Abstract
Summary
The heterogeneous cell types of the tumor-immune microenvironment (TIME) play key roles in determining cancer progression, metastasis and response to treatment. We report the development of TIMEx, a novel TIME deconvolution method emphasizing on estimating infiltrating immune cells for bulk transcriptomics using pan-cancer single-cell RNA-seq signatures. We also implemented a comprehensive, user-friendly web-portal for users to evaluate TIMEx and other deconvolution methods with bulk transcriptomic profiles.
Availability and implementation
TIMEx web-portal is freely accessible at http://timex.moffitt.org.
Supplementary information
Supplementary data are available at Bioinformatics online.
1 Introduction
The tumor-immune microenvironment (TIME) encompasses many heterogeneous cell types that engage in extensive crosstalk among the cancer, immune and stromal components. TIME plays a crucial role in determining cancer progression, metastasis and response to treatment. Recent advancement in single-cell transcriptomics (scRNA-seq) will be the ideal technique to characterize TIME; however, due to the cost and complicated experimental steps for sample preparation, it is not feasible to apply scRNA-seq in large-scale cohorts or existing patient samples (e.g. TCGA).
To overcome this limitation, several computational tools have been developed to deconvolute the different cell components from bulk transcriptomics profiles. Currently, the state-of-the-art computational tools for TIME deconvolution of bulk transcriptomics can be generally classified into two strategies (Finotello and Trajanoski, 2018): (i) gene set-based—gene set enrichment framework is used to compute the enrichment scores of immune cells represented as specific gene sets [e.g. ConsensusTME (Jiménez-Sánchez et al., 2019), xCell (Aran et al., 2017), MCP-counter (Becht et al., 2016) and ESTIMATE (Yoshida et al., 2013)]. (ii) Regression-based—given an additional signature matrix containing gene expression profiles from different immune cell types, these methods attempt to resolve the gene expression profiles of each immune cell present in the bulk expression profile by using various regression methods [e.g. Cibersort (Newman et al., 2015), TIMER (Li et al., 2020), EPIC (Racie et al., 2017) and quanTIseq (Finotello et al., 2019)]. As demonstrated by several benchmarking studies, none of the methods outperformed each other in different immune cell types estimation across multiple datasets (Sturm et al., 2019; Jiménez-Sánchez et al., 2019). However, ConsensusTME, which used an ensemble approach, was the best overall performing tool in three cancer-related benchmark datasets (Jiménez-Sánchez et al., 2019).
Here, we introduce TIMEx, a novel TIME deconvolution method for bulk transcriptomics using pan-cancer scRNA-seq signatures. TIMEx differs from the existing methods in the following aspects: (i) this is the first deconvolution method based on scRNA-seq cell-type signatures. (ii) The cell-type signatures were derived from multiple pan-cancer scRNA-seq studies. (iii) TIMEx web-portal provides a one-stop deconvolution site for users to simultaneously test 10 widely adopted methods and generates pairwise comparisons between these methods across different cell types. We will describe the TIMEx developmental workflow, the functionality of the web-portal, comparisons of TIMEx with other tools and an example illustrating TIMEx utility.
2 Timex methods
2.1 Extraction of cell-type signatures from pan-cancer scRNA-seq
We extracted cell-type signatures from Tumor Immune Single Cell Hub (TISCH), a recently published large-scale curated scRNA-seq database (Sun et al., 2020). All the data in TISCH have been uniformly processed, normalized, clustered and annotated using MAESTRO workflow (Wang et al., 2020). We focused the extraction of the TIMEx cell types from 47 scRNA-seq datasets that were profiled by six different platforms. In total, >1.35 million cells obtained from 427 patients across 16 solid cancer types were used in the analysis. For each cell type, we defined the gene markers as >2-fold-change and detected in >70% of the cells with a cell type. We performed a hierarchical filtering approach to select genes that most reliably represent a particular cell type. We defined 37 TIMEx cell-type signatures, which are classified into 4 main cell groups (malignant, immune, stromal and other), 14 major myeloid and lymphoid immune cell lineages (e.g. conventional CD4 T cells), 15 minor immune cell lineages (e.g. naïve CD4, effector CD4) and 4 stromal cell types. See Supplementary Methods for details.
2.2 Estimating cell types with TIMEx
We used single sample GSEA (ssGSEA) to score each sample based on the scRNA-seq cell-type signatures. ssGSEA is a robust approach and has been used in the other deconvolution methods (e.g. ESTIMATE, xCell and ConsensusTME). Because the enrichment scores are not linearly correlated with the abundance of cell types, we transformed TIMEx enrichment scores to linear scores using the previously published approach (Aran et al., 2017) to allow better interpretable comparisons between samples. In brief, we synthesized expression datasets with different percentage of cell mixtures mimicking serial dilution from purified peripheral blood mononuclear cells (PBMCs) to estimate the enrichment scores. Using these simulated mixtures, we calculated the power coefficient between the mixtures and enrichment scores for selected cell types. We then transformed the enrichment scores to linear scores applying power coefficient for each cell type. These linear scores can be directly compared between samples as estimates of cell-type abundance, such as calculating fold-change between samples. See Supplementary Methods for details.
2.3 Implementation of TIMEx and TIMEx web-portal
TIMEx is implemented in Python, using GSEApy (https://github.com/zqfang/GSEApy/) as the underlying ssGSEA for computing the enrichment of signatures. We obtained source codes of xCell, Cibersort and ConsensusTME from the authors and used ‘immunedeconv’ R package for the other methods (Sturm et al., 2019). The TIMEx web-portal is implemented in a Python API web server and is freely accessible at http://timex.moffitt.org.
2.4 Functionality of the TIMEx web-portal
Users can submit bulk transcriptomic profiles (either microarray or RNA-seq) to the TIMEx web-portal to estimate the immune cell types in their samples. Users will receive the TIMEx enrichment scores and abundances for the estimated immune cell types. Users will also receive immune cell types deconvoluted by other methods and can easily compare the outputs across methods to better interpret the results. In addition, we used ESTIMATE (Yoshida et al., 2013) to calculate the tumor tissue percentage in a particular bulk sample as ‘purity score’ to assist users in interpreting their results. See Supplementary Methods and TIMEx User Manual for details.
3 Validation and use case example
3.1 Purified PBMC datasets
We tested TIMEx estimation in two bulk RNA-seq of immune cell types collected from purified PBMCs: GSE107011 (Monaco et al., 2019) and E-MTAB-2319 (Bonnal et al., 2015) as validation. Overall, TIMEx correctly identified the immune cell types from these datasets. See Supplementary Results for details.
3.2 TCGA pan-cancer datasets
To compare TIMEx with existing tools in estimating various immune cell types, we performed comparative analyses of the state-of-the-art methods in 9898 patients across 31 TCGA cancer types. Normalized pan-cancer RNA-seq data were downloaded from cBioPortal and submitted to TIMEx web-portal for immune cell types deconvolution. For each cancer type, we performed pairwise comparisons between TIMEx and the other eight methods. We computed Pearson’s correlation coefficient for each method within each cell type. From the results, TIMEx estimation of immune cell types is highly correlated with ConsensusTME, followed by Cibersort (Absolute mode), xCell and TIMER in TCGA samples. For the 11 cell types commonly identified by TIMEx, ConsensusTME and xCell, the overlapping genes for these cell-type signatures ranges from 1.5 to 21%. This suggests that TIMEx signatures are capturing additional genes from scRNA-seq that are not included in ConsensusTME and xCell signatures, which are derived from bulk transcriptomics (see Supplemental Results). Overall, TIMEx and ConsensusTME performed similarly in estimating immune cell types in TCGA cohort.
3.3 Immune cell types deconvolution in pre- and on-immunotherapy melanoma patients
Immunotherapy has dramatically improved overall survival in multiple cancer types, including melanoma. To validate TIMEx on bulk RNA-seq from solid tumors, we analyzed the data from GSE91061, which contains 42 paired melanoma samples (24 responders and 18 non-responders) of pre-and on-therapy of nivolumab (Riaz et al., 2017). We used TIMEx to estimate the immune cell types in these patients and compared the fold-change between on-therapy versus pre-treatment samples for responders versus non-responders. We identified several immune cell types including DC, CD4eff, CD4Tn, CD8Tcm, CD8Teff, CD8Tem, MAIT, monocytes and Th17 that increased in responders compared to non-responders during treatment (Mann–Whitney–Wilcoxon test, P < 0.05, see Supplementary Results). Our analysis corroborates with Riaz et al. (2017) that responders’ tumors were characterized by an increased infiltration of immune cells, particularly CD8+ cytotoxic T cells.
In conclusion, we developed TIMEx, a novel TIME deconvolution method for bulk transcriptomics using pan-cancer scRNA-seq signatures. We also implemented a user-friendly TIMEx web-portal to allow users to analyze bulk transcriptomics and to compare TIMEx with other deconvolution methods for understanding TIME.
Supplementary Material
Acknowledgements
We thank Guillermo Gonzalez-Calderon for technical support.
Funding
This work was supported in part by the Biostatistics and Bioinformatics Shared Resource at the Moffitt Cancer Center [NCI P30 CA076292], NCI T32 CA233399 and the James and Esther King Biomedical Research Grant [7JK02].
Conflict of Interest: none declared.
Contributor Information
Mengyu Xie, Department of Biostatistics and Bioinformatics, H. Lee Moffitt Cancer Center and Research Institute, Tampa, FL 33612, USA.
Kyubum Lee, Department of Biostatistics and Bioinformatics, H. Lee Moffitt Cancer Center and Research Institute, Tampa, FL 33612, USA.
John H Lockhart, Department of Molecular Oncology, H. Lee Moffitt Cancer Center and Research Institute, Tampa, FL 33612, USA.
Scott D Cukras, Department of Biostatistics and Bioinformatics, H. Lee Moffitt Cancer Center and Research Institute, Tampa, FL 33612, USA.
Rodrigo Carvajal, Department of Biostatistics and Bioinformatics, H. Lee Moffitt Cancer Center and Research Institute, Tampa, FL 33612, USA.
Amer A Beg, Department of Immunology H. Lee Moffitt Cancer Center and Research Institute, Tampa, FL 33612, USA.
Elsa R Flores, Department of Molecular Oncology, H. Lee Moffitt Cancer Center and Research Institute, Tampa, FL 33612, USA.
Mingxiang Teng, Department of Biostatistics and Bioinformatics, H. Lee Moffitt Cancer Center and Research Institute, Tampa, FL 33612, USA.
Christine H Chung, Department of Head and Neck-Endocrine Oncology, H. Lee Moffitt Cancer Center and Research Institute, Tampa, FL 33612, USA.
Aik Choon Tan, Department of Biostatistics and Bioinformatics, H. Lee Moffitt Cancer Center and Research Institute, Tampa, FL 33612, USA.
References
- Aran D. et al. (2017) xCell: digitally portraying the tissue cellular heterogeneity landscape. Genome Biol., 18, 220. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Becht E. et al. (2016) Estimating the population abundance of tissue-infiltrating immune and stromal cell populations using gene expression. Genome Biol., 17, 218. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Bonnal R.J.P. et al. (2015) De novo transcriptome profiling of highly purified human lymphocytes primary cells. Sci. Data, 2, 150051. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Finotello F., Trajanoski Z. (2018) Quantifying tumor-infiltrating immune cells from transcriptomics data. Cancer Immunol. Immunother., 67, 1031–1040. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Finotello F. et al. (2019) Molecular and pharmacological modulators of the tumor immune contexture revealed by deconvolution of RNA‐seq data. Gen. Med., 11, 34. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Jiménez-Sánchez A. et al. (2019) Comprehensive benchmarking and integration of tumor microenvironment cell estimation methods. Cancer Res., 79, 6238–6246. [DOI] [PubMed] [Google Scholar]
- Li T. et al. (2020) TIMER2.0 for analysis of tumor-infiltrating immune cells. Nucleic Acids Res., 48, W509–W514. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Monaco G. et al. (2019) RNA-seq signatures normalized by mRNA abundance allow absolute deconvolution of human immune cell types. Cell Rep., 26, 1627–1640. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Newman A.M. et al. (2015) Robust enumeration of cell subsets from tissue expression profiles. Nat. Methods, 12, 453–457. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Racie J. et al. (2017) Simultaneous enumeration of cancer and immune cell types from bulk tumor gene expression data. Elife, 6, e26476. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Riaz N. et al. (2017) Tumor and microenvironment evolution during immunotherapy with nivolumab. Cell, 171, 934–949. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Sturm G. et al. (2019) Comprehensive evaluation of transcriptome-based cell-type quantification methods for immune-oncology. Bioinformatics, 35, i436–i445. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Sun D. et al. (2020) TISCH: a comprehensive web resource enabling interactive single-cell transcriptome visualization of tumor microenvironment. Nucleic Acids Res., 49, D1420–D1430. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Wang C. et al. (2020) Integrative analyses of single-cell transcriptome and regulome using MAESTRO. Genome Biol., 21, 198. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Yoshida K. et al. (2013) Inferring tumour purity and stromal and immune cell admixture from expression data. Nat. Commun., 4, 2612. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.