Leukocytes

We performed two in silico experiments to assess the detection limits of different deconvolution algorithms. In the first experiment (Supplementary Fig. 6), we used the same cell line GEPs described above to compare CIBERSORT and RLR with five other GEP deconvolution methods^{4 (link)–8 (link)}. We evaluated detection limit using Jurkat cells (spike-in concentrations of 0.5%, 1%, 2.5%, 5%, 7.5%, and 10%), whose reference GEP (median of three replicates in GSE11103) was added into randomly created background mixtures of the other three blood cell lines. Five mixtures were created for each spike-in concentration. Predicted Jurkat fractions were assessed in the presence of differential tumor content, which we simulated by adding HCT116 (described above) in ten even increments, from 0% to 90%. Of note, we also used the same marker or signature genes described for simulated tumors (above). In a second experiment (Supplementary Fig. 7a), we compared CIBERSORT with QP^{5 (link)}, LLSR^{4 (link)}, PERT^{6 (link)}, and RLR. We spiked naïve B cell GEPs from the leukocyte signature matrix into four random background mixtures of the remaining 21 leukocyte subsets in the signature matrix. The same background mixtures were used for each spike-in. We also tested the addition of unknown content by adding defined proportions (0 to 90%) of randomly permuted expression values from a naïve B cell reference transcriptome (median expression profile from samples used to build LM22, Supplementary Table 1). We then repeated this analysis for each of the remaining leukocyte subsets in LM22 (Supplementary Fig. 7b).

In the following two sections, we describe how to create a custom leukocyte signature matrix and apply it to study cellular heterogeneity and TIL survival associations in melanoma tumors profiled by The Cancer Genome Atlas (TCGA). Readers can follow along by creating ‘LM6’, a leukocyte RNA-Seq signature matrix comprised of six peripheral blood immune subsets (B cells, CD8 T cells, CD4 T cells, NK cells, monocytes/macrophages, neutrophils; GSE60424 [20 ]). Key input files are provided on the CIBERSORT website (‘Menu>Download’).
A custom signature file can be created by uploading the Reference sample file and the Phenotype classes file (section 3.3.2) to the online CIBERSORT application (SeeFigure 2) or can be created using the downloadable Java package. To build a custom gene signature matrix with the latter, the user should download the Java package from the CIBERSORT website and place all relevant files under the package folder. To link Java with R, run the following in R:
Within R:

> library(Rserve)

> Rserve(args=“–no-save”)

Command line:

> java -Xmx3g -Xms3g -jar CIBERSORT.jar -M Mixture_file -P Reference_sample_file -c phenotype_class_file -f

The last argument (-f) will eliminate non-hematopoietic genes from the signature matrix and is generally recommended for signature matrices tailored to leukocyte deconvolution. The user can also run this step on the website by choosing the corresponding reference sample file and phenotype class file (seeFigure 2). The CIBERSORT website will generate a gene signature matrix located under ‘Uploaded Files’ for future download.
Following signature matrix creation, quality control measures should be taken to ensure robust performance (see ‘Calibration of in silico TIL profiling methods’ in Newman et al.) [17 (link)]. Factors that can adversely affect signature matrix performance include poor input data quality, significant deviations in gene expression between cell types that reside in different tissue compartments (e.g., blood versus tissue), and cell populations with statistically indistinguishable expression patterns. Manual filtering of poorly performing genes in the signature matrix (e.g., genes expressed highly in the tumor of interest) may improve performance.
To benchmark our custom leukocyte matrix (LM6), we compared it to LM22 using a set of TCGA lung squamous cell carcinoma tumors profiled by RNA-Seq and microarray (n = 130 pairs). Deconvolution results were significantly correlated for all cell subsets shared between the two signature matrices (P < 0.0001). Notably, since LM6 was derived from leukocytes isolated from peripheral blood [20 ,21 (link)], we restricted the CD4 T cell comparison to naïve and resting memory CD4 T cells in LM22. Once validation is complete, a CIBERSORT signature matrix can be broadly applied to mixture samples as described in section 3.3 (e.g., SeeFigure 4).

We applied LM22 to a publicly available dataset (GSE29832^{5 (link)}) to measure stability of deconvolution results over defined levels of blood admixed with breast tissue. To confirm reported fractions of blood admixed with breast tissue, we compared these proportions with an ‘LM22 normalized immune index’, defined for each sample as the median gene expression value of all genes in LM22 (Supplementary Table 1) divided by the median expression level of the transcriptome, and normalized into the range of known leukocyte content across the datasets (Fig. 2e). As a consistency metric, we compared deconvolution results for each sample with results from the sample with highest immune purity (Fig. 2e).