Spatial distribution of the proteome in the human body and in cancers

Main

Grounded in a genetic blueprint, diverse human tissues and organs with distinct functions arise during development. These functions can become dysregulated in pathological conditions, such as tumours. Characterizing proteomic variation across tissue types in both developmental and pathological contexts is essential for enhancing our understanding of human biology and advancing therapeutic development. Although transcriptomic repositories, such as ArrayExpress 5 , RNA-Seq Atlas 6 and the BioGPS portal 7 , have provided initial annotations for tissue expression, and the Adult GTEx project has further expanded this using genomic and transcriptomic data from tissue sites that are not affected by disease 8 , 9 , 10 , mRNA abundance correlates only moderately with the expression of proteins, which are the main functional and druggable molecules.

The Human Protein Atlas (HPA), which was launched in 2005, began by incorporating immunohistochemistry-based proteomic data from healthy and cancerous tissues, and has since expanded continuously 11 . By 2015, the HPA had integrated transcriptomic data from 32 healthy tissue types and proteomic data, based on 20,456 antibodies, from 44 healthy tissue types 3 . Although antibody-based protein measurement has an advantage in that it provides localized protein information, its semi-quantitative nature limits the reliable quantification of thousands of proteins, particularly those for which effective antibodies are lacking. By contrast, mass spectrometry (MS) offers a comprehensive, multiplexed and unbiased alternative for quantitative proteomic measurement 12 , 13 .

In 2014, two MS-based human proteome drafts reported the identification of approximately 85% of proteins encoded by human genes across around 30 tissue types and cell lines 1 , 2 . A few years later, another study 14 characterized 15,210 protein groups across 29 tissues using label-free data-dependent acquisition MS, and, more recently 4 , researchers quantified 12,027 proteins across 32 tissue types using tandem mass tag-based MS. Although these studies advanced human tissue proteomics, they focused on a limited set of approximately 30 major tissues, leaving many uncharted, and lacked comprehensive comparisons between healthy and cancerous tissue.

Consortia such as The Cancer Genome Atlas (TCGA), the International Cancer Genome Consortium and the Clinical Proteomic Tumor Analysis Consortium (CPTAC) have generated extensive multi-omics datasets for specific tumours 15 , 16 ; however, challenges in cross-tumour type comparisons limit the insights into differences between cancers that can be obtained from these resources 17 . Comprehensive proteome profiling across diverse human tissues and states requires broad tissue coverage and in-depth, high-throughput proteomics, and this need is addressed effectively by data-independent acquisition mass spectrometry (DIA-MS) 18 , 19 , 20 . Here, using DIA-MS, we present a rich data resource detailing the spatial distribution of 13,609 proteins. This coverage includes 58 healthy adult tissues, paired tumour and non-tumour samples from 25 cancer types, and 22 fetal tissue types, encompassing nearly all solid human tissues, body fluids and major cancer types (Fig. 1 and Supplementary Information ). This resource provides a foundation for navigating the human body proteome 21 , and will facilitate cancer drug discovery.

Fig. 1: Overview of the draft human proteome. The alternative text for this image may have been generated using AI.

Full size image

Summary of the sample types in our data. Each sample type is marked with a corresponding number. The specific sampling positions of the eye, blood, human embryo, ear, nose and brain are listed on the left. Twenty-five types of cancer are shown at the bottom. All samples are labelled with the full and abbreviated names of the organ, tissue or cancer, and the numbers in parentheses indicate the number of proteins retained in each submatrix that were quantified in more than 50% of the samples in the specific tissue types. In this figure, ‘female genitalia’ is used as a category label for the fallopian tubes, uterus and vagina, and ‘male genitalia’ is used as a category label for the epididymis, seminal vesicle, seminiferous duct and Cowper’s gland. Components of this figure were created in BioRender; Guomics Lab https://BioRender.com/hwyxuac and https://BioRender.com/96s0zl0 (2026); other components were created using Magnific.

Source data

Samples and proteomic profiling

We collected 2,856 samples from 9 post-mortem adult donors, 8 healthy participants, 9 post-mortem fetal donors and 1,015 patients with cancer (Supplementary Table 1 ). We first constructed a comprehensive human spectral library from these samples, containing 15,332 protein groups (Extended Data Fig. 1a and Supplementary Information ). To ensure the quality of large-scale proteomic data, we searched the DIA-MS raw data against a combined spectral library that inte…