Signature scoring: Signature estimates were constructed as the median of z-scored (log2) expression values of each signature gene component except for the NK markers (see below).
TCD8 (CD8+ T cells): (CD8A, CD8B) Source: Mining of immune signatures in tumors using CD8A as sentinel marker. Reciprocal-Mutual-Rank methods were used to identify transcripts most intimately associated with sentinel markers. Caveats: CD8A is also expressed in a fraction of dendritic cells, some NK cells, and occasionally (rarely) in tumors.
Treg (Regulatory T Cells): (FOXP3, CCR8) Source: Mining of immune signatures in tumors using FOXP3 as sentinel marker. Reciprocal-Mutual-Rank methods were used to identify transcripts most intimately associated with sentinel markers. Caveats: Although CCR4 and CCR8 seem to be most predominantly co-expressed with FOXP3 in tumors, in sorted immune cells these receptors can also be seen in activated populations of CD4+ and CD8+ T cells.
Tcell (Pan T-Cell): (CD3D, CD3E, CD2) Mining of immune signatures in tumors using CD3 family members as sentinel markers. Reciprocal-Mutual-Rank methods were used to identify transcripts most intimately associated with CD3 epsilon (CD3E).
Bcell (B-cell): (CD19, CD79A, MS4A1) Source: Mining of immune signatures in tumors using CD19 as sentinel marker. Reciprocal-Mutual-Rank methods were used to identify transcripts most intimately associated with sentinel markers.
Mono (Monocyte lineage): (CD86, CSF1R, C3AR1) Source: Examination of correlation between antigen presenting cell-related genes across TCGA. Caveats: may not discriminate well between monocytes, macrophages, and other related members of the lineage.
M2mf (M2 Macrophage): (CD163, VSIG4, MS4A4A) Source: cross-referencing of Fantom/Hacohen/Rooney macrophage marker sets with mutual rank distance measures across TCGA[21 (link)]. The initial set was expanded with neighboring genes, cross-referenced with the literature and Mouse Immunological Genome Project (http://immgen.org ) expression profiles to reduce to a small list of macrophage markers.
NK (Natural Killer cells): (KIR2DL1, KIR2DL3, KIR2DL4, KIR3DL1, KIR3DL2, KIR3DL3, KIR2DS4) Source: Mutual-rank correlation analysis of Natural Killer Group (NKG) and Killer-Cell Immmunoglobulin-Like Receptor (KIR) receptor families in TCGA tumor data revealed co-regulation of multiple members of the KIR family. However, any specific KIR gene was often observed to be at the lower limit of detection set by the TCGA RNA-seq pipeline. Compared to other cellular signatures, a larger collection of (KIR) markers was selected, a mean instead of median summarization was used to estimate NK cell content, and a small Gaussian noise component was added (mean 0.16, standard deviation 0.08) to improve the normality of the NK signature score distribution.
TregCD8 and NKCD8 signatures were constructed by subtracting the TCD8 estimate from Treg estimate, or the TCD8 from the NK estimate, respectively.
TCD8 (CD8+ T cells): (CD8A, CD8B) Source: Mining of immune signatures in tumors using CD8A as sentinel marker. Reciprocal-Mutual-Rank methods were used to identify transcripts most intimately associated with sentinel markers. Caveats: CD8A is also expressed in a fraction of dendritic cells, some NK cells, and occasionally (rarely) in tumors.
Treg (Regulatory T Cells): (FOXP3, CCR8) Source: Mining of immune signatures in tumors using FOXP3 as sentinel marker. Reciprocal-Mutual-Rank methods were used to identify transcripts most intimately associated with sentinel markers. Caveats: Although CCR4 and CCR8 seem to be most predominantly co-expressed with FOXP3 in tumors, in sorted immune cells these receptors can also be seen in activated populations of CD4+ and CD8+ T cells.
Tcell (Pan T-Cell): (CD3D, CD3E, CD2) Mining of immune signatures in tumors using CD3 family members as sentinel markers. Reciprocal-Mutual-Rank methods were used to identify transcripts most intimately associated with CD3 epsilon (CD3E).
Bcell (B-cell): (CD19, CD79A, MS4A1) Source: Mining of immune signatures in tumors using CD19 as sentinel marker. Reciprocal-Mutual-Rank methods were used to identify transcripts most intimately associated with sentinel markers.
Mono (Monocyte lineage): (CD86, CSF1R, C3AR1) Source: Examination of correlation between antigen presenting cell-related genes across TCGA. Caveats: may not discriminate well between monocytes, macrophages, and other related members of the lineage.
M2mf (M2 Macrophage): (CD163, VSIG4, MS4A4A) Source: cross-referencing of Fantom/Hacohen/Rooney macrophage marker sets with mutual rank distance measures across TCGA[21 (link)]. The initial set was expanded with neighboring genes, cross-referenced with the literature and Mouse Immunological Genome Project (
NK (Natural Killer cells): (KIR2DL1, KIR2DL3, KIR2DL4, KIR3DL1, KIR3DL2, KIR3DL3, KIR2DS4) Source: Mutual-rank correlation analysis of Natural Killer Group (NKG) and Killer-Cell Immmunoglobulin-Like Receptor (KIR) receptor families in TCGA tumor data revealed co-regulation of multiple members of the KIR family. However, any specific KIR gene was often observed to be at the lower limit of detection set by the TCGA RNA-seq pipeline. Compared to other cellular signatures, a larger collection of (KIR) markers was selected, a mean instead of median summarization was used to estimate NK cell content, and a small Gaussian noise component was added (mean 0.16, standard deviation 0.08) to improve the normality of the NK signature score distribution.
TregCD8 and NKCD8 signatures were constructed by subtracting the TCD8 estimate from Treg estimate, or the TCD8 from the NK estimate, respectively.
Free full text: Click here
