Analysis of immune subtypes across the epithelial-mesenchymal plasticity spectrum

Abstract

Epithelial-mesenchymal plasticity plays a critical role in many solid tumor types as a mediator of metastatic dissemination and treatment resistance. In addition, there is also a growing appreciation that the epithelial/mesenchymal status of a tumor plays a role in immune evasion and immune suppression. A deeper understanding of the immunological features of different tumor types has been facilitated by the availability of large gene expression datasets and the development of methods to deconvolute bulk RNA-Seq data. These resources have generated powerful new ways of characterizing tumors, including classification of immune subtypes based on differential expression of immunological genes. In the present work, we combine scoring algorithms to quantify epithelial-mesenchymal plasticity with immune subtype analysis to understand the relationship between epithelial plasticity and immune subtype across cancers. We find heterogeneity of epithelial-mesenchymal transition (EMT) status both within and between cancer types, with greater heterogeneity in the expression of EMT-related factors than of MET-related factors. We also find that specific immune subtypes have associated EMT scores and differential expression of immune checkpoint markers.

Keywords: Gene expression signature; Hybrid epithelial/mesenchymal cells; Immune invasion; Immune subtypes; Tumor heterogeneity.

Graphical abstract

Fig. 1

Carcinomas have more epithelial EMT scores. The KS algorithm scores samples on a scale of [-1, 1], with more positive scores representing more mesenchymal samples. (A) KS algorithm applied to RNA-Seq data from the Cancer Cell Encyclopedia (CCLE). (B) KS algorithm applied to RNA-Seq data from the Cancer Genome Atlas (TCGA); COAD = Colon adenocarcinoma, READ = Rectum adenocarcinoma, CESC = Cervical squamous cell carcinoma and endocervical adenocarcinoma, BLCA = Bladder urothelial carcinoma, STAD = stomach adenocarcinoma, ESCA = esophageal carcinoma, UCEC = uterine corpus endometrial carcinoma, LUAD = lung adenocarcinoma, HNSC = head and neck squamous cell carcinoma, PRAD = prostate adenocarcinoma, CHOL = cholangiocarcinoma, PAAD = pancreatic adenocarcinoma, KICH = kidney chromophobe, OV = ovarian serous cystadenocarcinoma, LUSC = lung squamous cell carcinoma, THCA = thyroid carcinoma, BRCA = breast invasive carcinoma, KIRP = kidney renal papillary cell carcinoma, LIHC = liver hepatocellular carcinoma, THYM = thymoma, LAML = acute myeloid leukemia, KIRC = kidney renal clear cell carcinoma, TGCT = testicular germ cell tumors, MESO = mesothelioma, UCS = uterine carcinosarcoma, DLBC = lymphoid neoplasm diffuse large B-cell lymphoma, UVM = uveal melanoma, SKCM = skin cutaneous melanoma, PCPG = pheochromocytoma and paraganglioma, LGG = brain lower grade glioma, SARC = sarcoma, GBM = glioblastoma multiforme. Violin plots show the median and interquartile range.

Fig. 2

EMT factors are more heterogeneous across cancers than MET factors. EMT and MET marker expression across TCGA tumor types; COAD = Colon adenocarcinoma, READ = Rectum adenocarcinoma, CESC = Cervical squamous cell carcinoma and endocervical adenocarcinoma, BLCA = Bladder urothelial carcinoma, STAD = stomach adenocarcinoma, ESCA = esophageal carcinoma, UCEC = uterine corpus endometrial carcinoma, LUAD = lung adenocarcinoma, HNSC = head and neck squamous cell carcinoma, PRAD = prostate adenocarcinoma, CHOL = cholangiocarcinoma, PAAD = pancreatic adenocarcinoma, KICH = kidney chromophobe, OV = ovarian serous cystadenocarcinoma, LUSC = lung squamous cell carcinoma, THCA = thyroid carcinoma, BRCA = breast invasive carcinoma, KIRP = kidney renal papillary cell carcinoma, LIHC = liver hepatocellular carcinoma, THYM = thymoma, LAML = acute myeloid leukemia, KIRC = kidney renal clear cell carcinoma, TGCT = testicular germ cell tumors, MESO = mesothelioma, UCS = uterine carcinosarcoma, DLBC = lymphoid neoplasm diffuse large B-cell lymphoma, UVM = uveal melanoma, SKCM = skin cutaneous melanoma, PCPG = pheochromocytoma and paraganglioma, LGG = brain lower grade glioma, SARC = sarcoma, GBM = glioblastoma multiforme (A) Normalized EMT and MET marker expression across all TCGA tumor types. (B) Pairwise correlations between expression values of all EMT-EMT, MET-MET, and EMT-MET factor pairs across a subset of TCGA tumor types. (C) Plot of mean pairwise correlation coefficients for all EMT-EMT factor pairs (x-axis) versus mean pairwise correlation coefficients for all MET-MET factor pairs (y-axis) across all TCGA tumor types.

Fig. 3

Immune subtypes are associated with EMT scores. EMT scores across TCGA immune subtypes; C1 = wound healing, C2 = IFN-γ dominant, C3 = inflammatory, C4 = lymphocyte depleted, C5 = immunologically quiet, C6 = TGF-β dominant. (A) Plot of calculated KS score across all immune subtypes. (B) Leave-one-out-analysis: pairwise comparison of each cancer immune subtype’s KS score to the KS scores of all other immune subtypes. (C) Application of Singscore to C5 immune subtype samples to quantify epithelial and mesenchymal scores separately. (D) Singscore epithelial and mesenchymal scores across immune subtypes.

Fig. 4

EMT and MET factor gene expression varies across immune subtypes. (A-E) Expression of EMT factors across immune subtypes; (F-J) Expression of MET factors across immune subtypes; (A) SNAI1, (B) SNAI2, (C) TWIST1, (D) ZEB1, (E) ZEB2, (F) GRHL2, (G) OVOL1, (H) OVOL2, (I) ESRP1, and (J) ESRP2 expression. Unless specified by n.s. (not significant), all comparisons are statistically reliable (p < 0.05; Wilcoxon rank means test with Benjamini-Hochberg adjustment for multiple comparisons).

Fig. 5

Immune checkpoint expression varies across immune subtypes. (A) CTLA4, (B) CD274, (C) LAG3, (D) HAVCR2, (E) CD47, (F) LGALS9, and (G) CD276 expression across immune subtypes. (H) Correlation matrix of EMT and pseudo-EMT scores with immune checkpoint markers.