Estimating the population abundance of tissue-infiltrating immune and stromal cell populations using gene expression.
Becht E, Giraldo NA, Lacroix L, Buttard B, Elarouci N, Petitprez F, Selves J, Laurent-Puig P, Sautès-Fridman C, Fridman WH, de Reyniès A. · 2016
Read Estimating the population abundance of tissue-infiltrating immune and stromal cell populations using gene expression. by Becht E, Giraldo NA, Lacroix L, Buttard B, Elarouci N, Petitprez F, Selves J, Laurent-Puig P, Sautès-Fridman C, Fridman WH, de Reyniès A. (2016) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/europmc-pmc5073889.txt.
Introduction
Etienne Becht and ten co-authors introduce MCP-counter in Genome Biology in 2016, a method that counts cell populations inside a tissue from its gene activity alone. Knowing how many immune and stromal cells infiltrate a tumor or organ guides prognosis and therapy, but direct measurement needs fresh samples and lab assays. MCP-counter instead reads marker genes in a bulk transcriptome and returns abundance scores for eight immune and two stromal populations.
The team validated the scores against lab-made RNA mixtures of known makeup and against stained tissue sections, and showed the method avoids weaknesses of earlier computational approaches. Applied broadly, it sketched immune landscapes across healthy tissues and non-blood cancers, recovering patient groupings tied to survival in lung, colorectal, and breast cancers. Running on existing data, it lets researchers mine old datasets for such clues and may help select patients for immunotherapy.
Catalog record
| CREATOR | Becht E, Giraldo NA, Lacroix L, Buttard B, Elarouci N, Petitprez F, Selves J, Laurent-Puig P, Sautès-Fridman C, Fridman WH, de Reyniès A. |
|---|---|
| YEAR | 2016 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 81 KB |
| LICENSE | CC BY 4.0 |
| RIGHTS | CC BY |
| SOURCE | Europe PMC |
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Estimating the population abundance of tissue-infiltrating immune and stromal cell populations using gene expression By Etienne Becht, Nicolas A Giraldo, Laetitia Lacroix, Bénédicte Buttard, Nabila Elarouci, Florent Petitprez, Janick Selves, Pierre Laurent-Puig, Catherine Sautès-Fridman, Wolf H Fridman, Aurélien de Reyniès Published in: Genome Biology DOI: 10.1186/s13059-016-1070-5 ABSTRACT Abstract We introduce the Microenvironment Cell Populations-counter (MCP-counter) method, which allows the robust quantification of the absolute abundance of eight immune and two stromal cell populations in heterogeneous tissues from transcriptomic data. We present in vitro mRNA mixture and ex vivo immunohistochemical data that quantitatively support the validity of our method’s estimates. Additionally, we demonstrate that MCP-counter overcomes several limitations or weaknesses of previously proposed computational approaches. MCP-counter is applied to draw a global picture of immune infiltrates across human healthy tissues and non-hematopoietic human tumors and recapitulates microenvironment-based patient stratifications associated with overall survival in lung adenocarcinoma and colorectal and breast cancer. Electronic supplementary material The online version of this article (doi:10.1186/s13059-016-1070-5) contains supplementary material, which is available to authorized users. Keywords: Transcriptomic markers, Gene signatures, Deconvolution, Tumor microenvironment BACKGROUND The abundance of tissue-infiltrating immune and non-immune stromal cell populations is highly informative of the types of inflammatory, angiogenic, and desmoplastic reactions occurring in a diseased tissue. In cancer, multiple studies focusing on a limited number of immune populations have reported an association between the extent of infiltrating immune cells and prognosis (reviewed in [1]). For instance, it has been shown that T-cell infiltration correlates with favorable outcome in colorectal [2, 3] and many other cancers [4] and with poor outcome in clear-cell renal cell carcinoma [5]. Other immune cell types, such as macrophages, B cells and natural killer (NK) cells [1], as well as endothelial cells and fibroblasts [6, 7], have been shown to positively or negatively influence the prognosis of cancer patients. Yet, as quantifying just a single cell population is demanding, little is known about how all these cell populations collectively predict the prognosis of cancer patients. The ability to simultaneously quantify multiple cell populations within a tissue sample thus appears critical to identify clinically relevant classes of diseased tissues based on their inflammatory and stromal profiles. In cellularly heterogeneous tissue samples, transcriptomic measurements average signals originating from the distinct underlying cell populations. The deconvolution of these signals can yield estimates of cell population proportions in a sample [8]. Tens of thousands of transcriptomic profiles are readily available for retrospective analyses from public repositories such as Gene Expression Omnibus (GEO), ArrayExpress, or The Cancer Genome Atlas (TCGA). Different transcriptome-based computational methods were recently proposed to characterize the proportions of immune populations within leukocytes but omit the critical parameter of the overall abundance of leukocytes in the sample [9, 10]. Other published methods lack quantitative validation [11] or are limited to only two cell populations [12]. To overcome these limitations, we introduce Microenvironment Cell Populations (MCP)-counter, a transcriptome-based computational method that robustly quantifies the abundance of immune and non-immune stromal cell populations in a heterogeneous tissue sample. MCP-counter is available as an R package. From a gene expression matrix, it produces for each sample an abundance score for CD3+ T cells, CD8+ T cells, cytotoxic lymphocytes, NK cells, B lymphocytes, cells originating from monocytes (monocytic lineage), myeloid dendritic cells, neutrophils, as well as endothelial cells and fibroblasts. MCP-counter estimates are “single sample” scores, in the sense that they are computed on each sample independently. These scores can then be used for direct comparisons of the abundance of the corresponding cell type across samples within a cohort. We show that our approach either complements or outperforms previously published methods. We quantitatively validated MCP-counter both in vitro, by using mRNA mixtures, and ex vivo, by using immunohistochemical cell quantifications on paraffin-embedded tissue sections. We illustrate its application to assess tissue-infiltration in 47 healthy tissue types and in 32 non-hematological malignancies. We show that our method is able to reproduce immunological and stromal prognostic classifications in lung adenocarcinoma and colorectal and breast cancers. RESULTS Development and validation process of the MCP-counter method We designed a method, called MCP-counter, with the objective of measuring the inter-sample relative abundance of different cell populations in a microenvironment. To develop and validate our method, we designed a seven step strategy (Fig. 1). MCP-counter is based on the methodological framework [8] of transcriptomic markers (TM), defined as gene expression features expressed in one and only one cell population and whose expression shows little variation within the population of interest (step 1). It directly follows that, within a sample composed of many different cell populations, the abundance of the population of interest is proportional to the sample’s expression of its related TM. Given their restrictive definition, TM are not guaranteed to exist for all populations. We set up a discovery series by curating Microenvironment Cell Population (MCP) transcriptomes from 81 public datasets derived using Affymetrix Human Genome U133 Plus 2.0 microarrays (step 2). These transcriptomes were normalized [13] to allow their integration into a large meta-dataset which includes 1194 immune or non-immune microenvironment cell population samples; 742 tumor cell line samples were included as negative controls (Additional file 1: Table S1). No hematopoietic tumor cell line was included as a negative control as the transcriptome of these cells is closely related to that of non-malignant hematopoietic cells [14]. After manual curation, all the samples were annotated using 63 labels, of which 42 correspond to microenvironment cell populations and 21 represent cell lines from 21 non-hematopoietic cancer types (Fig. 2a). We organized the 63 cell populations into a pyramidal graph according to rules of inclusion (Additional file 2: Figure S1). We added hematopoiesis-inspired (e.g., lymphoid and myeloid lineages) or functional (e.g., cytotoxic lymphocytes) categories to this pyramid, resulting in a total of 67 nodes corresponding to potential cellular populations in the tissue microenvironment (step 3) (Additional file 2: Figure S1). This organization was validated by its consistency with clusters obtained with transcriptome-based principal component analysis (step 4; Additional file 2: Figure S2). [Figure: Purpose and development of the MCP-counter method. a Comparison of MCP-counter estimates and CIBERSORT-estimated leukocytic fractions. b Framework of MCP-counter’s development and validation] [Figure: Identification and qualitative validation of transcriptomic markers. a The MCP discovery series. pDC plasmacytoid dendritic cell, PBMC peripheral blood mononuclear cell. b Quartiles of MCP-counter scores on positive and control samples in the discovery and validation microenvironment series. Gray indicates missing values. c Representative transcriptomic markers and their corresponding expression patterns in the MCP discovery series] To identify TM of a given cell population (a node in our cell population pyramid; step 5), we defined as “positive” the samples included in this population and we defined as “negative” the samples that do not contain this population. Samples containing both positive and negative cells are omitted from the analysis for this node. Three criteria were then calculated for each feature (probe set) within the discovery set: a) the mean log2-expression difference between positive and negative samples (a threshold of 2 was applied); b) the area under the ROC curve (AUC) of the feature for the identification of the positive samples (threshold of 0.97); and c) a measure of the signal to noise ratio between positive and negative samples (threshold of 1.5) (“Methods”; Additional file