Molecular and pharmacological modulators of the tumor immune contexture revealed by deconvolution of RNA-seq data

Abstract

We introduce quanTIseq, a method to quantify the tumor immune contexture, determined by the type and density of tumor-infiltrating immune cells. quanTIseq is based on a novel deconvolution algorithm for RNA sequencing data that was validated with independent data sets. Complementing the deconvolution output with image data from tissue slides enables in silico multiplexed immunodetection and provides an efficient method for the immunophenotyping of a large number of tumor samples.Cancer immunotherapy with antibodies targeting immune checkpoints has shown durable benefit or even curative potential in various cancers 1,2 . As only a fraction of patients are responsive to immune checkpoint blockers, efforts are underway to identify predictive markers as well as mechanistic rationale for combination therapies with synergistic potential. Thus, comprehensive and quantitative immunological characterization of tumors in a large number of clinical samples is of utmost importance, but it is currently hampered by the lack of simple and efficient methods. Therefore, we developed quanTIseq, a computational pipeline for the quantification of the Tumor Immune contexture using RNA-seq data and images of haematoxylin and eosin (H&E)-stained tissue slides (Fig. 1a). As part of quanTIseq, we first developed a deconvolution algorithm based on constrained least squares regression 12 . We then designed a signature matrix from a compendium of 51 RNA-seq data sets (Supplementary (Fig. 1c).To validate quanTIseq we first used both simulated data and published data. We simulated 1,700RNA-seq data sets from human breast tumors by mixing various numbers of reads from tumor and immune-cell RNA-seq data, considering different immune compositions and sequencing depths.quanTIseq obtained a high correlation between the true and the estimated fractions and accurately quantified tumor content (measured by the fraction of "other" cells) (Supplementary Figure 1). We then validated quanTIseq using experimental data from a previous study 13 , in which peripheral blood mononuclear cell (PBMC) mixtures were subjected to both, RNA-seq and flow cytometry. A high accuracy of quanTIseq estimates was also observed with this data set ( Fig. 1d and Supplementary Figure 2). Additionally, we successfully validated quanTIseq using two previous published data sets (Supplementary Figures 3 and 4).As most of the validation data sets available in the literature are based on microarray data or consider a limited number of phenotypes, we generated RNA-seq and flow cytometry data from mixtures of peripheral-blood immune cells collected from nine healthy donors. Flow cytometry was used to quantify all the immune sub-populations considered by quanTIseq signature matrix except macrophages, which are not present in blood. Comparison between quanTIseq cell estimates and flow cytometry fractions showed a high correlation at a single and multiple cell-type level ( Fig. 1e and Supplementary Figure 5).We then validated quanTIseq using two independent data sets. The first data...

Gene networks in cancer are biased by aneuploidies and sample impurities

“…When inferring gene networks in the context of cancer, it is important to not only keep in mind the potential technical variability between batches that may induce false positive correlations and hence edges when using network inference methods, but also the biologically intrinsic confounding factors of gene expression, like the one induced by DNA copy number changes 26,55 or a mixture of different proportions of different cell types 60,61 .…”

neoANT-HILL: an integrated tool for identification of potential neoantigens

“…Potential neoantigens arising from genes showing an expression level under 1 TPM are excluded. In addition, neoANT-HILL also offers the possibility of estimating quantitatively, via deconvolution, the relative fractions of tumor-infiltrating immune cell types through the use of quanTIseq [37].…”

Identification of non-cancer cells from cancer transcriptomic data

“…Out of the considered stromal populations, EPIC was recommended for the deconvolution of B cells, CD4 + and CD8 + T cells, NK cells, CAFs and endothelial cells [ 52 ]. quanTIseq [ 51 ] was the first method to derive its signature matrix entirely from bulk RNA-seq data of purified cell populations. Marker genes were selected based on their differential expression between cell types and filtered out if highly expressed in tumour cells.…”

“…It is usually based on least square regression to minimise the differences between the bulk expression values and the product of the reference expression profiles with the estimated fractions [ 57 ]. Tools implementing least square regression include PERT [ 58 ], DeconRNASeq [ 59 ], TIMER [ 60 ], EPIC [ 50 ], and quanTIseq [ 51 ]. Machine learning based on nu-support vector regression (nu-SVR) has also been applied in the context of partial deconvolution, such as CIBERSORT [ 48 ] and Mysort [ 61 ].…”

“…As a result, partial deconvolution approaches may produce under- or over-estimated cell fractions [ 57 ]. To address this, some methods avoid restricting the cell fraction estimation to the populations under consideration [ 50 , 51 , 69 ]. Instead, they estimate the fraction of uncharacterised cells within the tumour bulk to provide more accurate estimations.…”