Immunocytochemistry/IF - cells
The Subcellular Section displays high-resolution, multicolor images of proteins labeled by indirect immunocytochemistry/immunofluorescence (ICC-IF). This provides spatial information on protein expression patterns to define the subcellular localization to cellular organelles and structures at single cell level.
Originally three cell lines, U-2 OS, A-431 and U-251 MG, originating from different human tissues were chosen to be included in the analysis of protein subcellular localization by ICC-IF. The cell line panel has since been expanded to cover more cell types and lineages, e.g. tumor cell lines from mesenchymal, epithelial and glial tumors, as well as cell lines that have immortalized by introduction of telomerase. The selection was furthermore based on morphological characteristics and widespread use of these cell lines. Information regarding sex and age of the donor, cellular origin and source is listed here. In order to localize the whole human proteome on a subcellular level in one specific cell line, most proteins are stained in U-2 OS. Two additional cell lines are selected based on mRNA expression data. In addition to the human cell lines, many proteins have been stained in the mouse cell line NIH 3T3, given that the human and mouse genes are orthologous.
The standard immunostaining protocol for ICC can be found on the open access repository for science methods at protocols.io. For the great majority of antibodies, fixation is achieved with paraformaldehyde (PFA), but for a few antibodies, this is replaced by methanol (3 times 5 minutes) in order to better preserve the morphology of certain cellular structures. For each gene, the use of PFA or methanol, as well as dilution factors for the antibodies, are stated in the Antibodies and Validation section. In order to facilitate the annotation of the subcellular localization of the protein targeted by the HPA antibody, the cells are also stained with reference markers: (i) DAPI for the nucleus, (ii) anti-tubulin antibody for microtubules, and (iii) anti-calreticulin or anti-KDEL for the endoplasmic reticulum (ER).
The resulting confocal images are single slice images representing one optical section of the cells. The microscope settings are standardized, but the detector gain is optimized for each sample. The different organelle probes are displayed as different channels in the multicolor images; the HPA antibody staining is shown in green, nucleus in blue, microtubules in red and ER in yellow.
In order to provide an interpretation of the staining patterns, all images generated by ICC-IF are manually annotated. For each cell line and antibody, the staining is described in terms of intensity, subcellular location and single-cell variability (SCV). The staining intensity is classified as negative, weak, moderate or strong based on the detector gain settings used for image acquisition in combination with the visual appearance of the image. The table below lists the subcellular locations used for annotation, with links to the cell structure dictionary entry and corresponding GO terms. SCVs within an immunofluorescence image are classified as intensity variation (variation in their expression level) or as spatial variation (variation in the spatial distribution).
The knowledge-based annotation aims to provide an interpretation of the detected subcellular localization of a protein. In the first step, stainings in different cell lines with the same antibody are reviewed and the results are compared with external experimental protein/gene characterization data for subcellular localization, available in the UniProtKB/Swiss-Prot database. In the second step, all antibodies targeting the same protein are taken in consideration for a final annotation of the subcellular distribution of the protein.
Each location is separately given one of the four reliability scores (Enhanced, Supported, Approved, or Uncertain) based on available protein/RNA/gene characterization data from both HPA and the UniProtKB/Swiss-Prot database. The reliability score also encompass several additional factors, including reproducibility of the antibody staining in different cell lines, correlation between staining intensity and RNA expression levels, and assays for enhanced antibody validation. Enhanced validation in achieved by using antibodies binding to different epitopes on the same target protein (independent antibody validation), by assessing staining intensity upon knockdown/knockout of the target protein (genetic validation) and/or by matching of the signal with a GFP-tagged protein (recombinant expression validation), and experimental evidence for subcellular location described in literature. The individual location relibility scores are summarized in an overall gene reliability score.
There are four different reliability scores:
Immunohistochemistry - tissues
The Human Protein Atlas contains images of histological sections from normal and cancer tissues obtained by immunohistochemistry. Antibodies are labeled with DAB (3,3'-diaminobenzidine) and the resulting brown staining indicates where an antibody has bound to its corresponding antigen. The section is furthermore counterstained with hematoxylin to enable visualization of microscopical features.
Tissue microarrays are used to show antibody staining in samples from 144 individuals corresponding to 44 different normal tissue types, and samples from 216 cancer patients corresponding to 20 different types of cancer (movie about tissue microarray production and immunohistochemical staining). Each sample is represented by 1 mm tissue cores, resulting in a total number of 576 images for each antibody. Normal tissues are represented by samples from three individuals each, one core per individual, except for endometrium, skin, soft tissue and stomach, which are represented by samples from six individuals each and parathyroid gland, which is represented by one sample. Protein expression is annotated in 76 different normal cell types present in these tissue samples. For cancer tissues, two cores are sampled from each individual and protein expression is annotated in tumor cells. A small fraction of the 576 images are missing for most antibodies due to technical issues. Specimens containing normal and cancer tissue have been collected and sampled from anonymized paraffin embedded material of surgical specimens, in accordance with approval from the local ethics committee.
For selected proteins extended tissue profiling is performed in addition to standard tissue microarrays. Examined tissues include mouse brain, human lactating breast, eye, thymus and extended samples of adrenal gland, skin and brain.
In order to provide an overview of protein expression patterns, all images of tissues stained by immunohistochemistry are manually annotated by a specialist followed by verification by a second specialist. Annotation of each different normal and cancer tissue is performed using fixed guidelines for classification of immunohistochemical results. Each tissue is examined for representability, and subsequently immunoreactivity in the different cell types present in normal or cancer tissues was annotated. Basic annotation parameters include an evaluation of i) staining intensity (negative, weak, moderate or strong), ii) fraction of stained cells (<25%, 25-75% or >75%) and iii) subcellular localization (nuclear and/or cytoplasmic/membranous). The manual annotation also provides two summarizing texts describing the staining pattern for each antibody in normal tissues and in cancer tissues.
Knowledge-based annotation aims to create a comprehensive overview of protein expression patterns in normal human tissues. This is achieved by stringent evaluation of immunohistochemical staining pattern, RNA-seq data from internal and external sources and available protein/gene characterization data, with special emphasis on RNA-seq. Annotated protein expression profiles are performed using single antibodies as well as independent antibodies (two or more independent antibodies directed against different, non-overlapping epitopes on the same protein). For independent antibodies, the immunohistochemical data from all the different antibodies are taken into consideration. The immunohistochemical staining pattern in normal tissues is subjectively annotated according to strict guidelines. It is based on the experienced evaluation of positive immunohistochemical signals in the 76 normal cell types analyzed. The review also takes suboptimal experimental procedures and interindividual variations into consideration.
A reliability score is manually set for all genes and indicates the level of reliability of the analyzed protein expression pattern based on knowledge-based evaluation of available RNA-seq data, protein/gene characterization data and immunohistochemical data from one or several antibodies designed towards non-overlapping sequences of the same gene. The reliability score is based on the 44 normal tissues analyzed, and is displayed on both the Tissue Atlas and the Pathology Atlas.
The reliability score is divided into Enhanced, Supported, Approved, or Uncertain. If there is available data from more than one antibody, the staining patterns of all antibodies are taken into consideration during the evaluation of the reliability score.
Immunohistochemistry/IF - mouse brain
As a complement to the immunohistochemically stained tissues, the protein atlas also includes the mouse brain atlas as a sub compartment of the normal tissue atlas. In which comprehensive profiles are available in mouse brain. A selected set of targets have been analyzed by using the antibodies in serial sections of mouse brain which covers 129 areas and subfields of the brain, several of these regions difficult to cover in the human brain. In addition pituitary, retina and trigeminal ganglions are included in recent and future image series but not annotated yet.
The tissue microarray method used within the human protein atlas enabled the global mapping of proteins in the human body, including the brain. Currently, the human tissue atlas covers four areas of the human brain: cerebral cortex, hippocampus, caudate and cerebellum. Due to the heterogeneous structure of the brain, with many nuclei and cell-types organized in complex networks, it is difficult to achieve a comprehensive overview in a 1 mm tissue sample. Analysis of more human brain samples, including smaller brain nuclei, is thus desirable in order to generate a more detailed map of protein distribution in the brain. Therefore, we here complemented the human brain atlas effort with a more comprehensive analysis of the mouse brain. A series of mouse brain sections is explored for protein expression and distribution in a large number of brain regions.
Antibodies are selected against protein involved in normal brain physiology, brain development and neuropathological processes. A limit of 60% homology (human vs mouse) is used as cut off when comparing the PrEST sequence for the antibody targets.
Selected antibodies are applied to test-sections containing brain regions or cell types with known expression based on in situ hybridization (Allen Brain Atlas) and single cell RNAseq data (Linnarsson Lab and Barres Lab). Staining patterns are evaluated based on consistency between staining patterns of multiple antibodies against the same target and match to transcriptomics data. Antibody immunoreactivity is visualized using tyramid signal amplification shown in green. A nuclear reference staining (DAPI) is visualized in blue. The immunofluorescence protocol is standardized through antibody concentration and incubation time are variable depending on protein abundance and antibody affinity determined during the test staining. The complete mouse brain profile is represented by serial coronal sections of adult mouse brain, 16 µm thick. Stained slides are then scanned and digitalized before further processing.
Table 1. Brain regions. Abbreviations are based on The Mouse Brain in Stereotaxic Coordinates, Third Edition: The coronal plates and diagrams (ISBN: 9780123742445)
The digitalized images are processed (axel-adjusted and tissue edges defined) and regions of interest (ROIs) are then marked according to the table above. These ROIs are then used for image analysis and the relative fluorescence intensity is listed for each region. The relative fluorescence is defined intensity of the annotated region relative to the intensity of the region with highest intensity.
The overview and preserved orientation in the mouse brain has enabled us to annotate additional cell classes (ependymal), glial subpopulations (microglia, oligodendrocytes, and astrocytes), and additional brain specific subcellular locations (axon, dendrite, synapse, and glia endfeet) for each investigated protein.
All images of immunofluorescence stained sections were manually annotated by specially educated personnel followed by review and verification by a second qualified member of the staff. The cellular and subcellular location of the immunoreactivity is defined and a summarizing text is provided describing the general staining pattern.
Specificity is validated by comparing the data with in situ hybridization data (Allen brain atlas) and/or available literature; support from other data leads to a supportive reliability score, while more unknown targets are viewed as uncertain and awaits further validation.
A reliability score is set for all genes and indicates the level of reliability of the analyzed protein expression pattern based on available protein/RNA/gene characterization data.
The reliability score of the antibodies in mouse brain atlas is scored as Supported or Uncertain depending on support from in situ hybridization data (Allen brain atlas) and/or previous published data, UniProtKB/Swiss-Prot database.
All purified antibodies are analyzed on antigen microarrays. The specificity profile for each antibody is determined based on the interaction with 384 different antigens including its own target. The antigens present on the arrays are consecutively exchanged in order to correspond to the next set of 384 purified antibodies. Each microarray is divided into 21 replicated subarrays, enabling the analysis of 21 antibodies simultaneously. The antibodies are detected through a fluorescently labeled secondary antibody and a dual color system is used in order to verify the presence of the spotted proteins. A specificity profile plot is generated for each antibody, where the signal from the binding to its own antigen is compared to the eventual off target interactions to all the other antigens. The vast majority (86%) of antibodies are given a pass and the remaining are failed either due to low signal or low specificity.
Western blot analysis of antibody specificity has been done using a routine sample setup composed of IgG/HSA-depleted human plasma and protein lysates from a limited number of human tissues and cell lines. Antibodies with an uncertain routine WB have been revalidated using an over-expression lysate (VERIFY Tagged Antigen(TM), OriGene Technologies, Rockville, MD) as a positive control. Antibody binding was visualized by chemiluminescence detection in a CCD-camera system using a peroxidase (HRP) labeled secondary antibody.
Antibodies included in the Human Protein Atlas have been analyzed without further efforts to optimize the procedure and therefore it cannot be excluded that certain observed binding properties are due to technical rather than biological reasons and that further optimization could result in a different outcome.
HPA RNA-seq data
In total, 69 cell lines, 51 human tissues and 18 blood cell types as well as total peripheral blood mononuclear cells (PBMC) have been analyzed by RNA-seq to estimate the transcript abundance of each protein-coding gene. Additionally, 19 mouse tissue samples and 32 pig tissue samples collected from the brain and retina of the animals were sampled and analyzed by RNA-seq.
For normal tissue and blood samples, specimens were collected with consent from patients and all samples were anonymized in accordance with approval from the local ethics committee (ref #2011/473 and ref #2015/1552-32) and Swedish rules and legislation. All tissues were collected from the Uppsala Biobank and RNA samples were extracted from frozen tissue sections. Blood samples were enriched for PBMC and granulocytes, labeled with antibodies and separated into subpopulation by flow sorting. For cell lines, early-split samples were used as duplicates and total RNA was extracted using Qiagen RNeasy mini kit. Information regarding cellular origin and source of each cell line is listed here.
For mouse tissue, samples were collected and handled in accordance with Swedish laws and regulation, and all experiments were approved by the local ethical committee (Stockholms Norra Djurförsöksetiska Nämd N183/14). The animal experiments conformed to the European Communities Council Directive (86/609/EEC), and all efforts were made to minimize the suffering and the number of animals used. WT male (n = 2) and female (n = 2) C57BL/6J mice (2 month old) were obtained from Charles River Laboratories and maintained under standard conditions on a 12-hour day/night cycle, with water and food ad libitum. After washing out the blood, brains, pituitary gland, and spinal cord were quickly removed from the skull and spine and placed in ice-cold sterile PBS to make the tissue stiff and easier to dissect. The entire brain was carefully dissected into 17 sub-regions on an ice-cold surface. Retina samples were collected by separating the retina from the pigment layer in warm (37°C) PBS, pH 7.4. All dissected regions were placed in a 1.5 ml Eppendorf tube and snap-frozen in liquid nitrogen. Samples were stored at -80°C until further processing for the RNA extraction. Transcript expression of all brain regions, pituitary and retina were analysed. Tissue was homogenized mechanically using a TissueLyser LT (Qiagen) and total RNA was prepared using the RNeasy Mini isolation kit (Qiagen). This generated high-quality RNA, with 84% of the samples having RNA Integrity Number (RIN) values higher than 8.0 and only one sample removed due to a very low RIN value (less than 6.0). In total, 75 samples were subsequently used for library construction with Illumina TruSeq Stranded mRNA reagents. The Illumina HiSeq2500 platform was used for sequencing at approximately 20 million reads depth.
For a total number of 141 cell line samples, 375 normal tissue samples, and 109 blood samples, mRNA sequencing was performed on Illumina HiSeq2000 and 2500 machines (Illumina, San Diego, CA, USA) using the standard Illumina RNA-seq protocol with a read length of 2x100 bases. Blood cells mRNA sequencing was performed on an Illumina NovaSeq 6000 System in four S4 lanes with a read length of 2x150 bases. Transcript abundance estimation was performed using Kallisto v0.46.2. The 18 blood cell types are classified into six different lineages including B-cells, T-cells, NK-cells, monocytes, granulocytes and dendritic cells.
The HPA Human brain sample set contains 1324 samples of >200 regions of the human brain. The analysis is a collaboration with Human Brain Tissue Bank (HBTB; Semmelweis University, Budapest) in accordance with approval from the Committee of Science and Research Ethic of the Ministry of Health Hungary (ETT TUKEB: 189/KO/02.6008/2002/ETT) and the Semmelweis University Regional Committee of Science and Research Ethic (No. 32/1992/TUKEB) to remove human brain tissue samples, collect, store and use them for research. Samples were collected by Prof. Palkovits and RNA was extracted from frozen brain punches. The human prefrontal cortex dataset includes 165 samples from 3 male and 3 female donors providing a detailed overview of protein expression in 17 subregions of the prefrontal cortex and 3 reference cortical regions was analyzed using the Illumina sequencing platform, all other samples were analyzed using the MGI DNBSEQ-T7 platform.
The pig tissue samples were collected and analyzed in collaboration with BGI. Pig brain used for mRNA analysis were collected and handled in accordance with national guidance for large experimental animals and under permission of the local ethical committee (ethical permission numbers No.44410500000078 and BGI-IRB18135) as well as conducted in line with European directives and regulations. The experimental minipigs (Chinese Bama Minipig) were provided by the Peral Lab Animal Sci & Tech Co.,Ltd (Permit number SYXK2017-0123). Male (n = 2) and female (n = 2) Chinese Bama minipigs (1 year old), were housed in a specific pathogen-free stable facility under standard conditions. The brain was cut in coronal slabs at the level of 1) frontal lobe/olfactory tract, 2) optic chiasm and 3) between hypothalamus and cerebral peduncle. Slabs were divided in 2 hemispheres exposing all main brain structures. For mRNA analysis, pieces of cerebral cortex and cerebellum were collected, based on a sampling strategy collecting a representative sample that contained all cell layers. All other regions were dissected and collected completely. Two samples (somatosensory cortex and periaqueductal gray) are missing from female 1 due to the fact that these two regions could not be identified with 100% certainty, and thus were excluded. Duplicate samples were taken from olfactory bulb from female 2, resulting in totally 119 brain samples and additional 8 samples (retina and pituitary gland), all in all 127 samples. All samples were stored at -80° C until RNA was extracted within one month.
GTEx RNA-seq data
The Genotype-Tissue Expression (GTEx) project collects and analyzes multiple human post mortem tissues. RNA-seq data from 36 of their tissue types was mapped based on RSEMv1.3.0 (v8) and the resulting TPM values have been included in the Human Protein Atlas for all corresponding genes that could be mapped from Gencode v26 to Ensembl version 103.38. The GTEx retina data are based on EyeGEx data from Ratnapriya et al., Nature Genetics 2019 and transcript abundance estimation was performed using Kallisto v0.46.2 using Ensembl version 103.38 as reference genome.
FANTOM5 CAGE data
The Functional Annotation of Mammalian Genomes 5 (FANTOM5) project provides comprehensive expression profiles and functional annotation of mammalian cell-type specific transcriptomes using Cap Analysis of Gene Expression (CAGE) (Takahashi H et al. (2012)), which is based on a series of full-length cDNA technologies developed in RIKEN. CAGE data for 60 of their tissues was obtained from the FANTOM5 repository and mapped to Ensembl version 103.38.
Tissue Cell Type: Using GTEx bulk RNAseq data to profile gene cell type specificity
GTEx data was used in an integrative network analysis to determine the cell type specificity of all protein coding genes within a given tissue type. For more details on this analysis and the classifications, see the Tissue Cell Type section Methods Summary.
The single cell RNA sequencing dataset is based on meta-analysis of literature on single cell RNA sequencing and single cell databases that include healthy human tissue. To avoid technical bias and to ensure that the single cell dataset can best represent the corresponding tissue, the following data selection criteria were applied: (1) Single cell transcriptomic datasets were limited to those based on the Chromium single cell gene expression platform from 10X Genomics (version 2 or 3); (2) Single cell RNA sequencing was performed on single cell suspension from tissues without pre-enrichment of cell types; (3) Only studies with >4,000 cells and 20 million read counts were included, (4) Only dataset whose pseudo-bulk transcriptomic expression profile is highly correlated with the transcriptomic expression profile of the corresponding HPA tissue bulk sample were included. It should be noted that exceptions were made for lung (~7.3 million reads), pancreas (3,719 cells) and rectum (3,898 cells) to include various cell types in the analysis.
Single cell transcriptomics datasets
In total, single cell transcriptomics data for 25 tissues and peripheral blood mononuclear cells (PBMCs) were analyzed. These datasets were respectively retrieved from the Single Cell Expression Atlas, the Human Cell Atlas, the Gene Expression Omnibus, the Allen Brain Map, and the European Genome-phenome Archive. The complete list of references is shown in the table below.
Clustering of single cell transcriptomics data
For each of the single cell transcriptomics datasets, the quantified raw sequencing data were downloaded from the corresponding depository database based on the accession number provided by the corresponding study in the available format (total cells, read, and feature counts, or count tables). Unfiltered data were used as input for downstream analysis with an in-house pipeline using Scanpy (version 1.4.4.post1) in Python 3.7.3 for the 13 tissues and PBMC published in HPA v20 and Scanpy (version 1.7.2) in Python 3.8.5 for the 12 tissues published in HPA v21. In the pipeline, the data were filtered using two criteria: a cell is considered as valid if at least 200 genes are detected and a gene is considered as valid if it is expressed in at least 10% of the cells. Specially, in the HPA v21, tissues which contain more than 10,000 cells used 1000 cells as their cutoff. Subsequently, the cell counts were normalized to have a total count per cell of 10000. The valid cells were then clustered using Louvain clustering function within Single-Cell Analysis in Python (Scanpy). Default values of parameters were used in clustering. More in detail, the features of cells were projected into a PCA space with 50 components using UMAP, and a k-nearest neighbours (KNN) graph was generated. 15 neighbours were used in the network for Louvain, while the resolution of clustering was set as 1.0. The total read counts for all genes in each cluster was calculated by adding up the read counts of each gene in all cells belonging to the corresponding cluster. Finally, the read counts were normalized to transcripts per million protein coding genes (pTPM) for each of the single cell clusters. When calculating the expression profile for pseudo-bulk samples based on single cell transcriptomics, we added the read counts for all genes from all cells of the sample, and normalized it to pTPM in the same way as for the cluster ones.
Defining cell types
Each of the 444 different cell type clusters were manually annotated based on an extensive survey of >500 well-known tissue and cell type-specific markers, including both markers from the original publications, and additional markers used in pathology diagnostics. For each cluster, one main cell type was chosen by taking into consideration the expression of different markers. For a few clusters, no main cell type could be selected, and these clusters were not used for gene classification. The most relevant markers are presented in a heatmap on the Cell Type Atlas, in order to clarify cluster annotation to visitors.
Cell type dendrogram
The cell type dendrogram presented on the Single Cell Type section shows the relationship between the single cell types based on genome-wide expression. The dendrogram is based on agglomerative clustering of 1 - Spearman's rho between cell types using Ward's criterion. The dendrogram was then transformed into a hierarchical graph, and link distances were normalized to emphasize graph connections rather than link distances. Link width is proportional to the distance from the root, and links are colored according to cell type group if only one cell type group is present among connected leaves.
Normalization of transcriptomics data
For both the HPA and GTEx transcriptomics datasets, the average TPM value of all individual samples for each human tissue or human cell type was used to estimate the gene expression level. To be able to combine the datasets into consensus transcript expression levels, a pipeline was set up to normalize the data for all samples. In brief, all TPM values per sample were scaled to a sum of 1 million TPM (denoted pTPM) to compensate for the non-coding transcripts that had been previously removed. Next, all TPM values of all samples within each data source (HPA + GTEx human tissues, HPA immune cell types, HPA cell lines) were normalized separately using Trimmed mean of M values (TMM) to allow for between-sample comparisons. The resulting normalized transcript expression values, denoted nTPM, were calculated for each gene in every sample. nTPM values below 0.1 are not visualized on the Atlas sections.
For the brain dataset, an additional normalization was performed using linear regression to do the correction for inter-individual variation using the removeBatchEffect in the R package Limma with subject as a batch parameter. To reduce the technical variation between MGI and illumina platforms, 19 reference samples were included and run on both platforms. Intensity normalization based on reference samples was conducted to minimize technical variation between two platforms.
Consensus transcript expression levels for each gene were summarized in 54 human tissues based on transcriptomics data from the two sources HPA and GTEx. The consensus nTPM value for each gene and tissue type represents the maximum nTPM value based on HPA and GTEx. For tissues with multiple sub-tissues (brain regions, blood cells, lymphoid tissues and intestine) the maximum of all sub-tissues is used for the tissue type and the total number of tissue types in the human tissue consensus set is 36.
The FANTOM5 dataset was normalized separately on the sample level using TMM. The normalized Tags Per Million for each gene were calculated based on the average of all individual samples for each human tissue.
Mouse and pig transcriptomic data generated by the HPA in collaboration with BGI, were normalized separately, according to the same procedure used for human tissues and cell types, no Limma adjustment was performed on the mouse and pig data. Consensus transcript expression levels is summarized into 13 brain regions for mouse brain and 15 regions for pig brain, where sub-regional samples were combined and the maximum of sub-regions used for the brain region.
Single cell type clusters were normalized separately from other transcriptomics datasets using TMM. To generate expression values per cell type, clusters were aggregated per cell type by first calculating the mean nTPM in all cells with the same cluster annotation within a dataset. The values for the same cell types in different data sets were then mean averaged to a single aggregated value.
Classification of transcriptomics data
The consensus transcriptomics data was used to classify all genes according to their tissue-specific, single cell type-specific, brain region-specific, blood cell-specific or cell line-specific expression into two different schemas: specificity category and distribution category. These are defined based on the total set of all nTPM values in 256 tissues, 76 single cell types, 13 main regions of each mammalian brain,18 immune cell types or 69 cell lines and using a cutoff value of 1 nTPM as a limit for detection across all tissues or cell types.
Explanation of the specificity category
Explanation of the distribution category
External blood RNA-seq data
In addition to the immune cell type data from blood, generated within the Human Protein Atlas project, data from 15 immune cell types by Schmiedel et al. and 29 immune cell types as well as total PBMC by Monaco et al. have been incorporated into the Blood Atlas.
The Schmiedel dataset is available at the DICE (Database of Immune Cell Expression, Expression quantitative trait loci (eQTLs) and Epigenomics) database, which was established to address how genetic variants associated with risk for human diseases affect gene expression in various cell types. The TPM values per gene for 15 immune cell types were mapped to the corresponding genes in the Ensembl version used in the Human Protein Atlas.
The Monaco dataset contains data for 29 immune cell types within the peripheral blood mononuclear cell (PBMC) fraction of healthy donors using RNA-seq and flow cytometry. Raw data for 29 immune cells as well as total PBMC were analyzed using the same pipeline as for HPA-generated RNA-seq data and also normalized using TMM to allow for between-sample comparisons. Normalized gene expression values are reported as nTPM values.
Gene expression clustering of transcriptomics data
The RNA expression data has been used to classify protein-coding genes into expression clusters for tissues, single cell types, immune cells, and cell lines.
Pre-processing the data for clustering
For each dataset, genes detected at nTPM > 1 in at least one of the samples were selected, and the data was genewise scaled to z-scores to account for differences in dynamic ranges between genes across samples. After scaling, the expression data was projected into a lower dimensional space using Principal Component Analysis (PCA), where a number of components were selected to satisfy Kaiser’s rule and at least 80% of variance explained.
Gene to gene distances were calculated as the Spearman correlation of gene expression across samples, and transformed to Spearman distance (1 - Spearman correlation). The distances were transformed into a shared nearest neighbor graph and used for Louvain clustering to find clusters of genes with similar expression profiles within the graph. To account for stochasticity in the clustering process each clustering was run 100 times, and consequently collapsed into a single consensus clustering. Confidence of the gene-to-cluster assignment was calculated as the fraction of times that the gene was assigned to the cluster.
The clustering generated for each of the datasets is manually annotated to assign a specificity and function to each cluster. The annotation is based on overrepresentation analysis towards biological databases, including Gene Ontology, Reactome, PanglaoDB, TRRUST, and KEGG, as well as HPA classifications including subcellular location, protein class, secretion location and classification, and specificity toward tissues, single cell types, immune cells, brain regions, and cell lines. A reliability score is manually set for each cluster indicating the confidence of specificity and function assignment.
The clustering results are visualized in a UMAP. Colored polygons were generated to represent the main contiguous masses of genes corresponding to the same cluster. First, for each cluster, the two-dimensional density was estimated in the UMAP, and an area enveloping 95% of the total density was determined. The areas were moderated to include contiguous areas corresponding to at least 5% of the total area in the UMAP space. Finally, contiguous areas were converted to two-dimensional polygons per each cluster.
TCGA RNA-seq data
The Cancer Genome Atlas (TCGA) project of Genomic Data Commons (GDC) collects and analyzes multiple human cancer samples. RNA-seq data from 17 cancer types representing 21 cancer subtypes with a corresponding major cancer type in the Human Pathology Atlas were included to allow for comparisons between the protein staining data from the Human Protein Atlas and RNA-seq from TCGA data.
The TCGA RNA-seq data was mapped using the Ensembl gene id available from TCGA, and the FPKMs (number Fragments Per Kilobase of exon per Million reads) for each gene were subsequently used for quantification of expression with a detection threshold of 1 FPKM. Genes were categorized using the same classification as described above.
Based on the FPKM value of each gene, patients were classified into two expression groups and the correlation between expression level and patient survival was examined. The prognosis of each group of patients was examined by Kaplan-Meier survival estimators, and the survival outcomes of the two groups were compared by log-rank tests. Both median and maximally separated Kaplan-Meier plots are presented in the Human Protein Atlas, and genes with log rank P values less than 0.001 in maximally separated Kaplan-Meier analysis were defined as prognostic genes. If the group of patients with high expression of a selected prognostic gene has a higher observed event than expected event, it is an unfavorable prognostic gene; otherwise, it is a favorable prognostic gene. Genes with a median expression less than FPKM 1 were lowly expressed, and classified as unprognostic in the database even if they exhibited significant prognostic effect in survival analysis
Allen Mouse brain ISH dataset
The Allen Brain Atlas (ABA) is an open access database focusing on the brain, and includes both human and mouse expression data. The ABA is a part of the Allen Institute for Brain Science, which is one of the three branches of the Allen Institute. The Mouse brain In situ hybridization (ISH) data provides information on where in the adult mouse brain each gene is expressed (Lein ES et al. (2007)). We have imported the expression values available through the ABA API (© 2004 Allen Institute for Brain Science, Allen Mouse Brain Atlas) and show the regional expression grouped in the same manner as the other datasets visualized on the HPA Brain Atlas.
The Allen mouse brain ISH data was mapped to the mouse gene annotation of Ensembl version 103.38 using the probe nucleotide sequences provided through the Allen mouse brain API together with the blast program package. The mouse genes where then mapped to human genes using Ensembl orthologue data with a one-to-one restriction.
Protein evidence is calculated for each gene based on three different sources: UniProt protein existence (UniProt evidence); a Human Protein Atlas antibody- or RNA based score (HPA evidence); and evidence based on PeptideAtlas (MS evidence). In addition, for each gene, a protein evidence summary score is based on the maximum level of evidence in all three independent evidence scores (Evidence summary).
All scores are classified into the following categories:
UniProt evidence is based on UniProt protein existence data, which uses five types of evidence for the existence of a protein. All genes in the classes "Experimental evidence at protein level" or "Experimental evidence at transcript level" are classified into the first two evidence categories, whereas genes from the "Inferred from homology", "Predicted", or "Uncertain" classes are classified as "No evidence". Genes where the gene identifier could not be mapped to UniProt from Ensembl version 103.38 are classified as "Not available".
The HPA evidence is calculated based on the manual curation of Western blot, tissue profiling and subcellular location as well as transcript profiling. All genes with Data reliability "Supported" in one or both of the two methods immunohistochemistry and immunofluorescence, or standard validation "Supported" for the Western blot application (assays using over-expression lysates not included) are classified as "Evidence at protein level". For the remaining genes, all genes detected at NX > 1 in at least one of the tissue types, cell type or cell lines used in the RNA-seq analysis based on HPA, GTEx or FANTOM5 are classified as "Evidence at transcript level". The remaining genes are classified as "No evidence".
The MS evidence displays protein evidence based on PeptideAtlas for all proteins with protein evidence in NextProt and with a minimum of two distinct peptides. Each gene detected on the protein level according to PeptideAtlas is classified as "Evidence at protein level" and all remaining genes as "Not available".