Abstract
Tissues are complex milieu consisting of numerous cell-types. Numerous recent methods attempt to enumerate cell subsets from transcriptomes. However, available method used limited source for training and displayed only partial portrayal of the full cellular landscape. Here we present xCell, a novel gene-signature based method for inferring 64 immune and stroma cell-types. We harmonized 1,822 pure human cell-types transcriptomes from various sources, employed curve fitting approach for linear comparison of cell-types, and introduced a novel spillover compensation technique for separating between cell-types. Using extensive in silico analyses and comparison to cytometry immunophenotyping we show that xCell outperforms other methods.
Introduction
Understanding the cellular composition of tissues is fundamental to biology and medicine, particularly in oncology where the tumor microenvironment (TME) dictates therapeutic response. Tissues are not monolithic; they are intricate ecosystems of immune, stromal, and malignant cells. Traditional "deconvolution" methods aimed to estimate these populations from bulk RNA sequencing data but were often hampered by limited scope (focusing only on a few immune types), "cancer mimicry" (where malignant cells express genes typical of normal cells), and the "spillover effect" (difficulty distinguishing between closely related cell lineages).
Methods and Data Sources
The authors developed xCell, a computational tool designed to overcome these hurdles by integrating diverse datasets and novel mathematical corrections.
Signature Generation and Harmonization
The foundation of xCell is a massive compendium of 1,822 pure human cell-type transcriptomes integrated from six primary sources: FANTOM5, ENCODE, Blueprint epigenomics, IRIS, Novershtern et al., and the Human Primary Cells Atlas (HPCA). This diversity allows xCell to be robust against technical noise and batch effects inherent in single-platform studies.
From these sources, the team generated over 6,000 candidate gene signatures, ultimately selecting the 489 most reliable ones to represent 64 distinct cell types, including lymphoid, myeloid, stem, and stromal cells.
The xCell Pipeline
- Enrichment Scoring: Raw scores are calculated using single-sample Gene Set Enrichment Analysis (ssGSEA).
- Linear Transformation: To ensure scores reflect actual cell abundance, the researchers used curve-fitting from in silico simulations to linearize the non-linear ssGSEA outputs.
- Spillover Compensation: Borrowing from flow cytometry principles, xCell employs a "spillover matrix" to adjust for dependencies between related cell types. By mathematically "subtracting" overlapping signals, it provides a clearer, "deblurred" estimate for each specific population.
- Cancer Filtering: To mitigate false positives from malignant cells, genes overexpressed in 634 cancer cell lines (from CCLE) were filtered out.
Key Insights and Results
xCell's multi-source approach and unique compensation steps yield several advantages:
- Broad Landscape: Unlike previous tools like CIBERSORT, which focused primarily on 22 immune types, xCell provides insights into stromal cells (e.g., fibroblasts, endothelial cells), which are critical players in cancer progression.
- Superior Performance: In in silico benchmarks, xCell outperformed existing methods in 64 out of 67 comparisons.
- Validation against Ground Truth: Comparison with CyTOF (mass spectrometry) data from 165 individuals confirmed xCell's ability to identify specialized subsets, such as memory B-cells, where other methods struggled.
- Clinical Application: Applying xCell to nearly 10,000 TCGA samples revealed distinct "microenvironment fingerprints" for different cancers, such as unique stromal compositions in head and neck tumors.
Discussion and Limitations
While xCell represents a significant leap in digital tissue dissection, it provides enrichment scores rather than absolute cell percentages. Furthermore, while it performs exceptionally well on synthetic data, real-world samples present challenges as cell gene expression can be perturbed by the surrounding tissue environment. Future work involving single-cell RNA sequencing (scRNA-seq) will likely further refine these signatures and our understanding of cellular interactions within complex tissues.
For more information, visit the xCell web tool at http://xCell.ucsf.edu/.
Summary updated