Molecular and pharmacological modulators of the tumor immune contexture revealed by deconvolution of RNA-seq data
Francesca FinotelloiD, Clemens Mayer, Christina PlattneriD, Gerhard Thomas Laschober, Dietmar RiederiD, Hubert HackliD, Anne KrogsdamiD, Zuzana LoncováiD, Wilfried PoschiD, Doris WilflingsederiD, Sieghart SopperiD, Marieke E. IjsselsteijniD, Thomas P. BrouweriD, Douglas Buckner JohnsoniD, Yaomin XuiD, Yu WangiD, Melinda E. SandersiD, Monica Valeria EstradaiD, Paula I. González EricssoniD, Pornpimol CharoentongiD, Justin Micah BalkoiD, Noel F.C.C. de MirandaiD, Zlatko TrajanoskiiD
Abstract
The immune contexture has a prognostic value in several cancers and the study of its pharmacological modulation could identify drugs acting synergistically with immune checkpoint blockers. However, the quantification of the immune contexture is hampered by the lack of simple and efficient methods. We developed quanTIseq, a deconvolution method that quantifies the densities of ten immune cell types from bulk RNA sequencing data and tissue imaging data. We performed extensive validation using simulated data, flow cytometry data, and immunohistochemistry data from three cancer cohorts.\n\nAnalysis of 8,000 samples showed that the activation of the CXCR3/CXCL9 axis, rather than the mutational load is associated with cytotoxic T cell infiltration. We also show the prognostic value of deconvolution-based immunoscore and T cell/B cell score in several solid cancers. Finally, we used quanTIseq to show how kinase inhibitors modulate the immune contexture, and we suggest that it might have predictive value for immunotherapy.
02 /Specification & Classification
Click any tag to browse matching papers- Citations
- 1,546
- FWCI
- 44.60
- Field percentile
- top 1%
- Organism
- Assay
- Research Focus
- Approach
In-depth summary
Claude Opus 5.5 · xhigh effort · 8 Nov 2026 · from the full text of version 2 (17 Aug 2018) · CC BY 4.0
quanTIseq estimates absolute fractions of ten immune cell types from bulk RNA-seq; across TCGA, CXCL9/CXCR3 expression, not mutational load, tracks CD8+ T cell infiltration.
Signature-based tools such as MCP-counter and xCell give only enrichment scores, CIBERSORT reports fractions relative to immune cells and was built on microarrays, TIMER outputs are not comparable across cell types, and EPIC omits Treg, dendritic cells and M1/M2 macrophages. quanTIseq is designed for RNA-seq, returns absolute fractions for inter- and intra-sample comparison, and is validated against matched IHC/IF from solid tumors.
Key results
ShownOn new blood-cell mixtures from healthy donors, quanTIseq estimates closely matched flow cytometry for every measured population (macrophages excluded, as they are absent from blood), including closely related Treg and non-regulatory CD4+ T cells and scarce dendritic cells.
PBMC with 3-6% PMN spike-ins from nine healthy donors: overall correlation 0.87 vs flow cytometry. High concordance also on a published PBMC RNA-seq/flow cytometry set, two published microarray sets, and 1,700 simulated breast-tumor RNA-seq sets built with leave-K-out signatures. · Validation of quanTIseq with flow cytometry…; Validation of quanTIseq using simulated RNA-seq…; Generation of the simulated data sets; Generation of flow cytometry and RNA-seq data…
ShownIn three tumor cohorts with matched RNA-seq and stained slides, deconvolved fractions tracked image-based immune cell densities, and the uncharacterized 'other' fraction tracked independent TCGA tumor purity estimates.
Pearson r ranged 0.32-0.75 in melanoma (IHC, 32 patients imaged), 0.55-0.91 in lung cancer (IHC, 8 patients) and 0.58-0.90 in CRC (IF/IHC, ten patients); agreement was lower for Treg in melanoma. 'Other' fraction vs consensus purity: correlation from 0.29 (GBM) to 0.72 (SKCM), positive in every cancer type with at least 100 patients having both estimates. · Validation of quanTIseq with flow cytometry…; Analysis of Vanderbilt images with IHCount; Validation of quanTIseq using simulated RNA-seq…
ShownAcross TCGA solid tumors, immune cell fractions were largely unrelated to tumor genomic features, while CD8+ T cell fractions correlated strongly with expression of the chemokines CXCL9 and CXCL10 and their receptor CXCR3.
More than 8,000 TCGA samples across 19 solid cancers: low or no correlation of immune fractions with mutational load, neoantigen load, tumor heterogeneity or clonal/subclonal mutation fractions; high Pearson correlation of CD8+ T cell fraction with CXCL9 TPM, also with CXCL10 and CXCR3 (coefficients shown only in figures). · Activation of the CXCL9/CXCR3 axis is associated…; DISCUSSION
ShownMedian-split versions of the immunoscore (CD3+ and CD8+ T cells) and a T cell/B cell score, computed from deconvolved fractions, separated overall survival in several TCGA solid cancers but not in colorectal cancer.
Hi-Hi vs Lo-Lo groups, Kaplan-Meier and log-rank tests in 19 TCGA solid cancers: immunoscore significant in BRCA, CESC, HNSC, SKCM, UCEC; TB score in BRCA, CESC, HNSC, LUAD, PRAD (text says six cancers but names five); no significant association in CRC. · Deconvolution-based immune scores are associated…; Computation of the deconvolution-based…; Statistical analysis
SuggestedBRAF and/or MEK inhibition reshaped melanoma immune contextures, most clearly by raising dendritic cell fractions. In anti-PD1 responders, image-scaled densities pointed to more T cells and M1 macrophages, but the differences were not statistically significant.
Pre- vs on-treatment biopsies from seven inhibitor-treated patients: dendritic cells rose (p=0.043); CD8+ T cells (p=0.19) and M2 macrophages (p=0.07) did not reach significance. In 21 pre-anti-PD1 samples with H&E-scaled densities, M1, CD4+ and CD8+ T cell densities were higher in responders (p>0.09). · Pharmacological modulation of the tumor immune…
Methods, data and reproducibility
Raw reads are trimmed (Trimmomatic) and quantified (Kallisto TPM), then deconvolved by constrained least squares against the TIL10 signature (170 genes from 51 immune-cell RNA-seq data sets; a 153-gene subset is the RNA-seq default), with mRNA-content scaling and a Treg dropout heuristic; image-derived total cell densities optionally convert fractions to cells/mm^2.
- Public paired-end Illumina RNA-seq of blood-derived immune cells (ten types) plus MCF7 breast and SW480 colorectal cancer cell lines, used to build the TIL10 signature · human · n = 51 immune-cell data sets; 53 libraries in expression matrix · CGHub G41726.MCF7.5; G27202.SW480.1
- Simulated breast-tumor RNA-seq mixing immune-cell reads with MCF7 reads at 11 purity levels (0-100%), plus depths of 1 to 100 million read pairs at 60% purity · human · n = 1,700 simulated data sets
- Healthy-donor PBMC spiked with polymorphonuclear cells (3-6% of total cells), profiled by RNA-seq and flow cytometry · human · n = nine donors · GSE107572
- Vanderbilt melanoma and lung cancer tumors: RNA Access RNA-seq plus CD4/CD8 IHC (and FOXP3 in melanoma) quantified with IHCount · human · n = 70 melanoma and 8 lung cancer samples
- Leiden colorectal cancer specimens: rRNA-depleted RNA-seq with IF (CD3, CD8) and IHC (CD4, FOXP3, CD20) quantified by automated pathology imaging · human · n = ten patients
- TCGA RNA-seq of 19 solid cancer types with survival, mutation, neoantigen, purity and MSI annotations · human · n = 8,243 tumors
- Published melanoma RNA-seq from biopsies taken before and during BRAF and/or MEK inhibitor therapy · human · n = seven patients
- External purified immune-cell expression sets used to filter signature genes (two microarray, one RNA-seq, CIBERSORT LM22 compendium) · GSE28490; GSE2849
- Methods & tools
- quanTIseq (Docker image), TIL10 immune signature matrix (170 genes; 153-gene RNA-seq default), Trimmomatic, Kallisto (hg19_M_rCRS reference), limSolve lsei constrained least squares, IHCount (Ilastik, CellProfiler, OME bftools), Vectra 3.0 with inFORM Cell Analysis, BD LSRFortessa flow cytometry with FlowLogic 7.1, Rtsne (t-SNE), R survival package (Kaplan-Meier, log-rank), pROC, GEOquery, biomaRt, Cancer Cell Line Encyclopedia (CCLE)
- Software & environment
- quanTIseq pipeline packaged as a Docker image; Kallisto with hg19_M_rCRS reference; HGNC symbols downloaded April 2017; FlowLogic 7.1; R packages limSolve, Rtsne, pROC, survival, GEOquery, biomaRt (versions not given).
- Reproducibility gaps
- No accession given for raw RNA-seq of the Vanderbilt melanoma/lung or Leiden CRC cohorts (only quanTIseq outputs for melanoma and lung are in TCIA); CRC cell counts used proprietary software; most tool versions unstated.
Limitations (6)
- Treg fractions agreed less well with FOXP3 IHC in melanoma, which the authors attribute to low Treg abundance and small sample size; Treg vs CD4+ T cell multicollinearity required an ad hoc re-estimation heuristic for fractions below 0.02. · stated by the authors
- The survival analyses could not use spatial information (tumor core vs invasive margin), which the authors think explains the null CRC result. TCGA cohorts also vary in treatment and staging, which may confound the survival associations. · stated by the authors
- The anti-PD1 responder comparison is underpowered (21 samples; differences p>0.09), and the Vanderbilt samples were collected based on tissue availability without a prespecified power analysis. · stated by the authors
- Accuracy may drop on microarray or pre-computed expression matrices, because signature genes can be missing and hybridization arrays have a different dynamic range than RNA-seq. · stated by the authors
- The CXCL9/CXCR3 finding is a cross-sectional bulk correlation. CXCR3 is expressed by the infiltrating CD8+ T cells themselves, so part of the association may reflect cell abundance rather than a causal recruitment axis. · noted in this summary
- The main text reports no head-to-head accuracy numbers against CIBERSORT, EPIC or TIMER on the new validation sets. It also says the TB score was prognostic in six cancers but names only five. · noted in this summary
Who should read:Read this if you need absolute immune-cell fractions from bulk tumor RNA-seq backed by flow cytometry and IHC/IF validation, or pan-cancer analyses linking chemokines, survival and drug treatment to immune infiltration.
Reproducibility ledger
Nobody has reported trying to reproduce this paper yet. Did you replicate an experiment or run the code? Your report saves the next lab a week.
Sign in to file a reproducibility report.
Related papers
- Infino: a Bayesian hierarchical model improves estimates of immune infiltration into tumor microenvironment
Maxim Zaslavsky, Jacqueline L. Buros, … Jeff Hammerbacher·bioRxiv·Nov 21, 2017
- Comprehensive evaluation of computational cell-type quantification methods for immuno-oncology
Gregor Sturm, Francesca Finotello, … Tatsiana Aneichyk·BioinformaticsPeer-reviewed·May 9, 2019
- GEM-DeCan: Improved tumor immune microenvironment profiling through novel gene expression and DNA methylation signatures predicts immunotherapy response
Ting Xie, Jacobo Solórzano, … Véra Pancaldi·bioRxiv·Apr 9, 2021
- Informatics for Cancer Immunotherapy
Jeff Hammerbacher, Alexandra Snyder·Annals of OncologyPeer-reviewed·Oct 18, 2017
- MIXTURE: an improved algorithm for immune tumor microenvironment estimation based on gene expression data.
Elmer A. Fernández, Yamil Damian Mahmoud, … María Romina Girotti·bioRxiv·Aug 7, 2019
- Estimation of immune cell content in tumour tissue using single-cell RNA-seq data
Max Schelker, Sonia Feau, … Andreas Raue·Nature CommunicationsPeer-reviewed·Dec 5, 2017