Brain - Methods summaryThe Brain Resource provides an overview of gene and protein expression and distribution in the human brain, integrating externally sourced and in-house generated data. Regional gene expression is analyzed using bulk RNA sequencing, while spatial transcriptomics provides a more detailed view of RNA expression across anatomical niches. Single-nucleus RNA sequencing enables the exploration of gene expression in individual neuronal and non-neuronal cell populations, while antibody-based immunohistochemistry illustrates protein distribution at the cellular and tissue level. Together, these complementary datasets enable the exploration of gene and protein expression across brain regions, anatomical structures, and cell types, as well as the identification of genes with elevated expression in specific regions or cell populations. Key publications:
What type of data can I find in the brain resourceThere is 3 main types of transcriptomics data:
How has the data been generated?Bulk RNA expression data
Figure 1. Brain regions, areas and nuclei were micro-dissected from human, pig and mouse brains. RNA was extracted using QIAGEN RNEASY lipid tissue mini kit, RIN values were determined using QIAxcel and Qubit was used to measure mRNA concentrations. Samples were enriched for mRNA using poly(A) purification or ribosomal RNA depletion. After library preparation, the quality of the libraries was assessed using Qubit and Tapestation. Samples were sequenced using the illumina and MGI sequencing platfroms (PE100 or PE150). Transcriptomics data have been based on micro-dissected areas and subregions of the human brain (n=192) and human prefrontal cortex (n=17). Human brain samples were provided by the Human Brain Tissue Bank of the Semmelweis University in Hungary. Following RNA extraction, messenger RNA was enriched using either a poly(A) purification step (Human prefrontal cortex) or ribosomal RNA clean-up (other Human samples) strategy. Prefrontal cortex samples were sequenced on the Illumina RNAseq platform, while the other brain samples were sequenced on MGI RNAseq platforms. Reads were mapped to corresponding genes in the Ensembl version used in the Human Protein Atlas. Important note, the Tissue Resource also includes brain trascriptomics data, based on 9 brain regions and the retina from GTEx, and also provides the expression overview for 14 brain regions from FANTOM. Read more about generation of brain transcriptomics data Read more about GTEx and FANTOM data Spatial RNA expression dataBrain tissues were obtained from the Netherlands Brain Bank and the Human Brain Tissue Bank of the Semmelweis University in Hungary. A 10 micrometer thick section was cut from each frozen tissue block and placed on a Stereo-seq chips and processed using the STOmics Stereo-seq Transcriptomics Set (v1.2, v1.3). Sequencing reads were mapped to the human reference genome using the SAW pipeline and counted to generate a pixel-by-gene matrix for downstream analysis (0.5 micrometer resolution). Read more about the spatial RNA expression data Single nuclei RNA expression dataThe single nucleus RNAseq (snRNA-seq) data was integrated from externally published data, generated by Siletti K et al. (2023), comprising more than 3 million nuclei sampled across the adult human brain. For the HPA Brain Resource, approximately 2.6 million nuclei were integrated and organized into 11 major brain regions, with minor filtering of low-abundance or anatomically inconsistent cell populations. The data was split by neuron and non-neuronal cells across the 11 brain regions and re-clustered and manually annotated with cell type names and cell type groups. Gene expression profiles enable comparisons across brain regions, complemented by interactive UMAP visualizations of individual nuclei. Read more about the brain single nuclei data Immunohistochemistry on tissue microarraysThe protein expression tissue micro-array data, include cerebral cortex, hippocampal formation, caudate nucleus and cerebellum, and were derived from antibody-based protein profiling using immunohistochemistry (the Tissue resource contains data on 44 human tissue types). Tissue microarrays of 1mm diameter samples were stained with primary antibodies, visualized with DAB (3,3'-diaminobenzidine) and counterstained with hematoxylin. Each brain region is represented by samples from three individuals. For selected proteins, additional brain tissues were stained, such as eye, cerebral cortex, hypothalamus, cerebellum and substantia nigra. Immunohistochemically stained sections from tissue microarrays were scanned to allow for subsequent analysis and presentation at the HPA web portal. Read more about antibody generation Read more about immunohistochemistry How has the data been analyzed?Bulk RNA expression dataFigure 2. All reads of all samples were mapped to selected Ensembl version using Kallisto pseudoalignment algorithms. Transcripts Per Million for all protein coding transcripts was calculated. Technical variation (cohorts, platfrom) and individual (donor) variation was removed using a data normalization approach. Each protein coding gene is classified based on expression in brain vs. peripheral tissues and between regions of the brain. Expression (normalized TPM) is calculated for all regions and subregions. Brain and regional elevated genes are listed on the brain region pages.
Transcript expression levels are determined by counting the reads that match each protein coding sequence. Normalized data are presented on the gene summary pages as normalized Transcripts Per Million (nTPM). Samples are grouped in 11 main regions of the central nervous system (cerebral cortex, hippocampal formation, amygdala, basal ganglia, thalamus, hypothalamus, midbrain, pons, cerebellum, medulla oblongata, spinal cord).
Read more about normalization and classification of transcriptomics data Spatial RNA expression dataIdentification of tissue niches from spatial transcriptomic data primarily relies on neuronal gene expression, reflecting both the high transcriptional diversity and abundance of neurons in the brain. Stereo-seq data was binned into lanes of fixed width and variable length, with each bin normalized to a fixed number of target counts. Sample variation was harmonized using harmony and bin data from all samples per region were clustered simultaneously using leiden clustering. Several pons samples contain cerebellum, therefore cerebellum and pons were processed and clustered together. Spatial data was compressed in hex-bins with a diameter of 50 micrometer, with each bin assigned to a tissue niche containing the sum of detected transcript counts. For each tissue niche the sum of all protein coding transcript counts were collected and normalized to counts per million (nCPM). Protein expression in tissue niches was classified using expression classification method with a enrichment factor of 2.25 for elevated expression. Read more about the spatial RNA expression data Single nuclei RNA expression dataHere, the HPA imported the expression profiles and grouped them based on the cell type- strategy (providing bar charts of pooled data representing each cell type cluster and calculating the average normalized protein-coding transcripts per million). Neuronal and non-neuronal cells were clustered, annotated and named separately, resulting in 48 different main cell types in total. The expression profile of the grouped clusters are shown for each of the 11 different brain regions. More details, related to number of million reads and number of cells per brain region/UMAP can be found here. The published cerebral cortex data is represented by a larger number of cells and only a random selection of 500 thousand cells is included in the brain resource. In total, expression data for 2674926 brain cells is displayed in the Brain single nuclei resource, for browsing the gene expression and profile easy comparison to cell type expression in peripheral tissues. Read more about the brain single nuclei data Knowledge-based annotation of protein expressionHuman brain tissue microarray images have been annotated for positivity in glial cells, neuronal cells, endothelial cells and neuropil. For cerebellum, Purkinje cells and cells in the granular and molecular layer have been annotated. What can you learn from the Brain resource?
What is the difference between the Brain representation in the Tissue Resource vs the Brain Resource?Samples were sequenced on the Illumina (human prefrontal cortex) or MGI (other human brain samples) RNAseq platforms and reads were mapped to corresponding genes in the Ensembl version used in the Human Protein Atlas. Important note, the Tissue Resource also includes brain trascriptomics data, based on 10 brain regions and the retina from GTEx, and also provides the expression overview for 14 brain regions from FANTOM. The brain expression profiles presented in the Tissue Resource may differ from those in the Brain Resource due to differences in tissue sampling, datasets, and technical processing. Nevertheless, comparing brain expression with other tissues in the body provides an important complementary perspective, helping to identify genes with elevated expression in the brain relative to peripheral tissues. Read more about generation of brain transcriptomics data Read more about GTEx and FANTOM data What is the difference between the Brain representation in the Single Nuclei Brain Resource vs the Single Cell Type Resource?Both datasets are based on data published by Siletti K et al. (2023) and are single nucleus RNA-seq data. The single nuclei data included in the brain resource includes all brain regions and cells (except for neurons cerebral cortex were limited due to high cell numbers). In this way the full regional detail is available for each of the different cell types. For the brain representation under the single cell type resource, the goal is a global comparison across the whole body, and for it to be a cell type resolved complement to the Tissue resource data. For this reason, we chose to include only 6 regions (cerebral cortex, cerebellum, midbrain, basal ganglia, hypothalamus, and hippocampus), regions that are coved by the tissue resource with bulk transcritpomics and proteomics data. Additionally, for the single cell type resource we settled for a downsampled the dataset size: From each region, we randomly sampled 20,000 neurons and 20,000 non-neuron cells. Neurons labelled by the original publishers with the cluster_id 314 (Purkinje cells), 400 (cholinergic neurons), and 724, 725 (magnocellular secretory cells) were subsequently manually added to the included neurons. The sampled datasets were then split into two: a neuron dataset and a non-neuron dataset, for clustering and naming. Read more about the generation of the brain single nuclei data Read more about the generation of the single cell type data
Data overviewDownloadable lists and data information related bulk RNAseq Downloadable lists and data information related to integrated single nuclei RNAseq
|