The structure and function of the human brain are highly stereotyped, implying a conserved molecular program responsible for its development, cellular structure and function. We applied a correlation-based metric called differential stability to assess reproducibility of gene expression patterning across 132 structures in six individual brains, revealing mesoscale genetic organization. The genes with the highest differential stability are highly biologically relevant, with enrichment for brain-related annotations, disease associations, drug targets and literature citations. Using genes with high differential stability, we identified 32 anatomically diverse and reproducible gene expression signatures, which represent distinct cell types, intracellular components and/or associations with neurodevelopmental and neurodegenerative disorders. Genes in neuron-associated compared to non-neuronal networks showed higher preservation between human and mouse; however, many diversely patterned genes displayed marked shifts in regulation between species. Finally, highly consistent transcriptional architecture in neocortex is correlated with resting state functional connectivity, suggesting a link between conserved gene expression and functionally relevant circuitry.
Subscribe to Journal
Get full journal access for 1 year
only $18.75 per issue
All prices are NET prices.
VAT will be added later in the checkout.
Rent or Buy article
Get time limited or full article access on ReadCube.
All prices are NET prices.
Manolio, T.A. & Collins, F.S. The HapMap and genome-wide association studies in diagnosis and therapy. Annu. Rev. Med. 60, 443–456 (2009).
McCarthy, M.I. & Hirschhorn, J.N. Genome-wide association studies: potential next steps on a genetic journey. Hum. Mol. Genet. 17, R156–R165 (2008).
Hawrylycz, M.J. et al. An anatomically comprehensive atlas of the adult human brain transcriptome. Nature 489, 391–399 (2012).
Kang, H.J. et al. Spatio-temporal transcriptome of the human brain. Nature 478, 483–489 (2011).
Oldham, M.C. et al. Functional organization of the transcriptome in human brain. Nat. Neurosci. 11, 1271–1282 (2008).
Roth, R.B. et al. Gene expression analyses reveal molecular relationships among 20 regions of the human CNS. Neurogenetics 7, 67–80 (2006).
Voineagu, I. et al. Transcriptomic analysis of autistic brain reveals convergent molecular pathology. Nature 474, 380–384 (2011).
Shaw, G.T., Shih, E.S., Chen, C.H. & Hwang, M.J. Preservation of ranking order in the expression of human Housekeeping genes. PLoS ONE 6, e29314 (2011).
Langfelder, P. & Horvath, S. WGCNA: an R package for weighted correlation network analysis. BMC Bioinformatics 9, 559 (2008).
Zhang, B. & Horvath, S. A general framework for weighted gene co-expression network analysis. Stat. Appl. Genet. Mol. Biol. 4, 17 (2005).
Miller, J.A. et al. Transcriptional landscape of the prenatal human brain. Nature 508, 199–206 (2014).
Oldham, M.C., Horvath, S. & Geschwind, D.H. Conservation and evolution of gene coexpression networks in human and chimpanzee brains. Proc. Natl. Acad. Sci. USA 103, 17973–17978 (2006).
Ponomarev, I., Wang, S., Zhang, L., Harris, R.A. & Mayfield, R.D. Gene coexpression networks in human brain identify epigenetic modifications in alcohol dependence. J. Neurosci. 32, 1884–1897 (2012).
Bernard, A. et al. Transcriptional architecture of the primate neocortex. Neuron 73, 1083–1099 (2012).
Bray, N.J. et al. Screening the human protocadherin 8 (PCDH8) gene in schizophrenia. Genes Brain Behav. 1, 187–191 (2002).
Geschwind, D.H. & Konopka, G. Neuroscience in the era of functional genomics and systems biology. Nature 461, 908–915 (2009).
Konopka, G. et al. Human-specific transcriptional networks in the brain. Neuron 75, 601–617 (2012).
Hanashima, C., Fernandes, M., Hebert, J.M. & Fishell, G. The role of Foxg1 and dorsal midline signaling in the generation of Cajal-Retzius subtypes. J. Neurosci. 27, 11103–11111 (2007).
Siegenthaler, J.A. & Miller, M.W. Generation of Cajal-Retzius neurons in mouse forebrain is regulated by transforming growth factor beta-Fox signaling pathways. Dev. Biol. 313, 35–46 (2008).
Zapala, M.A. et al. Adult mouse brain gene expression patterns bear an embryologic imprint. Proc. Natl. Acad. Sci. USA 102, 10357–10362 (2005).
Nelson, T.H. et al. Autworks: a cross-disease network biology application for autism and related disorders. BMC Med. Genomics 5, 56 (2012).
Langfelder, P., Luo, R., Oldham, M. & Horvath, S. Is my network module preserved and reproducible? PLoS Comput. Biol. 7, e1001057 (2011).
Cahoy, J. et al. A transcriptome database for astrocytes, neurons, and oligodendrocytes: a new resource for understanding brain development and function. J. Neurosci. 28, 264–278 (2008).
Chen, J., Bardes, E.E., Aronow, B.J. & Jegga, A.G. ToppGene Suite for gene list enrichment analysis and candidate gene prioritization. Nucleic Acids Res. 37, W305–W311 (2009).
Gratacos, M. et al. Identification of new putative susceptibility genes for several psychiatric disorders by association analysis of regulatory and non-synonymous SNPs of 306 genes involved in neurotransmission and neurodevelopment. Am. J. Med. Genet. B Neuropsychiatr. Genet. 150B, 808–816 (2009).
Pinheiro, A.P. et al. Association study of 182 candidate genes in anorexia nervosa. Am. J. Med. Genet. B Neuropsychiatr. Genet. 153B, 1070–1080 (2010).
Lein, E.S. et al. Genome-wide atlas of gene expression in the adult mouse brain. Nature 445, 168–176 (2007).
Richiardi, J. et al. Correlated gene expression supports synchronous activity in brain networks. Science 348, 1241–1244 (2015).
Van Essen, D.C. et al. The WU-Minn Human Connectome Project: an overview. Neuroimage 80, 62–79 (2013).
Smith, S.M. et al. Functional connectomics from resting-state fMRI. Trends Cogn. Sci. 17, 666–682 (2013).
Yeo, B.T. et al. The organization of the human cerebral cortex estimated by intrinsic functional connectivity. J. Neurophysiol. 106, 1125–1165 (2011).
De Bacco, F. et al. The MET oncogene is a functional marker of a glioblastoma stem cell subtype. Cancer Res. 72, 4537–4550 (2012).
Hosokawa, M., Klegeris, A., Maguire, J. & McGeer, P.L. Expression of complement messenger RNAs and proteins by human oligodendroglial cells. Glia 42, 417–423 (2003).
Rosenmann, H. et al. A polymorphism in the complement component C1r is not associated with sporadic Alzheimer's disease. Neurosci. Lett. 336, 101–104 (2003).
Walker, D.G., Dalsing-Hernandez, J.E. & Lue, L.F. Human postmortem brain-derived cerebrovascular smooth muscle cells express all genes of the classical complement pathway: a potential mechanism for vascular damage in cerebral amyloid angiopathy and Alzheimer's disease. Microvasc. Res. 75, 411–419 (2008).
Miller, J.A., Horvath, S. & Geschwind, D.H. Divergence of human and mouse brain transcriptome highlights Alzheimer disease pathways. Proc. Natl. Acad. Sci. USA 107, 12698–12703 (2010).
Morris, J.A. et al. Divergent and nonuniform gene expression patterns in mouse brain. Proc. Natl. Acad. Sci. USA 107, 19049–19054 (2010).
Oberheim, N.A., Wang, X., Goldman, S. & Nedergaard, M. Astrocytic complexity distinguishes the human brain. Trends Neurosci. 29, 547–553 (2006).
Zeng, H. et al. Large-scale cellular-resolution gene profiling in human neocortex reveals species-specific molecular signatures. Cell 149, 483–496 (2012).
Damoiseaux, J.S. et al. Consistent resting-state networks across healthy subjects. Proc. Natl. Acad. Sci. USA 103, 13848–13853 (2006).
Greicius, M.D., Krasnow, B., Reiss, A.L. & Menon, V. Functional connectivity in the resting brain: a network analysis of the default mode hypothesis. Proc. Natl. Acad. Sci. USA 100, 253–258 (2003).
Richiardi, J. et al. Correlated gene expression supports synchronous activity in brain networks. Science 348, 1241–1244 (2015).
French, L. & Pavlidis, P. Relationships between gene expression and brain wiring in the adult rodent brain. PLoS Comput. Biol. 7, e1001049 (2011).
French, L., Tan, P.P. & Pavlidis, P. Large-scale analysis of gene expression and connectivity in the rodent brain: insights through data integration. Front. Neuroinform. 5, 12 (2011).
Smith, S.M. et al. Resting-state fMRI in the Human Connectome Project. Neuroimage 80, 144–168 (2013).
Langfelder, P., Zhang, B. & Horvath, S. Defining clusters from a hierarchical cluster tree: the Dynamic Tree Cut package for R. Bioinformatics 24, 719–720 (2008).
Miller, J.A. et al. Strategies for aggregating gene expression data: the collapseRows R function. BMC Bioinformatics 12, 322 (2011).
Salimi-Khorshidi, G. et al. Automatic denoising of functional MRI data: combining independent component analysis and hierarchical fusion of classifiers. Neuroimage 90, 449–468 (2014).
Beckmann, C.F. & Smith, S.M. Probabilistic independent component analysis for functional magnetic resonance imaging. IEEE Trans. Med. Imaging 23, 137–152 (2004).
The authors thank the Allen Institute for Brain Science founders, Paul G. Allen and Jody Allen, for their vision, encouragement and support. Research was supported by the Allen Institute for Brain Science. We also gratefully acknowledge support from the US National Institute of Drug Abuse, grant 4R33DA027644; D. Wall of Stanford University School of Medicine; and 1U54MH091657 (NIH Blueprint for Neuroscience Research).
The authors declare no competing financial interests.
Integrated supplementary information
Hierarchical clustering of potassium channels across the brain. This gene class has higher DS than expected by chance (Mann-Whitney U, p<1.70e-12). Gene expression data are z-score normalized. A wide variety of distinct expression profiles are represented from regions representing cortex, hippocampus, striatum, thalamus, cerebellum and a variety of small marker regions including inferior olivary (IO). Structural ordering same as Fig. 1B.
Transcriptional distinctness and topographic relationships between brain regions based on multidimensional scaling (MDS) under decreasing DS gene sets (size n=1735) at 10th (A), 25th (B), 50th (C), 75th (D) percentile. Anatomic structural separation becomes less prominent for lower DS ranked genes in each of the quartiles. Legend and scales are identical for all four panels.
(A) Maximum correlation to a module eigengene (M1-M32) is plotted for each gene in descending DS, with Lowess fit f=1/3 (www.r-project.org) shown in black. As high DS genes are used in the network construction it is expected that the vast majority of these genes (96.5%, g=8,372) will be highly correlated (ρ>0.4) to one of 32 core gene expression networks. However, a much larger class of genes (90.1%, 15,627/17,348) are well correlated with the 32 modules at ρ>0.4. (B) To measure the reproducibility of modules we performed a cross-validation study in which 6 consensus networks were constructed from all possible subsets of 5 brains, leaving a different one out in each network (e.g. network O1 omits brain 1). Using the same fully automated procedure described in the text, 33, 36, 38, 36, 39, and 36 modules were identified in networks O1-O6, respectively, consistent with, but slightly higher than, the 32 modules identified in the 6-brain network. Average module eigengene (MEs) were found in each network by independently calculating the MEs in each brain using the assigned genes and then averaging across the 5 brains. These vectors were then correlated with the corresponding patterns found in the held out brain as a consistency check. Overall, the patterns found in the held out brain largely agree with average MEs from the 5 brain networks (mean ρ=0.81, 0.86, 0.85, 0.87, 0.85, and 0.86 in O1-O6, respectively). The essential patterning of the gene set in the held out brain is preserved as 98.7% of comparisons have high correlation (ρ> 0.6).
(A) Functional interaction of canonical modules based on common GO gene ontology categories (Molecular Function, Biological Process, Cellular Component), KEGG and Reactome pathways. (Database details: https://toppgene.cchmc.org/navigation/database.jspwww.Methods). Size of colored squares between modules scales with number of common interactions. Notable common shared pathways and interactions are labeled. Neuronal processes, cell type and function, including neurogenesis and neurodevelopment are dominated by M1-M5. The dopaminergic pathway, from biosynthesis to storage and release is represented by the pair M10, M12. Other notable pathways include M19, M27, and M29 involved in mitochondrial and ribosomal function, M21 and M25 in immune response, and M30 and M32 for glial processes. The largest set of common pathways are shared by M21 and M25 centering on immune and defense response (1.92e-05, 5.85e-34) respectively and immune system development (5.98e-08, 8.60e-04). (B) The degree of the each module as a node in the graph of A. Modules having more specific neuronal or glial content have the highest degree of functionality connectivity with other modules, e.g. M4 and M25. The modules with mixed cellular content (M13-M18) including highly specific anatomic modules for pontine/arcuate nuceli, dentate gyrus, hypothalamus, and cerebellum share few functional annotations with other modules. The diameter of the graph, indicating the maximum path connecting any two modules in terms of common annotations, is 3. (C) By comparing this network with random networks of same size, neuronal modules M1-M15 are seen to substantially interact, and to a weaker extent M16-M32, with significantly fewer common interactions than by chance associations (see Supplementary Analysis). Interaction matrix indicates shared annotations that exceed or are fewer than expected by chance, scored as z-score with binomial test. See Table 1 and Supplementary Table 8 for detailed module annotation.
(Bottom row, each panel) The consensus module eigengene expression for modules M1-M8 are shown with standard error of the mean across 6 brains, with prominent structures and a representative hub gene of high correlation to the pattern listed above the bar graph. All bar graphs are on the same scale. (Top row, each panel) Anatomic visualization of each module is shown, where red represents higher expression and blue represents lower expression. Anatomical plots for all genes available through the Allen Human Brain Atlas are available as part of the resource.
(Bottom row, each panel) The consensus module eigengene expression for modules M9-M16 are shown with standard error of the mean across 6 brains, with prominent structures and a representative hub gene of high correlation to the pattern listed above the bar graph. All bar graphs except M16 are on the same scale. (Top row, each panel) Anatomic visualization of each module is shown, where red represents higher expression and blue represents lower expression. Anatomical plots for all genes available through the Allen Human Brain Atlas are available as part of the resource.
(Bottom row, each panel) The consensus module eigengene expression for modules M17-M24 are shown with standard error of the mean across 6 brains, with prominent structures and a representative hub gene of high correlation to the pattern listed above the bar graph. All bar graphs except M22 are on the same scale. (Top row, each panel) Anatomic visualization of each module is shown, where red represents higher expression and blue represents lower expression. Anatomical plots for all genes available through the Allen Human Brain Atlas are available as part of the resource.
(Bottom row, each panel) The consensus module eigengene expression for modules M25-M32 are shown with standard error of the mean across 6 brains, with prominent structures and a representative hub gene of high correlation to the pattern listed above the bar graph. All bar graphs except M26, M27, and M31 are on the same scale. (Top row, each panel) Anatomic visualization of each module is shown, where red represents higher expression and blue represents lower expression. Anatomical plots for all genes available through the Allen Human Brain Atlas are available as part of the resource.
Counts of the number of uniquely associated ontology and functional terms by module for major categories of gene ontology, gene family, pathway, and phenotype. The detailed terms and p-values can be found in Supplementary Table 8.
Supplementary Figures 1–9, Supplementary Table 1 and Supplementary Analysis (PDF 4608 kb)
Supplementary Table 1: Neuroanatomical sampling overview of the Allen Human Brain Atlas and of each analysis in the manuscript.
A hierarchical ontology spanning all major architectural subdivisions was created to support the microarray sampling strategy. Each structure in this tree is designated a specific RGB color used throughout this paper. The number of samples isolated from each brain from brain region is shown in the first worksheet, along with a summary of the specific subdivisions sampled and the sample isolation method used (Macro = scalpel macrodissection, LMD = laser microdissection). The complete hierarchical ontology and fine structure sampling for each brain is provided in the second worksheet, collapsed down to a single column (complete version available at the Allen Brain Atlas data portal, or can be reconstructed using the “structure ID” and “parent structure ID” columsn). The table contains the structure ID, acronym, hemisphere, color hex triplet, and the number of samples for that structures in brains 1-6. Asterisks (*) in the “structure acronym” column indicate the 96 brain regions that were sampled sufficiently to be included in the analysis in Figure 6. Additionally, the “subregion for analysis” column specifies the portions of the ontology which were averaged together to form the 132 broad brain structures in the analyses in Figures 2–5. Asterisks (*) in this column indicate the 65 brain regions sampled across all six brains that were shown in visualizations (** note: CPLV, Pa, and CGS were present in fewer than 6 samples but were also shown in the visualizations to highlight ependymal regions in module M26). The third worksheet shows which mouse structures were matched with each human structures for comparison between species (Figure 6). Structures listed in the “Mouse counterparts” column are structure names from the ontology of the Allen Mouse Brain Atlas. A red “none” indicates that the listed human region did not have comparable structures in mouse. Note that, although cc and RaM had comparable mouse counterparts, ISH quantification was not available for these structures. (PDF 1924 kb)
Several different metrics of differential stability are provided for all 17,348 genes included in the analysis. Genes and the corresponding probes are listed and ordered descending by Pearson correlation (this is the metric used in the manuscript). Alternative metrics include MaxDiff (the maximum occurring differential between pairs of structures), Tau (the average Kendall Tau correlation across regions for each pair of brains), Euclid (the average Euclidean distance between expression levels in pairs of brains), and AvgVar (the average across-region variability between the six brains.) (CSV 952 kb)
Supplementary Table 3: Genes with high expression but low variability are enriched for housekeeping functions.
2,236 genes have expression in the lowest quartile of variability and also have relatively low differential stability (DS<0.5). These genes are sorted descending by DS and also include their associated probes, their average log2 expression levels, and log2 standard deviations. These stable (but not differentially stable) genes are enriched for housekeeping functions such as RNA binding (p<3.26e-21), KEGG spliceosome pathway (p<6.4e-13), and mitochondrial ribosomal proteins (p<1.32e-10). (CSV 114 kb)
Supplementary Table 4: Enrichment analysis for the top 10th percentile set of genes (n=1735) ranked by differential stability.
The complete list of enrichments for the blue bars shown in Figure 2d, including significant enrichments for gene ontology categories, transcription factor binding sites, mrRNA targets, and drug targets. The first three columns indicate the category, ID, and name of the enrichment list. The next four columns show p-values from a hypergeometric test for enrichment and q-values after correction for multiple comparison, as well as the overlapping and total number of genes in the list. The final column shows all overlapping genes in each category. (CSV 2327 kb)
Supplementary Table 5: Disease enrichments for the top 10th percentile set of DS genes are primarily brain related.
The complete list of significant disease enrichments for the data shown in Figure 2f, based on 2289 gene sets from the Autworks database. Column B shows the disease tested for enrichment. The next two columns show Bonferroni corrected p-values from a hypergeometric test for enrichment, as well as the overlapping number of genes in the list. The final column shows all overlapping genes in each category. (CSV 68 kb)
Supplementary Table 6: Module assignments and eigengene correlations for each gene assigned to a module in the consensus network.
The great majority of all genes (90.1%, 15,627) are correlated with 32 modules with ME correlation > 0.4. This table includes these genes and their associated probes, their final module assignment color (column C), label (column D), and recoloring based on neuronal content (column E). The genes correlation to the corresponding ME is shown (column F) along with the DS metric (column G; reproduced from Supplementary Table 2). Finally, the initial module assignment is included (column H) to allow regeneration of the ME from the expression data. (CSV 1042 kb)
The number of genes showing at least 1.5-fold enrichment for astrocytes (column C), neurons (column D), and oligodendrocytes (column E; Ref. #23) is unevenly distributed across modules. Modules are labeled based on the percent of neuron-related genes (100 * column D / column B) in each module (see Fig. 3C). The number of genes in each module associated with discrete cellular subcompartments was also determined (columns F-O; Foster, LJ, de Hoog, CL, Zhang, Y, Zhang, Y, Xie, X, Mootha, VK, Mann, M. 2006. A Mammalian Organelle Map by Protein Correlation Profiling. Cell 125-1: 187-199); however, these data were not used in the manuscript. (CSV 1 kb)
The complete list of enrichments for the heatmap shown in Figure 3e. We used the ToppGene portal to identify significant enrichments in gene ontology, pathways, cytoband, disease association, transcription factor binding sites, micro RNAs, drug targets, and protein-protein interactions. The first three columns indicate the category, ID, and name of the enrichment list. The first three columns indicate the category, ID, and name of the enrichment list. The next four columns show p-values from a hypergeometric test for enrichment and q-values after correction for multiple comparison, as well as the overlapping and total number of genes in the list. Column H shows the module being tested. The final column shows all overlapping genes in each category. All enrichments shown have an FDR corrected q-value < 0.05. (CSV 3868 kb)
Supplementary Table 9: Enrichment analysis for the 302 genes with high DS that are not assigned to any module.
The complete list of enrichments for the heatmap shown in Figure 3e. The “Singular Genes” tab lists the genes and their differential stability metric, while the “Annotation” tab lists the enrichments. We used the ToppGene portal to identify significant enrichments in gene ontology, pathways, transcription factor binding sites, and other categories. The first four columns indicate the category, ID, and name of the enrichment list, as well as the source of the category. The first three columns indicate the category, ID, and name of the enrichment list. The next four columns show p-values from a hypergeometric test for enrichment and q-values after correction for multiple comparison, as well as the overlapping and total number of genes in the list. The final column shows all overlapping genes in each category. All enrichments shown have a B&H FDR corrected q-value < 0.01. (XLSX 91 kb)
All 2,651 genes with reliable expression patterns in both mouse and human data sets (Methods), as well as their corresponding human probes are shown. Column C indicates whether the pattern agrees between species (correlated to correct module eigengene in mouse with ρ > 0.4), or disagrees between species (correlated to the correct module eigengene in mouse with ρ < 0.4, but highly correlated to a different module with ρ > 0.8). All remaining genes, which cannot definitively be definitively listed as agreeing or strongly disagreeing between species, are listed as ambiguous (or uncorrelated). The final column lists the original module assignment in human. Note that unassigned genes which are correlated to any module with ρ > 0.8 in mouse are listed as disagreeing between species. (XLSX 101 kb)
Local differential stability metrics provided for all 17,348 genes included in the analysis. Genes are listed and ordered descending by Pearson correlation (metric used in the manuscript). The remaining columns list differential stability calculated using only the subregions of a particular brain regions. Note that for cortical and cerebellar regions, these metrics include the full set of subregions and not the data averaged by lobe. Figure 7 uses the DS values for cerebral cortex that are listed in column E. (CSV 2306 kb)
Supplementary Table 12: Parcel assignment from AHBA to Human Connectome functional imaging parcellation.
For each of 52 parcels of the functional connectome from the number of samples from each of the 6 AHBA brains is shown followed by the average number over 6 brains. As each AHBA parcel may belong to multiple regions, the next six columns give the number of unique AHBA parcels assigned to each functional connectome parcel. The final two columns give the size in voxels of corresponding functional connectome parcel and the percentage of total cortex voxels of that parcel. Analyses in the main manuscript are presented in for all AHBA parcels and in the Supplementary methods using unique parcels. (XLSX 3044 kb)
Zip file containing code and input required to reproduce figures. (ZIP 25557 kb)
About this article
Cite this article
Hawrylycz, M., Miller, J., Menon, V. et al. Canonical genetic signatures of the adult human brain. Nat Neurosci 18, 1832–1844 (2015). https://doi.org/10.1038/nn.4171
Molecular Autism (2020)
Journal of Neuroscience Methods (2020)
Cerebral Cortex (2020)
A molecular gradient along the longitudinal axis of the human hippocampus informs large-scale behavioral systems
Nature Communications (2020)
Landscape of immune cell gene expression is unique in predominantly WHO grade 1 skull base meningiomas when compared to convexity
Scientific Reports (2020)