Identification of an Immune-specific Class of Hepatocellular Carcinoma, Based on Molecular Features

Abstract

Background & aims: Agents that induce an immune response against tumors by altering T-cell regulation have increased survival times of patients with advanced-stage tumors, such as melanoma or lung cancer. We aimed to characterize molecular features of immune cells that infiltrate hepatocellular carcinomas (HCCs) to determine whether these types of agents might be effective against liver tumors.

Methods: We analyzed HCC samples from 956 patients. We separated gene expression profiles from tumor, stromal, and immune cells using a non-negative matrix factorization algorithm. We then analyzed the gene expression pattern of inflammatory cells in HCC tumor samples. We correlated expression patterns with the presence of immune cell infiltrates and immune regulatory molecules, determined by pathology and immunohistochemical analyses, in a training set of 228 HCC samples. We validated the correlation in a validation set of 728 tumor samples. Using data from 190 tumors in the Cancer Genome Atlas, we correlated immune cell gene expression profiles with numbers of chromosomal aberrations (based on single-nucleotide polymorphism array) and mutations (exome sequence data).

Results: We found approximately 25% of HCCs to have markers of an inflammatory response, with high expression levels of the CD274 molecule (programmed death-ligand 1) and programmed cell death 1, markers of cytolytic activity, and fewer chromosomal aberrations. We called this group of tumors the Immune class. It contained 2 subtypes, characterized by markers of an adaptive T-cell response or exhausted immune response. The exhausted immune response subclass expressed many genes regulated by transforming growth factor beta 1 that mediate immunosuppression. We did not observe any differences in numbers of mutations or expression of tumor antigens between the immune-specific class and other HCCs.

Conclusions: In an analysis of HCC samples from 956 patients, we found almost 25% to express markers of an inflammatory response. We identified 2 subclasses, characterized by adaptive or exhausted immune responses. These findings indicate that some HCCs might be susceptible to therapeutic agents designed to block the regulatory pathways in T cells, such as programmed death-ligand 1, programmed cell death 1, or transforming growth factor beta 1 inhibitors.

Keywords: Immune Checkpoint; Immune Regulation; Molecular Subgroups; Virtual Microdissection.

Figure 1.. Flow chart of the study.

A total of 956 HCC samples were used in this study. A training cohort (Heptromic) including 228 HCCs was virtually microdissected to identify an Immune class. Validation was then performed in 7 independent datasets.

Figure 2.. Identification of the Immune class of HCC.

(A) Consensus-clustered heatmap of HCC samples (training dataset, n=228) using exemplar genes of the immune expression pattern and refined by Random Forest. In the heatmap, high and low gene set enrichment scores are represented in red and blue, respectively. Positive prediction of signatures is indicated in red and absence in grey. Note: only the 28-gene signature will be shown in following heatmaps. Similar results were obtained with both signatures. (B) Representative images of immune cell infiltration, PD-1 and PD-L1 staining in a patient of the Immune class (M515) and a patient outside of the Immune class (B319). Images were captured with 20X.

Figure 3.. The Immune class contains two distinct microenvironment-based subtypes.

NTP analysis of whole-tumor gene expression data using a molecular signature able to capture activated inflammatory stromal response identified two distinct subtypes of Immune class – the Active (blue color bar) and the Exhausted (green color bar) Immune Response Subtypes. In the heatmap, high and low gene set enrichment scores are represented in red and blue, respectively; same representation is used for high and low gene expression. Positive prediction of signatures as calculated by NTP is indicated in red and absence in grey.

Figure 4.. Kaplan-Meier estimates of overall survival according to the immune response type status and robustness of the Immune class.

(A) Kaplan-Meier estimates of overall survival according to the Active Immune Response status in the Heptromic cohort (Active Immune Response vs rest plus Exhausted Immune Response). (B) Kaplan-Meier estimates of overall survival according to the Active Immune Response status in the TCGA cohort. (C) External validation of the Immune class was conducted in the publicly available TCGA dataset.

Figure 5.. Association of the Immune class with copy number aberrations, presence of neo-antigens and mutations in driver genes.

Patients of the Immune class showed significantly lower burden of gains (A) and losses (B), both broad (left panels) and focal (right panels).(C) The rate of mutations predicted to yield a neo-antigen was similar between the Immune class and the rest of the cohort and (D) between the two microenvironment-based subtypes. (E) Heatmap representation of the distribution of mutations in known driver genes between patients of the Immune class and rest of TGCA cohort.

Figure 6.. The intratumoral immune profile does not correspond to the immune infiltration of the surrounding non-tumoral liver.

(A) Gene expression of the tumor (upper panel) and matched surrounding non-tumoral liver (lower panel) was available for 167 patients of the Heptromic cohort (training dataset). Heatmap represents enrichment scores for immune signatures in the tumors (upper panel) and corresponding surrounding tissues (bottom panel). Multi-nodularity was more frequent in patients positive for the immune classifier [25/55 (45%) vs 24/110 (22%), p=0.01]. (B) Kaplan-Meyer estimates of overall survival according to the presence of the Immune Classifier in the surrounding liver.