CODEX multiplexed imaging of immunotherapy in human and mouse melanomas
Data files
Dec 21, 2023 version files 6.27 GB
-
23_07_04_day013512_neighborhoods.csv
908.81 MB
-
23_10_11_Melanoma_Marker_Cell_Neighborhood.csv
5.01 GB
-
Day3_Markers_Dryad.csv
350.98 MB
-
metadata.csv
1.51 KB
-
README.md
13.87 KB
Abstract
Our research used CODEX (Co-Detection by Indexing) multiplexed imaging to gain insights into melanoma tumors in both murine models and human samples. CODEX imaging involves an iterative process of annealing and stripping fluorophore-labeled oligonucleotide barcodes, complementing the barcodes attached to over 40 antibodies used for tissue staining. Subsequently, images underwent standard CODEX image processing (tile stitching, drift compensation, cycle concatenation, background subtraction, deconvolution, and determination of best focal plane), single cell segmentation, and column marker z-normalization by tissue.
Our datasets comprise individual cells as rows, each characterized by 40+ antibody fluorescence values quantified from various markers evaluated for each study. These markers correspond to the antibodies targeting specific proteins within the tissue, quantified at the single-cell level. The values represent per-cell/area-averaged fluorescent intensities, z-normalized along each column. Each cell is mapped with its cell type and cellular neighborhood, defined by x and y coordinates representing pixel locations in the original image. Refer to the table in the "Usage Notes" section below for further details. The CODEX multiplexed imaging data is organized into three distinct files, each representing key aspects of our research and the studies detailed in our manuscript.
We then used this data investigate the major cellular organization of the tumor sections we imaged, with downstream spatial statistics and analyses like cellular neighborhood analysis and cell-cell interaction analysis. These data could be used to understand the cellular interactions, composition, and structure of anti-tumor melanoma responses induced by antigen-specific immunotherapy either with adoptive T cell transfer for checkpoint blockade immunotherapy. These datasets offer valuable insights for researchers interested in anti-tumor microenvironments, immune responses, and therapeutic interventions such as T cell therapies.
1. Time-course of tumor microenvironment following antigen-specific T cell therapy in mice
We investigate the dynamic interplay between immune responses, antigen-specific T cell interactions, and tumor progression in a murine melanoma model. We activated PMEL CD8+ T cells with cognate antigen gp100 and IL-2 for 10 days ex vivo and transferred into mice with established B16-F10 tumors. Tumors were harvested and imaged with CODEX imaging at 0-, 1-, 3-, 5-, and 12-days post-treatment (n=3-7 per time point). Our 42-plex CODEX antibody panel characterizes immune cell types, T cell phenotypes, stromal cell types, and tumor cell phenotypes, resulting in a rich dataset of 1,052,125 cells across 42 marker channels.
2. Tumor microenvironment following antigen-specific T cell therapies with different phenotypes in mice
We delve deeper into the modulation of the tumor microenvironment by manipulating T cell phenotypes. By comparing activated T cells stimulated with and without 2-hydroxycitrate (2HC), a metabolic inhibitor of acetyl CoA production, we explore the impact of phenotype on tumor progression. Our datasets from mice treated with 2HC T cells or T cells provide insights into the role of T cell phenotype manipulation in the tumor microenvironment (n=4-7 per group).
3. Tumor microenvironment before and after checkpoint blockade in human melanoma patients of both responders and non-responders
Our research extends to human melanoma patients with advanced, metastatic, stage IV tumors. We examine 12 FFPE tumor samples from six patients, each with samples taken before and after checkpoint inhibitor therapy. Our CODEX multiplexed imaging, using a panel of 58 antibodies, reveals changes in immune, stromal, and tumor compartments. We segmented 5,019,159 individual cells from the 12 CODEX images, facilitating unsupervised clustering to identify 39 major cell types based on their expression profiles. Our accompanying donor metadata table links donor IDs to essential clinical information, including treatment response, demographics, and sample details.
README: CODEX multiplexed imaging of immunotherapy in human and mouse melanomas
https://doi.org/10.5061/dryad.k0p2ngfcc
Our research used CODEX (Co-Detection by Indexing) multiplexed imaging to gain insights into melanoma tumors in both murine models and human samples. CODEX imaging involves an iterative process of annealing and stripping fluorophore-labeled oligonucleotide barcodes, complementing the barcodes attached to over 40 antibodies used for tissue staining. Subsequently, images underwent standard CODEX image processing (tile stitching, drift compensation, cycle concatenation, background subtraction, deconvolution, and determination of best focal plane), single cell segmentation, and column marker z-normalization by tissue.
Description of the data and file structure
Our datasets comprise individual cells as rows, each characterized by 40+ antibody fluorescence values quantified from various markers evaluated for each study. These markers correspond to the antibodies targeting specific proteins within the tissue, quantified at the single-cell level. The values represent per-cell/area-averaged fluorescent intensities, z-normalized along each column. Each of these antibodies are described in detail in the Supplementary Table 2 of the published manuscript. Each cell is mapped with its cell type and cellular neighborhood, defined by x and y coordinates representing pixel locations in the original image.
We then used this data investigate the major cellular organization of the tumor sections we imaged, with downstream spatial statistics and analyses like cellular neighborhood analysis and cell-cell interaction analysis. These data could be used to understand the cellular interactions, composition, and structure of anti-tumor melanoma responses induced by antigen-specific immunotherapy either with adoptive T cell transfer for checkpoint blockade immunotherapy. These datasets offer valuable insights for researchers interested in anti-tumor microenvironments, immune responses, and therapeutic interventions such as T cell therapies.
Specifics about each dataset and specific columns is found below:
1. Time-course of tumor microenvironment following antigen-specific T cell therapy in mice
Filename: 23_07_04_day013512_neighborhoods.csv
Tumors were harvested and imaged with CODEX imaging at 0-, 1-, 3-, 5-, and 12-days post-T cell therapy (n=3-7 per time point). Our CODEX antibody panel characterizes immune cell types, T cell phenotypes, stromal cell types, and tumor cell phenotypes, resulting in a rich dataset of 1,052,125 cells across marker channels. Antibody panels were not consistent across all experiments, hence why some markers have Nan values within the columns after concatenation of the data. Which antibodies were used for which experiment is also described in detail in the Supplementary Table 2 of the manuscript. We opted to include these extra markers (even though there is missing data and these were not used in clustering, for potential use in the future). All other descriptive columns are described below:
x | Tissue x position in each region imaged |
---|---|
y | Tissue y position in each region imaged |
array | Experiment group for each set of samples |
date_array | Tissue array from which each region was imaged |
Xcorr | Corrected x position in each array imaged |
Ycorr | Corrected y position in each array imaged |
replicate | Which replicate of the treated mice the cells came from |
unique_region | Label for unique region from both day_replicate and region (can be used for x, y coordinates) |
region_num | Number region from initial imaging data |
day harvested | Day that the tumors were taken from mice |
Cell common | Cell type labels used for the paper analysis |
neigh_name_all | Neighborhood labels used for the paper analysis (done on unified dataset) |
prior_index | Original cell index from previous experiments |
original cell type | Cell type labels that were created on a per experiment basis that were later merged into 'Cell common' for comparison between experiments |
original neighborhood | Neighborhood labels that were created on a per experiment basis that were later merged into 'Neighborhood common' for comparison between a subset of experiments |
Neighborhood common | Neighborhood labels from subset of the data taken together |
2. Tumor microenvironment following antigen-specific T cell therapies with different phenotypes in mice
Filename: Day3_Markers_Dryad.csv
We compared activated T cells stimulated with and without 2-hydroxycitrate (2HC), a metabolic inhibitor of acetyl CoA production as a therapy for established melanomas in mice. Our datasets from mice treated with 2HC T cells or T cells provide insights into the role of T cell phenotype manipulation in the tumor microenvironment (n=4-7 per group). Our CODEX antibody panel characterizes immune cell types, T cell phenotypes, stromal cell types, and tumor cell phenotypes, resulting in a rich dataset of 520,385 cells across marker channels. Antibody panels were not consistent across all experiments, hence why some markers have Nan values within the columns after concatenation of the data. Which antibodies were used for which experiment is also described in detail in the Supplementary Table 2 of the manuscript. We opted to include these extra markers (even though there is missing data and these were not used in clustering, for potential use in the future). All other descriptive columns are described below:
x | Tissue x position in each region imaged |
---|---|
y | Tissue y position in each region imaged |
date_array | Tissue array from which each region was imaged |
x_array | Corrected x position in each array imaged |
y_array | Corrected y position in each array imaged |
treatment | What treatment condition the mice received |
replicate | Which replicate of the treated mice the cells came from |
unique_region | Label for unique region from both day_replicate and region (can be used for x, y coordinates) |
region_num | Number region from initial imaging data |
day harvested | Day that the tumors were taken from mice |
experiment | Which experiment the samples were from |
tissue | What type of tissue was imaged |
Cell common | Cell type labels used for the paper analysis |
Cell cat | Major cell categories |
Major cat | Immune vs. non-immune cell category label for cells |
prior_index | Original cell index from previous experiments |
original cell type | Cell type labels that were created on a per experiment basis that were later merged into 'Cell common' for comparison between experiments |
original neighborhood | Neighborhood labels that were created on a per experiment basis that were later merged into 'Neighborhood common' for comparison between a subset of experiments |
Neighborhood common | Neighborhood labels used for the paper analysis |
3. Tumor microenvironment before and after checkpoint blockade in human melanoma patients of both responders and non-responders
Filename: 23_09_11_Melanoma_Marker_Cell_Neighborhoods.csv
Metadata Filename: metadata.csv
We examined 12 FFPE tumor samples from six patients with advanced, metastatic, stage IV tumors, each with samples taken before and after checkpoint inhibitor therapy. Our CODEX multiplexed imaging, using a panel of 58 antibodies, reveals changes in immune, stromal, and tumor compartments. We segmented 5,019,159 individual cells from the 12 CODEX images, facilitating unsupervised clustering to identify 39 major cell types based on their expression profiles. Our accompanying donor metadata table links donor IDs to essential clinical information, including treatment response, demographics, and sample details. Antibody panels were not consistent across all experiments, hence why some markers have Nan values within the columns after concatenation of the data. We opted to include these extra markers (even though there is missing data and these were not used in clustering, for potential use in the future). All other descriptive columns are described below:
x | Tissue x position in each region imaged |
---|---|
y | Tissue y position in each region imaged |
cellid | Original segmentation label for the cell |
donor | Unique donor label that connects with the metadata csv also attached |
Overall_Cell_Type | Lowest resolution labels of cell types |
Cell_Type_Common | 2nd lowest resolution labels of cell types |
Cell_Type_Sub | 3rd lowest resolution labels of cell types |
Cell_Type | 4th lowest resolution labels of cell types and original cell type labels from clustering |
filename | Unique label for individual CODEX Phenocylcer runs |
region | Number region from initial imaging data |
Neighborhood | Neighborhood labels used for the paper analysis |
Sharing/Access information
N/a
Code/Software
N/a
Methods
For a detailed description of each of the steps of protocols and processes to obtain this data see the detailed materials and methods in the associated manuscript. Briefly, murine melanomas were extracted from sacrificed mice at indicated timepoint post therapy and frozen in OCT. These were assembled into an array of 4-10 tissues, cut into 7 um slices, and stained with a panel of 40+ CODEX DNA-oligonucleotide barcoded antibodies. Tissues were imaged with a Keyence microscope at 20x objective and then processed using image stitching, drift compensation, deconvolution, and cycle concatenation. For human melanoma samples, we took available tumor tissue and cut with a 3µm section thickness and mounted onto Superfrost PLUS slides for further staining with a panel of 58 CODEX DNA-oligonucleotide barcoded antibodies imaged with the Phenocycler Fusion machine and processed using manufacturer’s v1.6.0 (e.g., background-subtracted, stitched, and shading correction).
Processed data were then segmented using CellVisionSegmenter, a neural network R-CNN-based single-cell segmentation algorithm. Cell type analysis was completed on B004, 5, and 6 by z normalization of protein markers used for clustering and then overclustered using leiden-based clustering. The cell type labels were verified by looking back at the original image. With set cell type labels we performed neighborhood analysis by clustering windows of the 10 nearest neighbors around a given cell and were named based on cell type enrichment and location in the tissue. Broad categories for cell types, neighborhoods, and communities were expert annotated based on epithelial, immune, or other stromal compartments.