-eu.researchsquare.com/files/rs-5425486/v1/a9dbd5a0330301aadf7f8456.docx"}],"financialInterests":"There is \u003cb\u003eNO\u003c/b\u003e Competing Interest.","formattedTitle":"Genetic neurocognitive profile of autism unveiled with gene transcription","fulltext":[{"header":"Main","content":"\u003cp\u003eAutism spectrum disorder (ASD)\u0026nbsp;is a highly heritable and heterogeneous neurodevelopmental disorder\u0026nbsp;characterized by impairments in social communication and interaction, anomalies in sensory responses, repetitive behaviors, and restricted interests\u003csup\u003e1,2\u003c/sup\u003e. The heterogeneity is also complicated by the manifestation of\u0026nbsp;comorbidities, highlighting the mental burden in ASD\u003csup\u003e3\u003c/sup\u003e. Distinct conditions within ASD may exhibit varying pathophysiological processes\u003csup\u003e4\u003c/sup\u003e, as identified by genetic and magnetic resonance imaging (MRI) analyses\u003csup\u003e5,6\u003c/sup\u003e. The co-occurring ASD conditions complicate the disease course and lead to functional deterioration and poorer clinical outcomes\u003csup\u003e7\u003c/sup\u003e. Thus, clarifying the neural mechanisms of\u0026nbsp;comorbidity within ASD benefits from discovering new therapeutic approaches for different types of ASD.\u003c/p\u003e\n\u003cp\u003eFunctional MRI (fMRI) studies have uncovered that atypical activity in the visual cortex and\u0026nbsp;salient network\u0026nbsp;is involved in dysregulation with social cognition and language processing in ASD\u003csup\u003e8-10\u003c/sup\u003e, and that aberrant functional connectivity in the corticostriatal and frontoparietal network\u0026nbsp;can characterize different types of ASD\u003csup\u003e6,11\u003c/sup\u003e. Cortex-wide gradient mapping techniques described a disruption in macroscale hierarchy affecting integration and segregation of unimodal and transmodal networks in ASD, especially involving sensory and higher-order default mode regions\u003csup\u003e12\u003c/sup\u003e. Moreover, the neuroimaging basis of phenotypic heterogeneity has been demonstrated in ASD, exhibiting that the subgroups defined by anatomical features can enhance the prediction of symptom severity in subtyping ASD\u003csup\u003e13\u003c/sup\u003e, that the abnormal brain connections distinguish individuals with ASD from healthy controls\u003csup\u003e14\u003c/sup\u003e and that atypical connectivity patterns underlie different forms of ASD\u003csup\u003e6\u003c/sup\u003e.\u0026nbsp;However,\u0026nbsp;no studies attempt to uncover the neurocognitive architecture\u0026nbsp;of phenotypic heterogeneity\u0026nbsp;for ASD. In other words, the neurobiological basis of comorbidities in ASD remains poorly understood.\u003c/p\u003e\n\u003cp\u003eModern advancements in neuroimaging, coupled with the promotion of global data-sharing\u0026nbsp;initiatives, have fundamentally enhanced our ability to investigate neurocognitive processes and their underlying molecular mechanisms. Meta-analyses by synthesizing human neuroimaging data have informed how brain regions respond to a spectrum of cognitive, perceptual, and emotional tasks\u003csup\u003e15-17\u003c/sup\u003e. Moreover, advanced high-throughput transcriptomes have provided precise gene expression maps distributed over the whole brain, highly linking the spatial distribution of molecular dynamics with macroscale cortical organization of anatomical properties\u003csup\u003e18-20\u003c/sup\u003e. Therefore, integrating brain-mapping initiatives and high-throughput transcriptomes into neuroimaging analyses offers an unheard-of opportunity to unravel spatial associations between the brain\u0026rsquo;s gene expression profiles and the neurocognitive architecture\u0026nbsp;of phenotypic heterogeneity for ASD.\u003c/p\u003e\n\u003cp\u003eHere, we obtain probabilistic functional association maps linked to the\u0026nbsp;phenotypic heterogeneity of ASD\u0026nbsp;from Neurosynth\u003csup\u003e15\u003c/sup\u003e to construct the\u0026nbsp;neurocognitive\u0026nbsp;architecture. Then, we test whether the neurocognitive architecture is related to the properties of macroscale hierarchy. In parallel, we apply partial least-squares (PLS) analysis\u003csup\u003e21\u003c/sup\u003e to identify the spatial associations between the transcriptional data (Allen Human Brain Atlas (AHBA)\u003csup\u003e20\u003c/sup\u003e) and the neurocognitive architecture. This procedure can generate the weighted gene expression maps for these neurocognitive processes. Gene enrichment analysis was further employed to investigate transcriptomic associations with known pathogenic variants linked to ASD and explore transcriptomic associations with other psychiatric risks\u003csup\u003e22\u003c/sup\u003e. Considering that ASD is a neurodevelopmental disorder, we evaluated spatiotemporal gene expression dynamics from prenatal to adulthood stages using a developmental genetic dataset of BrainSpan\u003csup\u003e23,24\u003c/sup\u003e. Finally, we identify the brain\u0026ndash;behavior, and brain organization (see methods) for explaining the macroscale hierarchy of phenotypic heterogeneity in ASD using a large-scale resting-state fMRI dataset (Autism Brain Imaging Data Exchange (ABIDE) I\u003csup\u003e11\u003c/sup\u003e).\u003c/p\u003e"},{"header":"Results","content":"\u003cp\u003e\u003cstrong\u003eNeurocognitive profiles of ASD\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eTo establish the neurocognitive architecture, we acquired the functional association maps, which represent probabilistic measures of whether cognitive terms (such as restricted, social interactions, and anxiety) from Neurosynth are functionally related to specific brain regions\u003csup\u003e15\u003c/sup\u003e. The terms were chosen based on our literature reviews concerning heterogeneous phenotypes in ASD\u003csup\u003e1,2,5,7\u003c/sup\u003e (\u003cstrong\u003eSupplementary Table 1\u003c/strong\u003e). The functional association maps were subsequently projected on the cerebral cortex using HCP_MMP1.0 with 360 parcels\u003csup\u003e25\u003c/sup\u003e (\u003cstrong\u003eFig. 1a\u003c/strong\u003e). Then, we applied the principal component analysis (PCA) to determine the dominant component of spatial variation across these functional association maps,\u0026nbsp;thereby\u0026nbsp;defining the dominant\u0026nbsp;components as neurocognitive profiles for ASD (\u003cstrong\u003eSupplementary Fig. 1\u003c/strong\u003e). The principal neurocognitive profile exhibits higher PCA scores in the superior medial frontal cortex, middle frontal cortex, insular cortex, and parietal cortex while showing lower PCA scores in the inferior temporal cortex and inferior medial frontal cortex (\u003cstrong\u003eFig. 1b\u003c/strong\u003e). Then, we assigned parcels of HCP_MMP1.0 to six networks, followed by projecting the PCA scores into the networks\u003csup\u003e25\u003c/sup\u003e. At the network level, the temporal cortex (TPC) shows a significantly lower PCA score than other networks (\u003cstrong\u003eFig. 1c\u003c/strong\u003e), including the frontal cortex (FTC), auditory cortex (ADC), sensorimotor cortex (SMC), posterior cortex (POC) and visual cortex (VSC). The second neurocognitive profile demonstrates higher PCA scores in the FTC, TPC, and posterior cingulate cortex while exhibiting lower PCA scores in the SMC, POC, and VSC (\u003cstrong\u003eSupplementary Fig. 2\u003c/strong\u003e). Moreover, we also employed the Laplacian eigenmaps to reproduce the\u0026nbsp;neurocognitive profiles (\u003cstrong\u003eFig. 1d\u003c/strong\u003e). The Laplacian eigen score is significantly correlated with the PCA score of the dominant component (\u003cem\u003eR\u003c/em\u003e = 0.8687, \u003cem\u003eP\u003csub\u003ePerm\u003c/sub\u003e\u003c/em\u003e \u0026lt; 0.0001, \u003cstrong\u003eFig. 1d\u003c/strong\u003e).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eNeurocognitive profiles reflect cortical hierarchies\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eHaving identified neurocognitive profiles for phenotypic heterogeneity in ASD, we next investigated whether these topographic patterns relate to cortical hierarchical properties\u003csup\u003e26\u003c/sup\u003e. Here, we first calculated the spatial correlation between the principal neurocognitive profiles and the map of geodesic distance from the early visual cortex that defines a posterior-anterior gradient, exhibiting a strong negative correlation (\u003cem\u003eR\u003c/em\u003e = -0.2388, \u003cem\u003eP\u003csub\u003ePerm\u003c/sub\u003e\u003c/em\u003e = 0.0019; \u003cstrong\u003eFig. 2\u003c/strong\u003e\u003cstrong\u003ea\u003c/strong\u003e). We then averaged a measurement of the T1w/T2w ratio (a widely used proxy for intracortical myelin) from the left hemisphere cortex across 1113 subjects from the Human Connectome Project (HCP). The principal neurocognitive profile shows a significant positive spatial correlation with the T1w/T2w ratio map (\u003cem\u003eR\u003c/em\u003e = 0.3291, \u003cem\u003eP\u003csub\u003ePerm\u003c/sub\u003e\u003c/em\u003e \u0026lt; 0.0001; \u003cstrong\u003eFig. 2b\u003c/strong\u003e). Finally, we applied the diffusion map embedding on a group-averaged functional connectivity (FC) matrix computed from the 1,084 HCP participants to generate the principal functional gradient.\u0026nbsp;The principal neurocognitive profile has a strong negative spatial correlation with the principal functional gradient (\u003cem\u003eR\u003c/em\u003e = -0.3415, \u003cem\u003eP\u003csub\u003ePerm\u003c/sub\u003e\u003c/em\u003e \u0026lt; 0.0001; \u003cstrong\u003eFig. 2c\u003c/strong\u003e). In addition, the second neurocognitive profile also shows strong spatial correlations with cortical hierarchical properties (\u003cstrong\u003eSupplementary Fig. 3\u003c/strong\u003e).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTranscriptomic correlates of neurocognitive profiles\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eTo further explore the transcriptomic signatures of the neurocognitive profiles, we mapped normative regional gene expression profiles for 10027 microarray probes in the AHBA, including gene expression data for 1290 samples from 6 healthy donors (N = 5 males, N = 1 female, aged 24~57 years)\u003csup\u003e20\u003c/sup\u003e, to the HCP_MMP1.0 atlas using a standard processing pipeline\u003csup\u003e27\u003c/sup\u003e. Next, we used the PLS regression analysis to test the associations of weighted gene expression with the spatial distribution of the principal neurocognitive profile. Two PLS components explained 17.7% (PLS-1: P\u003csub\u003ePerm\u003c/sub\u003e =\u0026nbsp;0.004) and 19% (PLS-2: P\u003csub\u003ePerm\u003c/sub\u003e \u0026lt; 0.001) of the covariance between the principal neurocognitive profile and AHBA gene expression (\u003cstrong\u003eFig. 3a and Supplementary Fig. 4\u003c/strong\u003e). Also, both PLS-1 (\u003cem\u003eR\u003c/em\u003e = 0.37, \u003cem\u003eP\u003c/em\u003e\u0026lt; 0.0001) and PLS-2 (\u003cem\u003eR\u003c/em\u003e = 0.39, \u003cem\u003eP\u003c/em\u003e\u0026lt; 0.0001) are spatially correlated with the principal neurocognitive profile (\u003cstrong\u003eFig. 3b\u003c/strong\u003e). PLS-1 represents genes with high-weighted gene expression in the somatosensory and visual cortex but showing low expression in the medial frontal, insular, and temporal cortex\u0026nbsp;(\u003cstrong\u003eFig. 3a\u003c/strong\u003e). PLS-2 is characterized by high-weighted gene expression in the middle frontal cortex but shows low expression in the visual cortex (\u003cstrong\u003eFig. 3a\u003c/strong\u003e).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eGene enrichment analysis results exhibit that PLS-1 is mainly enriched in ADHD and autism ASD-linked metabolic pathways and SNP, head development, regulation of metal ion transport, and regulation of trans-synaptic signaling, indicating that the neurocognitive profile reflects the neurobiological basis of phenotypic heterogeneity in ASD\u003csup\u003e28\u003c/sup\u003e; PSL-2 is primarily involved in response to growth factor, neuron projection development, cell junction organization and extracellular matrix organization, foreboding the neurocognitive profile may be linked to neural connections formation during neuron development\u003csup\u003e29\u003c/sup\u003e (\u003cstrong\u003eFig. 3c\u003c/strong\u003e). Specific analysis reveals that PLS-1 is enriched in autism pathogenic risk and other neuropsychiatric risks, such as depression and epilepsy (\u003cstrong\u003eFig. 3d\u003c/strong\u003e), indicating that the weighted gene expression shapes the neurocognitive processes related to phenotypic heterogeneity of ASD. Furthermore, the pattern of PLS-1 covarying with specific terms, such as restricted, repetition, attention task, sensory, anxiety, and depression, is strongly correlated with heterogenous phenotypes or clinical symptoms comorbidity in ASD.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThe PLS regression analysis was also performed to test the associations of weighted gene expression with the spatial distribution of the second neurocognitive profile, displaying weighted gene expression maps of two PLS components (\u003cstrong\u003eSupplementary Fig. 5a\u003c/strong\u003e). PLS-1 is enriched in the regulation of neuron projection development, regulation of monatomic ion transport, and head development, while PLS-2 is mainly involved in the regulation of cell junction assembly and inorganic ion transmembrane transport (\u003cstrong\u003eSupplementary Fig. 5b\u003c/strong\u003e). Importantly, both PLS-1 and PLS-2 are enriched in autism pathogenic risk and other neuropsychiatric risk such as depression (\u003cstrong\u003eSupplementary Fig. 5c\u003c/strong\u003e). Furthermore, the pattern of PLS is also covarying with terms related to heterogenous phenotypes or clinical symptoms of comorbidity in ASD (\u003cstrong\u003eSupplementary Fig. 5d\u003c/strong\u003e).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eIdentify topographic networks of phenotypic heterogeneity in ASD\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eHaving identified the gene expression signatures for the neurocognitive profiles, we next test whether the weighted gene expressions distributed in the cerebral cortex contribute to these neurocognitive processes. Accordingly, we set a range of thresholds (from 0 to |0.1|) with a step of 0.01 for the PLS score to identify the topographic networks of phenotypic heterogeneity in ASD. Specifically, we began by applying a threshold to the PLS scores, which allowed us to generate a topographic network. We next selected functional features that are regional homogeneity (ReHo) calculated from a large-scare resting-state fMRI dataset of ASD (see methods) based on the topographic network. Finally, these features feed into the support vector machine (SVM) classification model for distinguishing ASD patients from healthy controls. The topographic network can reflect neurocognitive processes of the phenotypic heterogeneity in ASD if the model can accurately classify ASD patients from healthy controls. We found that the PLS-1 score of the principal neurocognitive profile threshed at 0.07 can better explain the neurocognitive processes with a significant classifying accuracy of 63.67% (\u003cem\u003eP\u003c/em\u003e = 0.002, 1000 times permutation; \u003cstrong\u003eFig. 4a and b\u003c/strong\u003e). The topographic network features regions with higher weighted gene expression in the somatosensory and visual cortices, while regions with lower weighted gene expression are found in the medial frontal and temporal cortices (\u003cstrong\u003eFig. 4b\u003c/strong\u003e). Furthermore, we showed the regions with higher gene expression are significantly related\u0026nbsp;to social behaviors (\u003cem\u003eR\u003c/em\u003e = -0.2309, \u003cem\u003eP\u003c/em\u003e = 0.0407) and age (\u003cem\u003eR\u003c/em\u003e = -0.3252, \u003cem\u003eP\u003c/em\u003e = 0.0035), and regions with lower gene expression are strongly related to stereotyped behaviors and restricted interests (\u003cem\u003eR\u003c/em\u003e = 0.2262, \u003cem\u003eP\u003c/em\u003e = 0.045, \u003cstrong\u003eFig. 4c\u003c/strong\u003e), indicating the topographic network involving in ASD’s neurocognitive processes. We also found that the PLS-2 score of the second neurocognitive profile threshed at 0.01 (\u003cstrong\u003eSupplementary Fig. 6\u003c/strong\u003e), which gets a significant classifying accuracy of 59.21% (\u003cem\u003eP\u003c/em\u003e = 0.028, 1000 times permutation). However, no clinical behaviors of ASD are correlated with the topographic network, indicating that the topographic network cannot well explain the neurocognitive processes for phenotypic heterogeneity of ASD.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTranscriptomic signatures in neurocognitive profiles over development\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eGiven the continued development of neurocognitive processes across the lifespan, we sought to investigate the spatiotemporal trajectories of transcriptomic signatures through human development. Here, we analyzed the spatiotemporal trajectory of the identified topographic network, including the frontal, temporal, somatosensory, and visual cortex, over brain development. We employed the BrainSpan\u003csup\u003e23,24\u003c/sup\u003e in the following analyses, a gene expression database of brain tissue across development covering the period from 8 post-conception weeks to 40 years across 16 cortical regions. The ages were binned into nine-time windows (W1 to W9), encompassing the embryonic period, fetal development, infancy, childhood, adolescence, and adulthood. The spatiotemporal transcriptomic signatures were defined as the projection of gene expression of BrainSpan onto the PLS analysis-defined gene expression (such as PLS-1 and PLS-2), generating an estimated gene expression per region and time window (see methods). For PLS-1 of the principal neurocognitive profile, gene expression decreased during embryonic and early to middle fetal stages (W1-W3), followed by an increase throughout the remaining life stages. In addition, the transcriptomic signature has gradually differentiated between brain regions throughout neurodevelopment (\u003cstrong\u003eFig. 5a\u003c/strong\u003e). In contrast, the spatiotemporal trajectory of PLS-2 (\u003cstrong\u003eFig. 5b\u003c/strong\u003e) mirrors that of PLS-1 from embryonic to late childhood stages (W1-W7) but shows a decline from late adolescence to adulthood (W8-W9).\u0026nbsp;Interestingly, the estimated gene expression of the second neurocognitive profile has an opposite spatiotemporal trajectory to that of the principal neurocognitive profile (\u003cstrong\u003eSupplementary Fig. 7\u003c/strong\u003e).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTopographic network organization in ASD\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eHaving identified the transcriptomic signatures of the topographic network for the neurocognitive processes associated with the phenotypic heterogeneity of ASD, we next hypothesized that the functional organization occurs in the topographic network in individuals with ASD. To solve this issue,\u0026nbsp;we tested the network similarity using a data-driven approach in which we used the blood-oxygen-level-dependent (BOLD) time series from each parcel to train an SVM model for achieving six network classification tasks (\u003cstrong\u003eFig. 6a\u003c/strong\u003e). This was motivated by the observation that brain regions with similar features of the BOLD time series are more likely to support similar functions. For instance, the BOLD time series within the default mode network has a similar feature, such as the longest temporal autocorrelation delay, while those in the early visual network show the shortest temporal autocorrelation delay\u003csup\u003e30\u003c/sup\u003e. Therefore, the function of a given network can shift to being closer to or farther from that of another network if functional reorganization occurs between the two networks in the brains of individuals with ASD. This is proved by our classification analysis, which shows that the classification accuracy of the networks varies between ASD patients and healthy controls. We found that the classification accuracy of a given network has no difference between ASD patients and healthy controls (frontal cortex: [T = 1.29, \u003cem\u003eP\u003c/em\u003e = 0.196], \u003cstrong\u003eFig. 6b\u003c/strong\u003e; temporal cortex: [T = 1.64, \u003cem\u003eP\u003c/em\u003e = 0.104], \u003cstrong\u003eFig. 6c\u003c/strong\u003e; auditory cortex: [T = 1.316, \u003cem\u003eP\u003c/em\u003e = 0.189], \u003cstrong\u003eFig. 6d\u003c/strong\u003e; somatosensory cortex: [T = -0.203, \u003cem\u003eP\u003c/em\u003e = 0.839], \u003cstrong\u003eFig. 6\u003c/strong\u003e\u003cstrong\u003ee\u003c/strong\u003e; posterior cortex: [T = -0.233, \u003cem\u003eP\u003c/em\u003e = 0.816], \u003cstrong\u003eFig. 6f\u003c/strong\u003e; visual cortex: [T = -1.39, \u003cem\u003eP\u003c/em\u003e = 0.167], \u003cstrong\u003eFig. 6g\u003c/strong\u003e). We further analyzed whether the given networks are more likely to classify as other networks in individuals with ASD and healthy controls. Importantly, we exhibited that the TPC is more likely to classify as FTC (T = -2.059, \u003cem\u003eP\u003c/em\u003e = 0.0409) and POC (T = -2.325, \u003cem\u003eP\u003c/em\u003e = 0.0212) in individuals with healthy controls (\u003cstrong\u003eFig. 6c\u003c/strong\u003e), indicating that functional similarities between the TPC and both the FTC and POC are closer in healthy controls than that in ASD. The TPC is more likely to classify as SMC (T = 2.064, \u003cem\u003eP\u003c/em\u003e = 0.0404) in individuals with ASD (\u003cstrong\u003eFig. 6c\u003c/strong\u003e), indicating that the functional similarity between TPC and SMC is closer in ASD than that in healthy controls. These findings suggest that there is dysregulation in the functional interactions between the TPC and both the FTC and POC in individuals with ASD, while interactions between the TPC and SMC are enhanced in individuals with ASD.\u003c/p\u003e"},{"header":"Discussion","content":"\u003cp\u003eHere, we identified a neurocognitive profile for\u0026nbsp;phenotypic heterogeneity of ASD\u0026nbsp;and further analyzed its spatial association with gene expression.\u0026nbsp;Collectively, this pattern delineates a strong spatial distribution difference among networks, especially separating TPC from other networks. The neurocognitive profiles follow cortical hierarchical organization and are significantly related to multiple cortical properties. By integrating a large-scale fMRI dataset with a weighted gene expression map, we identified a topographic network associated with phenotypic heterogeneity in ASD. Further, we uncovered how the transcriptomic signatures of this network evolve throughout neurodevelopment. Finally, we found that functional reorganization has occurred within the topographic network with the TPC as the central node in ASD. Our results directly bridge macroscale neurocognitive processes with microscale gene expression, highlighting the influence that functional reorganization underlying neurobiological mechanisms have on phenotypic heterogeneity in ASD.\u003c/p\u003e\n\u003cp\u003eThe present study was motivated by previous reports that link gene expression to structural and functional architectures in both healthy participants and ASD brains. Gene expression profiles have strong associations with cortical folding\u003csup\u003e31\u003c/sup\u003e, functional hierarchy\u003csup\u003e32\u003c/sup\u003e, and T1w/T1w ratio\u003csup\u003e33\u003c/sup\u003e in healthy human brains and have been linked to multiple microstructural features in brains with ASD\u003csup\u003e34,35\u003c/sup\u003e, potentially reflecting a hierarchical axis of microstructural and macroscale functional properties. In particular, disruptions in macroscale hierarchy affect the integration and segregation of unimodal and transmodal networks in ASD\u003csup\u003e12\u003c/sup\u003e, and further, these atypical functional connectomes can be explained by regional differences in the expression of distinct ASD-related gene sets\u003csup\u003e6\u003c/sup\u003e. While these reports exhibit a link between transcriptomic profiles and multiple functional and structural features, the neurocognitive processes reflected phenotypic heterogeneity of ASD of such trends are not well understood. Our results fill in this huge gap, showing statistically spatial associations of neurocognitive processes with cortical hierarchical properties and transcription. Collectively, spatial alignment of transcriptomic profiles and neurocognitive processes may appear to show a dominant underlying hierarchical axis, shaping\u0026nbsp;phenotypic heterogeneity in ASD.\u003c/p\u003e\n\u003cp\u003eOf note, the neurocognitive profile reveals a marked divergence in network distribution, with the TPC notably separated from other networks.\u0026nbsp;In correspondence with our result, a previous study reports\u0026nbsp;the inter-individual variability increased in ASD relative to healthy controls in default model, somatomotor, and attention networks but showed reduced variability in the lateral temporal cortex, highlighting the atypical pattern reflecting inter-individual heterogeneity in ASD\u003csup\u003e35\u003c/sup\u003e. The TPC may function as a node hub, forming dysfunctional connections with brain regions such as the prefrontal cortex, precuneus, and cuneus, thereby influencing emotion-related decision-making in ASD\u003csup\u003e36\u003c/sup\u003e.\u0026nbsp;This evidence is supported by our data-driven results, which demonstrate the TPC as a central hub facilitating functional reorganization within the FTC, SMC, and POC networks in ASD. Moreover, gray matter volume and cortical thickness decreased in the TPC may bring about its dysregulation with other brain areas\u003csup\u003e37-39\u003c/sup\u003e, ultimately resulting in failed language development and atypical object perception in ASD\u003csup\u003e40,41\u003c/sup\u003e. Thus, these findings suggest that the TPC may act as a central node to be involved in complex dysfunction in ASD.\u003c/p\u003e\n\u003cp\u003eIn the present study, we found that the neurocognitive profile has spatial correspondence with multiple cortical properties, demonstrating a cortical hierarchical axis that separates TPC from other brain networks. Patterns of gene expression spatially shape the cortical hierarchical properties, including microstructural distribution and functional organization\u003csup\u003e32,33,42\u003c/sup\u003e. In the AHBA dataset, we found that the neurocognitive profile can be explained by the transcription, showing that the weighted gene expressions (both PLS-1 and PLS-2 scores) are spatially correlated with the neurocognitive profile. Gene enrichment analysis revealed that the top 10\u003csup\u003eth\u003c/sup\u003e gene sets that shape the neurocognitive profile are enriched in ASD-related biological processes (PLS-1) and molecular processes in the formation of neural connections during neurodevelopment (PLS-2). In particular,\u0026nbsp;the weighted gene expression of PLS-1\u0026nbsp;is enriched in ASD pathogenic risk genes, unraveling that ASD risk gene sets shape the neurocognitive profile in the phenotypic heterogeneity in ASD.\u0026nbsp;We mapped the whole-genome transcription patterns to the spectrum of neurocognitive function across multiple brain areas, showing that the neurocognitive terms related to ASD behaviors, such as restricted, attention, and reaction\u003csup\u003e1\u003c/sup\u003e, contribute to the weighted gene expression of PLS-1. At the same time, the association between gene expression and behaviors has been approached from another study, delineating a ventromedial–dorsolateral axis and separating gene sets related to perceptual versus affective function\u003csup\u003e43\u003c/sup\u003e. Thus, these findings uncovered a cortical hierarchical axis, separating TPC from other brain networks, which is spatial covarying with transcription to characterize a gene-neurocognition signature of phenotypic heterogeneity in ASD.\u003c/p\u003e\n\u003cp\u003eIndeed, the gene sets in brain regions with highly weighted gene expression shape the neurocognitive profile, but whether these brain regions can characterize phenotypic heterogeneity in ASD. Using a data-driven approach combined with a large-scale fMRI dataset,\u0026nbsp;we identified a topographic network characterized by higher weighted gene expression in the somatosensory and visual cortices and lower weighted gene expression in the medial frontal and temporal cortices. All of these brain regions participate in brain pathological processes of ASD\u003csup\u003e44,45\u003c/sup\u003e. Moreover, brain regions within the topographic network can classify ASD from healthy controls and relate to ASD behaviors such as social cognition, stereotyped behaviors, and restricted interests, suggesting that the identified topographic network is involved in phenotypic heterogeneity in ASD. In line with our results, a review of the literature has reported the predictive models that phenotypic features within the brain regions that are spatially similar to the topographic network, differing across subtypes, may help triage individuals for better care management in ASD\u003csup\u003e46\u003c/sup\u003e. Therefore, how transcriptomic–neurocognitive links vary across subtypes is a key question for future studies.\u003c/p\u003e\n\u003cp\u003eGene expression profiles are involved in shaping cortical reorganization throughout neurodevelopment, comprising folding\u003csup\u003e31\u003c/sup\u003e and formation of corticocortical connection\u003csup\u003e47\u003c/sup\u003e. Using the BrainSpan dataset, we found that gene-neurocognition signature within the topographic network exhibits a spatiotemporal trajectory over neurodevelopment, gradually becoming most remarkable and differentiation between brain regions in adulthood (\u003cstrong\u003eFig. 5a\u003c/strong\u003e). This suggests that continued differentiation of neurocognitive processes during maturation gradually forms the phenotypic heterogeneity in ASD. Furthermore, we observed that functional reorganization has occurred within the topographic network with the TPC as the central node, suggesting that functional dysregulation within this network contributes to the phenotypic heterogeneity in ASD. Consequently, these findings suggest that gene expression shapes functional reorganization and ultimately forms the neurocognitive profile reflecting phenotypic heterogeneity in ASD.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eSeveral methodological factors may limit the scope of our findings. First, the main analysis is based on small samples of postmortem brains, potentially limiting the generalizability of the results. Thus,\u0026nbsp;more\u0026nbsp;comprehensive microarray gene expression datasets are necessary for future studies to ensure a more reliable understanding of the observed phenomena. Further, spatiotemporal trajectories of gene-neurocognition signature were analyzed based on BrainSpan data, which are limited by small cortical genetic samples. Finally, subcortical topographic changes were well documented in previous studies\u003csup\u003e48,49\u003c/sup\u003e. How gene-neurocognition signature shapes subcortical profiles of ASD should be investigated in future work.\u003c/p\u003e\n\u003cp\u003eIn summary, we demonstrated a neurocognitive profile related to phenotypic heterogeneity in ASD, showing a cortical hierarchical axis that separates the TPC from other networks. Alignment the patterns of gene expression with neurocognitive profile, we showed that gene-neurocognition signature shapes the neurocognitive processes and the ASD pathogenic risk, ultimately manifesting as a topographic network that gradually differentiates during neurodevelopment. Finally, we observed that functional reorganization has occurred within the topographic network with the TPC as the central node. Collectively, these results highlight that genomic-neurocognition shapes the abnormal functional reorganization to form a neurocognitive profile related to phenotypic heterogeneity in ASD.\u003c/p\u003e"},{"header":"Methods","content":"\u003cp\u003e\u003cstrong\u003eNeurocognitive profiles\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe meta-analytical functional association maps, of which probabilistic measures reflect the activation of specifical brain function, were obtained from Neurosynth\u003csup\u003e15\u003c/sup\u003e (https://neurosynth.org/analyses/terms/), a tool that synthesizes fMRI results from more than 15000 studies. Although more than 1000 terms are reported in Neurosynth, we focus primarily on neurocognitive processes related to phenotypic heterogeneity of ASD based on our comprehensive literature review\u003csup\u003e1,2,5,7\u003c/sup\u003e and therefore limit the terms of interests cognitive and behavioral terms. We used 40 terms, ranging from specific behaviors (reaction, repetition, and restricted) to specific cognitive processes (attention and cognitive control) and emotional states (anxiety and negative emotions). All functional association maps were parcellated into 360 bilateral hemisphere cortical regions using HCP_MMP1.0\u003csup\u003e25\u003c/sup\u003e. Then, we applied the PCA\u003csup\u003e50\u003c/sup\u003e to determine the dominant component of spatial variation across these functional association maps, thereby defining the dominant components as neurocognitive profiles for ASD. Moreover, we also employed the Laplacian eigenmaps\u003csup\u003e51\u003c/sup\u003e to reproduce the neurocognitive profiles for validation analysis. \u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eHCP dataset\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis study included data from 1206 healthy young adults (age range, 22 to 37 years) from the publicly released HCP Young Adult dataset\u003csup\u003e52\u003c/sup\u003e. The subjects who met the following criteria, such as a history of psychiatric disorder, substance abuse, cardiovascular disease, or any other severe health conditions, were excluded from the present study. Recruitment, data acquisition, and written consent are described in detail at https://www.humanconnectome.org/study/hcp-young-adult and in other publications\u003csup\u003e52\u003c/sup\u003e. \u003c/p\u003e\n\u003cp\u003eThe subjects who did not include high-resolution cortical T1w/T2w maps (0.7 mm\u003csup\u003e3\u003c/sup\u003e isotropic) were removed from the current study. Finally, 1113 subjects who contained T1w/T2w maps remained in the following analyses. Subsequently, T1w/T2w maps were averaged across subjects, and the averaged T1w/T2w map was projected on the left hemisphere with 180 areas using HCP multimodal parcellation. \u003c/p\u003e\n\u003cp\u003eThe minimally processed resting-state fMRI data of 1084 participants (age range, 22 to 37 years) from HCP were included for functional gradient analysis. The images were further processed to regress out nuisance variables from head movements and physiological noise using Analysis of Functional Neuroimages software (AFNI, https://afni.nimh.nih.gov/) and then nonlinearly registered to the Montreal MNI_ICBM152 standard space. The time series of fMRI data were averaged within each parcel using the HCP_MMP1.0 template, and Pearson correlation analysis was performed between each parcel to construct an FC matrix for each participant. Finally, the FC matrices were averaged across all participants to generate the averaged FC matrix.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eResting-state fMRI dataset of ASD and data preprocessing\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eOriginal fMRI and demographic data were obtained from the open-access data set, Autism Brain Image Data Exchange I (https://fcon_1000.projects.nitrc.org/indi/abide/abide_I.html). A total of 79 ASD patients (age range, 7 to 39 years) and 105 healthy controls (age range, 7 to 31 years) from NYU Langone Medical Center were included in the present study. The clinical assessment scores for the Autism Diagnostic Observation Schedule (ADOS), Social Communication Questionnaire (SCQ), and Social Responsiveness Scale (SRS) were collected. The detailed MRI scanning parameters and demographic information are available at https://fcon_1000.projects.nitrc.org/indi/abide/abide_I.html. \u003c/p\u003e\n\u003cp\u003ePreprocessing of the functional images involved the following steps: (1) discarding the first 10 time points of functional images to account for magnetic stabilization; (2) slice timing correction to compensate for temporal shifts of different slices; (3) within-subject fMRI image realignment to estimate and spatially correct for head motions of different volumes; (4) rigid-body T1 image registration to the functional mean image, then the T1 image normalized to the Montreal MNI_ICBM152 standard space to generate a transformation matrix, and finally aligned the fMRI image to the MNI_ICBM152 standard space using the transformation matrix; (5) functional image resampling to 3\u0026times;3\u0026times;3 mm\u003csup\u003e3\u003c/sup\u003e voxel size followed by regressing out confounding signals, including linear trends, white matter (WM), cerebrospinal fluid (CSF), and head motion parameters. All data preprocessing was performed on Analysis of Functional Neuroimages software (AFNI, https://afni.nimh.nih.gov/).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunctional gradient analysis\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eFunctional gradient analysis was performed using the Brainspace toolbox (https://github.com/MICA-MNI/BrainSpace). The averaged FC matrix from HCP was threshed to retain the top 10% connections of each node, and the cosine similarity between each pair of nodes was computed. Furthermore, the similarity matrix was scaled into a normalized angle matrix to avoid negative values\u003csup\u003e22\u003c/sup\u003e. The diffusion map embedding approach was finally applied to identify gradient components that explain most functional connectome variances. Following the previous recommendation, we set the manifold learning parameter \u0026alpha; = 0.5 in the diffusion process\u003csup\u003e53\u003c/sup\u003e. Finally, the gradient map was projected to the 360 bilateral hemisphere cortical regions using HCP_MMP1.0.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eGeodesic distance of cortex\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eWe calculated the Euclidean distance between each cortical parcel from the HCP_MMP1.0 atlas and the early visual cortex. The coordinates of each cortical parcel were acquired from the website at https://neuroimaging-core-docs.readthedocs.io/en/latest/pages/atlases.html. Then, the geodesic distance of each parcel was acquired to form a geodesic distance map. \u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eGene expression data and preprocessing\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eWe utilized left hemisphere microarray-based gene expression data from the Allen Human Brain Atlas\u003csup\u003e19,20\u003c/sup\u003e (AHBA) (http://human.brain-map.org, RRID: SCR_007416). The microarray gene expression data were obtained from six donors (mean age: 42.5 years, five males and one female), including two complete brains and four left hemispheres. None of the donors had a known history of neuropsychiatric or neurological conditions. Exclusion criteria included brain injury or disease, epilepsy, drug/alcohol dependency, \u0026gt; 1 hour on the ventilator, positive for infectious disease, prion disease, chronic renal failure, cancer deaths, brain cancer, and time since death \u0026gt; 30 hours. The gene expression of each sample from all donors was quantified across 58692 probes, resulting in 20,737 gene expression levels per sample. The tissue samples were also spatially registered to the Montreal Neurological Institute (MNI) coordinate space, and the locations of each sample were recorded with MNI coordinates. The gene expression data of the brain samples were preprocessed using an AHBA processing pipeline\u003csup\u003e27\u003c/sup\u003e (https://github.com/BMHLab/AHBAprocessing) with the recommended default setting. Specifically, probe-to-gene reannotation was performed using the latest sequencing database, and probes with values that did not exceed the background noise were filtered. When multiple probes for a gene were available, the probe with the highest correlation with the RNA-seq expression data was selected. Next, each sample was assigned to its nearest cortical parcel of HCP_MMP1.0 parcellation (left hemisphere) with 180 parcels. These procedures yielded 1290 brain tissue samples covering 176 regions within the left cortex, with each sample containing the expression data of 10,027 genes.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003ePLS analysis\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003ePLS regression analysis\u003csup\u003e21\u003c/sup\u003e was applied to investigate the spatial relationship between gene expression and neurocognitive profiles. PLS analysis, an unsupervised multivariate statistical technique, decomposes relationships between predictor variables and response variables into orthogonal sets of latent variables with maximum covariance. These latent variables are a linear combination of orthogonal variables. We first aligned the gene expression data (10027 genes) and neurocognitive profile map to the HCP_MMP1.0 atlas\u003csup\u003e25\u003c/sup\u003e. The gene expression data and neurocognitive profile map were used as the predictor variables and the response variables, respectively\u003csup\u003e54\u003c/sup\u003e. The rows (brain regions) of the neurocognitive profile matrix were randomly selected and replaced 10000 times\u003csup\u003e43\u003c/sup\u003e. PLS analysis was reperformed using a new bootstrapped neurocognitive profile matrix to generate a null distribution of the ratio of variance explained, ensuring that the PLS component was significantly greater than expected by chance. For each significant component, a bootstrapping method was employed to evaluate the estimation error associated with the weight of each gene. The weight of each gene was then divided by the estimated error to derive the adjusted weight\u003csup\u003e54\u003c/sup\u003e. Genes were ranked based on their corrected weights, reflecting their contributions to the PLS regression components.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eGene enrichment analysis \u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe top 10th percentile of 10027 ranked genes was applied to a Metascape analysis tool (https://metascape.org/gp/index.html#/main/step1) to uncover biological processes enriched in the list of genes\u003csup\u003e55\u003c/sup\u003e. The top 10 percent of PLS genes were input to the Metascape website, and the obtained enrichment pathways were thresholded for significance at 5%, corrected by the false discovery rate (FDR) approach.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eSpecificity analysis\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eSpecificity analysis was used to assess whether known psychiatric risk genes, such as bipolar disorder, depression, autism, schizophrenia, and intellectual disability genes, were enriched in the PLS components\u003csup\u003e56\u003c/sup\u003e. The disorder-related risk genes provided by the AHBA (https://help.brain-map.org/display/humanbrain/Documentation).\u003c/p\u003e\n\u003cp\u003eWe calculated the enrichment ratio (ER) for each PLS component. The ER is defined as the difference between the mean bootstrap weight of the candidate gene and the mean bootstrap weight of the same number of randomly permuted genes, which was further divided by the standard deviation weight of the permuted genes. Significance was determined by the percentile of the bootstrap weight of the candidate genes relative to the bootstrap weights of randomly selected genes from 10,000 permutations\u003csup\u003e56\u003c/sup\u003e. A positive/negative ER of a given condition indicates that the risk genes are expressed to a higher/lower degree relative to the baseline expression level.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eReHo analysis\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe ReHo was calculated using the DPABI toolbox\u003csup\u003e57\u003c/sup\u003e (https://rfmri.org/DPABI). The ReHo analysis was performed on the preprocessed images of ASD. Individual ReHo map was generated by calculating the Kendall coefficient concordance to measure the similarity of the BOLD time series of a given voxel and its 26 nearest neighbors in a voxel-wise way\u003csup\u003e58\u003c/sup\u003e. Then, a z-transformation was applied to the individual ReHo map to generate a normally distributed ReHo map. Finally, the normalized individual ReHo map was projected to the 360 bilateral hemisphere cortical regions using HCP_MMP1.0. \u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eSVM classification analysis\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe SVM classification analysis used a MATLAB toolbox (https://www.csie.ntu.edu.tw/~cjlin/libsvm/). Here, we opted for a linear function kernel for the SVM model and employed a grid search function to determine the optimal cost parameter (C) for the SVM hyperparameters. The performance of the SVM model was evaluated using a 10-fold cross-validation strategy, where the data were divided into ten partitions. The SVM model was trained using data from 9 partitions and tested on the remaining partition, and this process was repeated 10 times. The final performance assessment was determined by combining the results from these ten models, and accuracy was calculated based on the correct labeling assessments.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eIdentify network topographic pattern\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eHere, we set a range of thresholds (from 0 to |0.1|) with a step of 0.01 for the PLS score to identify the networks of phenotypic heterogeneity in ASD. Specifically, we began by applying a threshold to the PLS scores, which allowed us to generate a topographic network. We next selected ReHo within the topographic network as a training feature. Finally, these features feed into the SVM classification model for classifying ASD patients from healthy controls. Additionally, we randomly selected training features with the same dimensions as the topographic network to retrain the SVM model. This procedure was repeated 1000 times to generate a null distribution of accuracy. The observed accuracy was then compared with the null distribution to ascertain whether the topographic network was influenced by random effects and/or confounding factors. The topographic network can reflect neurocognitive processes of the phenotypic heterogeneity in ASD if the model can accurately distinguish ASD patients from healthy controls.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eSpatiotemporal gene expression over development \u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eWe used the PsychENCODE BrainSpan dataset\u003csup\u003e23,24\u003c/sup\u003e to calculate the spatiotemporal trajectories for each PLS component obtained in the PLS regression analysis. BrainSpan is a gene expression database of brain tissue across development (https://www.brainspan.org/static/download.html) covering the period from 8 postconception weeks to 40 years of age. The ages were binned into nine-time windows (W1 to W9), encompassing the embryonic period, fetal development, infancy, childhood, adolescence, and adulthood. Detailed information is available in a previous study\u003csup\u003e23\u003c/sup\u003e. In the current analysis, we used the gene expression data from 16 cortical regions across all nine-time windows to calculate the spatiotemporal profile. This profile is defined as the regional average of each BrainSpan gene expression level, weighted by its PLS analysis-defined weights\u003csup\u003e59\u003c/sup\u003e.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eNetwork classification analysis\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eTo identify the topographic network organization in individuals with ASD, we used the BOLD time series from each parcel to train an SVM model for achieving six network classification tasks. Specifically, we began by averaging the BOLD time series within each parcel associated with a given network. The resulting averaged time series for each parcel were then selected as features for training the SVM classification model (\u003cstrong\u003eFig. 6a\u003c/strong\u003e). We used a one-vs-one approach to classify the given network from other networks. Then, we constructed the confusion matrix to summarize the prediction results for each network. In this matrix, each row represents an instance of the actual class (i.e., an actual network), and each column represents an instance of the predicted class\u003csup\u003e60\u003c/sup\u003e (i.e., the predicted network). The diagonal elements indicate the number of points where the predicted label matches the true label, while off-diagonal elements represent mislabeled instances by the classifier (\u003cstrong\u003eFig. 6a\u003c/strong\u003e). Importantly, the given network is more likely classified as a specific network (i.e., the higher mislabeled instances in off-diagonal elements), demonstrating that the functional similarity of the given network shows closer with the specific network.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eStatistical analyses\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe differences in PCA score distribution between the TPC and other networks, such as FTC, ADC, SMC, POC, and VSC, were analyzed using a two-sample t-test. The spatial correlation between two spatial patterns using Pearson correlation analysis and the statistical significance of spatial correlation was calculated using a permutation test with 10000 times. The contribution of each neurocognitive term to weighted gene expression was calculated as the Pearson correlation between the term\u0026rsquo;s functional association map and PLS-1 or PLS-2. The classification accuracy of a given network between ASD and healthy controls was analyzed using a two-sample t-test. The relationships between ReHo and ASD behaviors were calculated using the Pearson correlation approach.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eData availability\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe meta-analytical terms used were obtained from Neurosynth (https://neurosynth.org/analyses/terms/). T1w/T2w maps and resting-state fMRI data for functional gradient analysis can be obtained online at https://www.humanconnectome.org/study/hcp-young-adult. The resting-state fMRI of ASD can be obtained from Autism Brain Image Data Exchange I (https://fcon_1000.projects.nitrc.org/indi/abide/abide_I.html). The human gene expression data used in the present study are available in the Allen Human Brain Atlas (‘Complete normalized microarray datasets, http://human.brain-map.org). The disorder-related risk genes were obtained from the Allen Human Brain Atlas and are available at https://help.brain-map.org/display/humanbrain/Documentation. BrainSpan dataset can be found at https://www.brainspan.org/static/download.html.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCode availability\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe resting-state fMRI data from the HCP were preprocessed using the HCP minimal processing pipeline (available at https://github.com/Washington-University/HCPpipelines). The resting-state fMRI data of ASD were preprocessed using Analysis of Functional Neuroimages software (AFNI, https://afni.nimh.nih.gov/). Functional gradient analysis was performed using the Brainspace toolbox (https://github.com/MICA-MNI/BrainSpace). The gene expression data of the brain samples were preprocessed using an AHBA processing pipeline (https://github.com/BMHLab/AHBAprocessing). Partial least squares regression analysis was performed using a standard pipeline (https://github.com/KirstieJane/NSPN_WhitakerVertes_PNAS2016). Gene enrichment analysis was conducted in a Metascape analysis tool (https://metascape.org/gp/index.html#/main/step1). The ReHo was calculated using the DPABI toolbox (https://rfmri.org/DPABI). The SVM classification analysis used a MATLAB toolbox (https://www.csie.ntu.edu.tw/~cjlin/libsvm/). Additional analyses were carried out using custom scripts written in MATLAB R2023b (available at https://github.com/DevlinHu/Autism-Profile).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAcknowledgements\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis research was supported by the Natural Science Research Projects of Anhui Provincial Department of Education (grant no. KJ2021A0580 to S.H., 2022AH050475 to F.L.), and Major Natural Science Research Projects of Anhui Provincial Department of Education (2024AH040143 to J.S.).\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n \u003cli\u003eLord, C.\u003cem\u003e, et al.\u003c/em\u003e Autism spectrum disorder. \u003cem\u003eNature Reviews Disease Primers\u003c/em\u003e \u003cstrong\u003e6\u003c/strong\u003e(2020).\u003c/li\u003e\n \u003cli\u003eLord, C., Elsabbagh, M., Baird, G. \u0026amp; Veenstra-Vanderweele, J. Autism spectrum disorder. \u003cem\u003eThe Lancet\u003c/em\u003e \u003cstrong\u003e392\u003c/strong\u003e, 508-520 (2018).\u003c/li\u003e\n \u003cli\u003eKohane, I.S.\u003cem\u003e, et al.\u003c/em\u003e The co-morbidity burden of children and young adults with autism spectrum disorders. \u003cem\u003ePloS one\u003c/em\u003e \u003cstrong\u003e7\u003c/strong\u003e, e33224 (2012).\u003c/li\u003e\n \u003cli\u003eLombardo, M.V., Lai, M.-C. \u0026amp; Baron-Cohen, S. Big data approaches to decomposing heterogeneity across the autism spectrum. \u003cem\u003eMolecular psychiatry\u003c/em\u003e \u003cstrong\u003e24\u003c/strong\u003e, 1435-1450 (2019).\u003c/li\u003e\n \u003cli\u003eJeste, S.S. \u0026amp; Geschwind, D.H. Disentangling the heterogeneity of autism spectrum disorder through genetic findings. \u003cem\u003eNature Reviews Neurology\u003c/em\u003e \u003cstrong\u003e10\u003c/strong\u003e, 74-81 (2014).\u003c/li\u003e\n \u003cli\u003eBuch, A.M.\u003cem\u003e, et al.\u003c/em\u003e Molecular and network-level mechanisms explaining individual differences in autism spectrum disorder. \u003cem\u003eNature Neuroscience\u003c/em\u003e \u003cstrong\u003e26\u003c/strong\u003e, 650-663 (2023).\u003c/li\u003e\n \u003cli\u003eMasi, A., DeMayo, M.M., Glozier, N. \u0026amp; Guastella, A.J. An Overview of Autism Spectrum Disorder, Heterogeneity and Treatment Options. \u003cem\u003eNeuroscience Bulletin\u003c/em\u003e \u003cstrong\u003e33\u003c/strong\u003e, 183-193 (2017).\u003c/li\u003e\n \u003cli\u003eKana, R.K. Sentence comprehension in autism: thinking in pictures with decreased functional connectivity. \u003cem\u003eBrain\u003c/em\u003e \u003cstrong\u003e129\u003c/strong\u003e, 2484-2493 (2006).\u003c/li\u003e\n \u003cli\u003eKoyama, M.S.\u003cem\u003e, et al.\u003c/em\u003e Resting-State Functional Connectivity Indexes Reading Competence in Children and Adults. \u003cem\u003eJournal of Neuroscience\u003c/em\u003e \u003cstrong\u003e31\u003c/strong\u003e, 8617-8624 (2011).\u003c/li\u003e\n \u003cli\u003eGreen, S.A., Hernandez, L., Bookheimer, S.Y. \u0026amp; Dapretto, M. Salience network connectivity in autism is related to brain and behavioral markers of sensory overresponsivity. \u003cem\u003eJournal of the American Academy of Child \u0026amp; Adolescent Psychiatry\u003c/em\u003e \u003cstrong\u003e55\u003c/strong\u003e, 618-626. e611 (2016).\u003c/li\u003e\n \u003cli\u003eDi Martino, A.\u003cem\u003e, et al.\u003c/em\u003e The autism brain imaging data exchange: towards a large-scale evaluation of the intrinsic brain architecture in autism. \u003cem\u003eMolecular psychiatry\u003c/em\u003e \u003cstrong\u003e19\u003c/strong\u003e, 659-667 (2014).\u003c/li\u003e\n \u003cli\u003eHong, S.-J.\u003cem\u003e, et al.\u003c/em\u003e Atypical functional connectome hierarchy in autism. \u003cem\u003eNature Communications\u003c/em\u003e \u003cstrong\u003e10\u003c/strong\u003e(2019).\u003c/li\u003e\n \u003cli\u003eBernhardt, B.C., Milham, M.P., Di Martino, A., Valk, S.L. \u0026amp; Hong, S.-J. Multidimensional Neuroanatomical Subtyping of Autism Spectrum Disorder. \u003cem\u003eCerebral Cortex\u003c/em\u003e \u003cstrong\u003e28\u003c/strong\u003e, 3578-3588 (2018).\u003c/li\u003e\n \u003cli\u003eKim, J.\u003cem\u003e, et al.\u003c/em\u003e Low-dielectric-constant polyimide aerogel composite films with low water uptake. \u003cem\u003ePolymer Journal\u003c/em\u003e \u003cstrong\u003e48\u003c/strong\u003e, 829-834 (2016).\u003c/li\u003e\n \u003cli\u003eYarkoni, T., Poldrack, R.A., Nichols, T.E., Van Essen, D.C. \u0026amp; Wager, T.D. Large-scale automated synthesis of human functional neuroimaging data. \u003cem\u003eNature Methods\u003c/em\u003e \u003cstrong\u003e8\u003c/strong\u003e, 665-670 (2011).\u003c/li\u003e\n \u003cli\u003eDock\u0026egrave;s, J.\u003cem\u003e, et al.\u003c/em\u003e NeuroQuery, comprehensive meta-analysis of human brain mapping. \u003cem\u003eelife\u003c/em\u003e \u003cstrong\u003e9\u003c/strong\u003e, e53385 (2020).\u003c/li\u003e\n \u003cli\u003eGorgolewski, K.J.\u003cem\u003e, et al.\u003c/em\u003e NeuroVault. org: a web-based repository for collecting and sharing unthresholded statistical maps of the human brain. \u003cem\u003eFrontiers in neuroinformatics\u003c/em\u003e \u003cstrong\u003e9\u003c/strong\u003e, 8 (2015).\u003c/li\u003e\n \u003cli\u003eBurt, J.B.\u003cem\u003e, et al.\u003c/em\u003e Hierarchy of transcriptomic specialization across human cortex captured by structural neuroimaging topography. \u003cem\u003eNature neuroscience\u003c/em\u003e \u003cstrong\u003e21\u003c/strong\u003e, 1251-1259 (2018).\u003c/li\u003e\n \u003cli\u003eHawrylycz, M.\u003cem\u003e, et al.\u003c/em\u003e Canonical genetic signatures of the adult human brain. \u003cem\u003eNature neuroscience\u003c/em\u003e \u003cstrong\u003e18\u003c/strong\u003e, 1832-1844 (2015).\u003c/li\u003e\n \u003cli\u003eHawrylycz, M.J.\u003cem\u003e, et al.\u003c/em\u003e An anatomically comprehensive atlas of the adult human brain transcriptome. \u003cem\u003eNature\u003c/em\u003e \u003cstrong\u003e489\u003c/strong\u003e, 391-399 (2012).\u003c/li\u003e\n \u003cli\u003eAbdi, H. Partial least squares regression and projection on latent structure regression (PLS Regression). \u003cem\u003eWiley interdisciplinary reviews: computational statistics\u003c/em\u003e \u003cstrong\u003e2\u003c/strong\u003e, 97-106 (2010).\u003c/li\u003e\n \u003cli\u003eHu, S.\u003cem\u003e, et al.\u003c/em\u003e Structural lesions and transcriptomic specializations shape gradient perturbations in Wilson disease. \u003cem\u003eBrain Communications\u003c/em\u003e \u003cstrong\u003e6\u003c/strong\u003e(2024).\u003c/li\u003e\n \u003cli\u003eLi, M.\u003cem\u003e, et al.\u003c/em\u003e Integrative functional genomic analysis of human brain development and neuropsychiatric risks. \u003cem\u003eScience\u003c/em\u003e \u003cstrong\u003e362\u003c/strong\u003e, eaat7615 (2018).\u003c/li\u003e\n \u003cli\u003eZhu, Y.\u003cem\u003e, et al.\u003c/em\u003e Spatiotemporal transcriptomic divergence across human and macaque brain development. \u003cem\u003eScience\u003c/em\u003e \u003cstrong\u003e362\u003c/strong\u003e, eaat8077 (2018).\u003c/li\u003e\n \u003cli\u003eGlasser, M.F.\u003cem\u003e, et al.\u003c/em\u003e A multi-modal parcellation of human cerebral cortex. \u003cem\u003eNature\u003c/em\u003e \u003cstrong\u003e536\u003c/strong\u003e, 171-178 (2016).\u003c/li\u003e\n \u003cli\u003eHuntenburg, J.M., Bazin, P.L. \u0026amp; Margulies, D.S. Large-Scale Gradients in Human Cortical Organization. \u003cem\u003eTrends Cogn Sci\u003c/em\u003e \u003cstrong\u003e22\u003c/strong\u003e, 21-31 (2018).\u003c/li\u003e\n \u003cli\u003eArnatkeviciute, A., Fulcher, B.D. \u0026amp; Fornito, A. A practical guide to linking brain-wide gene expression and neuroimaging data. \u003cem\u003eNeuroimage\u003c/em\u003e \u003cstrong\u003e189\u003c/strong\u003e, 353-367 (2019).\u003c/li\u003e\n \u003cli\u003eMartin, J.\u003cem\u003e, et al.\u003c/em\u003e Biological overlap of attention-deficit/hyperactivity disorder and autism spectrum disorder: evidence from copy number variants. \u003cem\u003eJournal of the American Academy of Child \u0026amp; Adolescent Psychiatry\u003c/em\u003e \u003cstrong\u003e53\u003c/strong\u003e, 761-770. e726 (2014).\u003c/li\u003e\n \u003cli\u003eTseng, Q.\u003cem\u003e, et al.\u003c/em\u003e Spatial organization of the extracellular matrix regulates cell\u0026ndash;cell junction positioning. \u003cem\u003eProceedings of the National Academy of Sciences\u003c/em\u003e \u003cstrong\u003e109\u003c/strong\u003e, 1506-1511 (2012).\u003c/li\u003e\n \u003cli\u003eRaut, R.V., Snyder, A.Z. \u0026amp; Raichle, M.E. Hierarchical dynamics as a macroscopic organizing principle of the human brain. \u003cem\u003eProceedings of the National Academy of Sciences\u003c/em\u003e \u003cstrong\u003e117\u003c/strong\u003e, 20890-20897 (2020).\u003c/li\u003e\n \u003cli\u003eAlexander-Bloch, A.F.\u003cem\u003e, et al.\u003c/em\u003e Imaging local genetic influences on cortical folding. \u003cem\u003eProceedings of the National Academy of Sciences\u003c/em\u003e \u003cstrong\u003e117\u003c/strong\u003e, 7430-7436 (2020).\u003c/li\u003e\n \u003cli\u003eXia, Y.\u003cem\u003e, et al.\u003c/em\u003e Development of functional connectome gradients during childhood and adolescence. \u003cem\u003eScience Bulletin\u003c/em\u003e \u003cstrong\u003e67\u003c/strong\u003e, 1049-1061 (2022).\u003c/li\u003e\n \u003cli\u003eBurt, J.B.\u003cem\u003e, et al.\u003c/em\u003e Hierarchy of transcriptomic specialization across human cortex captured by structural neuroimaging topography. \u003cem\u003eNat Neurosci\u003c/em\u003e \u003cstrong\u003e21\u003c/strong\u003e, 1251-1259 (2018).\u003c/li\u003e\n \u003cli\u003eDear, R.\u003cem\u003e, et al.\u003c/em\u003e Cortical gene expression architecture links healthy neurodevelopment to the imaging, transcriptomics and genetics of autism and schizophrenia. \u003cem\u003eNature Neuroscience\u003c/em\u003e \u003cstrong\u003e27\u003c/strong\u003e, 1075-1086 (2024).\u003c/li\u003e\n \u003cli\u003eBenkarim, O.\u003cem\u003e, et al.\u003c/em\u003e Connectivity alterations in autism reflect functional idiosyncrasy. \u003cem\u003eCommunications Biology\u003c/em\u003e \u003cstrong\u003e4\u003c/strong\u003e(2021).\u003c/li\u003e\n \u003cli\u003eRolls, E.T.\u003cem\u003e, et al.\u003c/em\u003e Effective connectivity in autism. \u003cem\u003eAutism Research\u003c/em\u003e \u003cstrong\u003e13\u003c/strong\u003e, 32-44 (2019).\u003c/li\u003e\n \u003cli\u003eWallace, G.L., Dankner, N., Kenworthy, L., Giedd, J.N. \u0026amp; Martin, A. Age-related temporal and parietal cortical thinning in autism spectrum disorders. \u003cem\u003eBrain\u003c/em\u003e \u003cstrong\u003e133\u003c/strong\u003e, 3745-3754 (2010).\u003c/li\u003e\n \u003cli\u003eZilbovicius, M.\u003cem\u003e, et al.\u003c/em\u003e Temporal lobe dysfunction in childhood autism: a PET study. \u003cem\u003eAmerican Journal of Psychiatry\u003c/em\u003e \u003cstrong\u003e157\u003c/strong\u003e, 1988-1993 (2000).\u003c/li\u003e\n \u003cli\u003eMcAlonan, G.M. Mapping the brain in autism. A voxel-based MRI study of volumetric differences and intercorrelations in autism. \u003cem\u003eBrain\u003c/em\u003e \u003cstrong\u003e128\u003c/strong\u003e, 268-276 (2004).\u003c/li\u003e\n \u003cli\u003eSchultz, R.T.\u003cem\u003e, et al.\u003c/em\u003e Abnormal ventral temporal cortical activity during face discrimination among individuals with autism and Asperger syndrome. \u003cem\u003eArchives of general Psychiatry\u003c/em\u003e \u003cstrong\u003e57\u003c/strong\u003e, 331-340 (2000).\u003c/li\u003e\n \u003cli\u003eEyler, L.T., Pierce, K. \u0026amp; Courchesne, E. A failure of left temporal cortex to specialize for language is an early emerging and fundamental property of autism. \u003cem\u003eBrain\u003c/em\u003e \u003cstrong\u003e135\u003c/strong\u003e, 949-960 (2012).\u003c/li\u003e\n \u003cli\u003eTong, C.\u003cem\u003e, et al.\u003c/em\u003e Multimodal analysis demonstrating the shaping of functional gradients in the marmoset brain. \u003cem\u003eNature Communications\u003c/em\u003e \u003cstrong\u003e13\u003c/strong\u003e(2022).\u003c/li\u003e\n \u003cli\u003eHansen, J.Y.\u003cem\u003e, et al.\u003c/em\u003e Mapping gene transcription and neurocognition across human neocortex. \u003cem\u003eNat Hum Behav\u003c/em\u003e \u003cstrong\u003e5\u003c/strong\u003e, 1240-1250 (2021).\u003c/li\u003e\n \u003cli\u003eClements, C.C.\u003cem\u003e, et al.\u003c/em\u003e Evaluation of the social motivation hypothesis of autism: a systematic review and meta-analysis. \u003cem\u003eJAMA psychiatry\u003c/em\u003e \u003cstrong\u003e75\u003c/strong\u003e, 797-808 (2018).\u003c/li\u003e\n \u003cli\u003ePhilip, R.C.\u003cem\u003e, et al.\u003c/em\u003e A systematic review and meta-analysis of the fMRI investigation of autism spectrum disorders. \u003cem\u003eNeuroscience \u0026amp; Biobehavioral Reviews\u003c/em\u003e \u003cstrong\u003e36\u003c/strong\u003e, 901-942 (2012).\u003c/li\u003e\n \u003cli\u003eHorien, C.\u003cem\u003e, et al.\u003c/em\u003e Functional Connectome\u0026ndash;Based Predictive Modeling in Autism. \u003cem\u003eBiological Psychiatry\u003c/em\u003e \u003cstrong\u003e92\u003c/strong\u003e, 626-642 (2022).\u003c/li\u003e\n \u003cli\u003eV\u0026aacute;\u0026scaron;a, F.\u003cem\u003e, et al.\u003c/em\u003e Conservative and disruptive modes of adolescent change in human brain functional connectivity. \u003cem\u003eProceedings of the National Academy of Sciences\u003c/em\u003e \u003cstrong\u003e117\u003c/strong\u003e, 3248-3253 (2020).\u003c/li\u003e\n \u003cli\u003eEvans, M.M., Kim, J., Abel, T., Nickl-Jockschat, T. \u0026amp; Stevens, H.E. Developmental Disruptions of the Dorsal Striatum in Autism Spectrum Disorder. \u003cem\u003eBiological Psychiatry\u003c/em\u003e \u003cstrong\u003e95\u003c/strong\u003e, 102-111 (2024).\u003c/li\u003e\n \u003cli\u003eMurayama, C.\u003cem\u003e, et al.\u003c/em\u003e Extrastriatal dopamine D2/3 receptor binding, functional connectivity, and autism socio-communicational deficits: a PET and fMRI study. \u003cem\u003eMolecular Psychiatry\u003c/em\u003e \u003cstrong\u003e27\u003c/strong\u003e, 2106-2113 (2022).\u003c/li\u003e\n \u003cli\u003eWold, S., Esbensen, K. \u0026amp; Geladi, P. Principal component analysis. \u003cem\u003eChemometrics and intelligent laboratory systems\u003c/em\u003e \u003cstrong\u003e2\u003c/strong\u003e, 37-52 (1987).\u003c/li\u003e\n \u003cli\u003eBelkin, M. \u0026amp; Niyogi, P. Laplacian eigenmaps for dimensionality reduction and data representation. \u003cem\u003eNeural computation\u003c/em\u003e \u003cstrong\u003e15\u003c/strong\u003e, 1373-1396 (2003).\u003c/li\u003e\n \u003cli\u003eGlasser, M.F.\u003cem\u003e, et al.\u003c/em\u003e The minimal preprocessing pipelines for the Human Connectome Project. \u003cem\u003eNeuroimage\u003c/em\u003e \u003cstrong\u003e80\u003c/strong\u003e, 105-124 (2013).\u003c/li\u003e\n \u003cli\u003eVos de Wael, R.\u003cem\u003e, et al.\u003c/em\u003e BrainSpace: a toolbox for the analysis of macroscale gradients in neuroimaging and connectomics datasets. \u003cem\u003eCommunications Biology\u003c/em\u003e \u003cstrong\u003e3\u003c/strong\u003e(2020).\u003c/li\u003e\n \u003cli\u003eWhitaker, K.J.\u003cem\u003e, et al.\u003c/em\u003e Adolescence is associated with genomically patterned consolidation of the hubs of the human brain connectome. \u003cem\u003eProc Natl Acad Sci U S A\u003c/em\u003e \u003cstrong\u003e113\u003c/strong\u003e, 9105-9110 (2016).\u003c/li\u003e\n \u003cli\u003eZhou, Y.\u003cem\u003e, et al.\u003c/em\u003e Metascape provides a biologist-oriented resource for the analysis of systems-level datasets. \u003cem\u003eNature communications\u003c/em\u003e \u003cstrong\u003e10\u003c/strong\u003e, 1523 (2019).\u003c/li\u003e\n \u003cli\u003eLee, H.M.\u003cem\u003e, et al.\u003c/em\u003e Multimodal mapping of regional brain vulnerability to focal cortical dysplasia. \u003cem\u003eBrain\u003c/em\u003e \u003cstrong\u003e146\u003c/strong\u003e, 3404-3415 (2023).\u003c/li\u003e\n \u003cli\u003eYan, C.-G., Wang, X.-D., Zuo, X.-N. \u0026amp; Zang, Y.-F. DPABI: data processing \u0026amp; analysis for (resting-state) brain imaging. \u003cem\u003eNeuroinformatics\u003c/em\u003e \u003cstrong\u003e14\u003c/strong\u003e, 339-351 (2016).\u003c/li\u003e\n \u003cli\u003eZang, Y., Jiang, T., Lu, Y., He, Y. \u0026amp; Tian, L. Regional homogeneity approach to fMRI data analysis. \u003cem\u003eNeuroimage\u003c/em\u003e \u003cstrong\u003e22\u003c/strong\u003e, 394-400 (2004).\u003c/li\u003e\n \u003cli\u003eQiu, B.\u003cem\u003e, et al.\u003c/em\u003e Transcriptomic profiles link corticostriatal microarchitecture to genetics of neurodevelopment and neuropsychiatric risks. (2024).\u003c/li\u003e\n \u003cli\u003eWang, X.\u003cem\u003e, et al.\u003c/em\u003e The Brain\u0026rsquo;s Topographical Organization Shapes Dynamic Interaction Patterns That Support Flexible Behavior Based on Rules and Long-Term Knowledge. \u003cem\u003eThe Journal of Neuroscience\u003c/em\u003e \u003cstrong\u003e44\u003c/strong\u003e(2024).\u003cstrong\u003e\u003c/strong\u003e\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"","lastPublishedDoi":"10.21203/rs.3.rs-5425486/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-5425486/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eThe neurobiological basis for elaborating phenotypic heterogeneity within autism spectrum disorder (ASD) remains unknown. Applying the principal component analysis to the Neurosynth database, we established neurocognitive profiles to characterize the phenotypic heterogeneity of ASD, revealing a cortical hierarchical axis that separates the temporal cortex from other networks. By integrating neurocognitive profiles with transcriptomic data, we found that gene sets shaping the patterns of neurocognitive profiles are enriched in ASD-related biological processes and ASD pathogenic risk. Using a data-driven approach, we identified a topographic network for ASD, comprising the temporal, frontal, somatosensory, and visual cortices, with its transcriptomic signatures differentiating between regions over neurodevelopment. Additionally, functional reorganization in ASD within the topographic network has occurred with the temporal cortex as the central node. Collectively, our results reveal spatially covarying transcriptomic and neurocognitive profiles, emphasizing the influence of functional reorganization and its underlying genetic mechanism on phenotypic heterogeneity in ASD.\u003c/p\u003e","manuscriptTitle":"Genetic neurocognitive profile of autism unveiled with gene transcription","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2024-11-26 05:39:55","doi":"10.21203/rs.3.rs-5425486/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"85584151-3cd1-4bc8-a6a5-5acdb21cacbe","owner":[],"postedDate":"November 26th, 2024","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[{"id":40683025,"name":"Health sciences/Diseases/Psychiatric disorders/Autism spectrum disorders"},{"id":40683026,"name":"Biological sciences/Neuroscience/Genetics of the nervous system"}],"tags":[],"updatedAt":"2024-11-27T11:53:34+00:00","versionOfRecord":[],"versionCreatedAt":"2024-11-26 05:39:55","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-5425486","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-5425486","identity":"rs-5425486","version":["v1"]},"buildId":"qtupq5eGEP_6zYnWcrvyt","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}