Single‐cell RNA‐seq and bulk RNA‐seq explore the prognostic value of exhausted T cells in hepatocellular carcinoma

Abstract

Hepatocellular carcinoma (HCC) remains a worldwide health problem. Mounting evidence indicates that exhausted T cells play a critical role in the progress and treatment of HCC. Therefore, a detailed characterisation of exhausted T cells and their clinical significance warrants further investigation in HCC. Based on the GSE146115, we presented a comprehensive single‐cell Atlas in HCC. Pseudo‐time analysis revealed that tumour heterogeneity progressively increased, and the exhausted T cells gradually appeared during tumour progression. Functional enrichment analysis revealed that the evolutionary process of exhausted T cells mainly contained the pathway of cadherin binding, proteasome, cell cycle, and T cell receptor regulation of apoptosis. In the International Cancer Genome Consortium database, we divided patients into three clusters with the T cell evolution‐associated genes. We found that the exhausted T cells are significantly related to poor outcomes through immunity and survival analysis. In The Cancer Genome Atlas database, the authors enrolled weighted gene co‐expression network analysis, univariate Cox analysis, and Lasso Cox analysis, then screened the 19 core genes in T cells evolution and built a robust prognostic model. This study offers a fresh view on evaluating the patients' outcomes from an exhausted T cells perspective and might help clinicians develop therapeutic systems.

Full text

Received: 26 April 2022
-
Revised: 8 June 2023
-
Accepted: 28 June 2023
-
IET Systems Biology
DOI: 10.1049/syb2.12072
ORIGINAL RESEARCH
Single‐cell RNA‐seq and bulk RNA‐seq explore the prognostic
value of exhausted T cells in hepatocellular carcinoma
Xiaolong Tang
1,2
|Yandong Miao
1,3
|Lixia Yang
4
|Wuhua Ha
1
|Zheng Li
5
|
Denghai Mi
1,4
1
The First Clinical Medical College, Lanzhou University, Lanzhou, China
2
The Second Department of Gastrointestinal Surgery, Afliated Hospital of North Sichuan Medical College, Nanchong, China
3
Department of Oncology, Yantai Afliated Hospital of Binzhou Medical University, The Second Clinical Medical College of Binzhou Medical University, Yantai, China
4
Gansu Academy of Traditional Chinese Medicine, Lanzhou, China
5
Institute of Modern Physics, Chinese Academy of Sciences, Lanzhou, China
Correspondence
Denghai Mi and Zheng Li.
Email: mi.dh@outlook.com and lizhenglys@126.
com
Funding information
Special Plan for Condition Construction of [Gansu
Provincial Scientic Research Institutes], Grant/
Award Number: 20JR10RA432; China Postdoctoral
Science Foundation, Grant/Award Number:
2019M663860
Abstract
Hepatocellular carcinoma (HCC) remains a worldwide health problem. Mounting evi-
dence indicates that exhausted T cells play a critical role in the progress and treatment of
HCC. Therefore, a detailed characterisation of exhausted T cells and their clinical sig-
nicance warrants further investigation in HCC. Based on the GSE146115, we presented
a comprehensive single‐cell Atlas in HCC. Pseudo‐time analysis revealed that tumour
heterogeneity progressively increased, and the exhausted T cells gradually appeared during
tumour progression. Functional enrichment analysis revealed that the evolutionary pro-
cess of exhausted T cells mainly contained the pathway of cadherin binding, proteasome,
cell cycle, and T cell receptor regulation of apoptosis. In the International Cancer
Genome Consortium database, we divided patients into three clusters with the T cell
evolution‐associated genes. We found that the exhausted T cells are signicantly related to
poor outcomes through immunity and survival analysis. In The Cancer Genome Atlas
database, the authors enrolled weighted gene co‐expression network analysis, univariate
Cox analysis, and Lasso Cox analysis, then screened the 19 core genes in T cells evolution
and built a robust prognostic model. This study offers a fresh view on evaluating the
patients' outcomes from an exhausted T cells perspective and might help clinicians
develop therapeutic systems.
KEYWORDS
bioinformatics, tumours
1
|
INTRODUCTION
Primary liver cancer is the sixth most commonly diagnosed
cancer and the third leading cause of cancer‐associated death
worldwide in 2020, with approximately 906,000 new cases and
830,000 deaths [1]. Hepatocellular carcinoma (HCC) accounts
for more than 90% of liver cancer cases [2]. Conventional
treatments include hepatic resection, liver transplantation, local
ablation with radiofrequency, and hepatic artery intervention,
such as transarterial chemoembolisation etc,. [3–5]. Currently,
systemic therapies including tyrosine kinase inhibitor, immune
checkpoint inhibitors,a and monoclonal antibodies have chal-
lenged the use of conventional therapies in the treatment of
liver cancer [6, 7]. However, as therapy advanced, the
Xiaolong Tang and Yandong Miao was contributed equally to this work.
This is an open access article under the terms of the Creative Commons Attribution‐NonCommercial‐NoDerivs License, which permits use and distribution in any medium, provided the
original work is properly cited, the use is non‐commercial and no modications or adaptations are made.
© 2023 The Authors. IET Systems Biology published by John Wiley & Sons Ltd on behalf of The Institution of Engineering and Technology.
228
-
IET Syst. Biol. 2023;17:228–244. wileyonlinelibrary.com/journal/syb2

prognosis of HCC patients is still poor, and 5‐year overall
survival rate is only about 10% [8]. Therefore, we need more
strategies to reveal the pathological process of tumour devel-
opment to help clinicians formulate therapeutic procedures
[9, 10].
Tumours have a high degree of heterogeneity, resulting in
people responding differently to the same therapy [11]. Be-
sides, multilesion HCC is frequent and may result from con-
current carcinogenesis, intrahepatic metastases, or clonal
spread [12]. Thus, lesions within the same tumour may exhibit
unique genetic changes, biological behaviours, and microenvi-
ronmental characteristics and react differentially to treatments.
Even tumour cells within the same lesion may have a variety of
somatic mutations [13]. Some researchers indicated that an
optimal strategy for multilesional HCC would manage all
subtype lesions into account [14]. Therefore, exploration of
intratumoral heterogeneity is a signicant issue in HCC.
Although much work has been spent explaining tumour
heterogeneity, our knowledge is still primarily conned to
tumour cells [15]. Recent evidence suggests that stromal cells,
including tumour‐inltrating immune cells, exhibit heteroge-
neity as well [16]. In addition, the tumour microenvironment
(TME) is becoming a more prominent target of pharmaco-
logical therapy methods [17]. The fact that naturally occurring
T cells with antitumour activity exist in human cancer has
rationalised immunotherapy's use in oncology. The correlation
between intratumoral T cell increase and better survival in
cancer indicates that tumour‐specic T cell activity plays a
critical anticancer function [18]. With the development of
single‐cell transcriptome analysis, T cells can be annotated and
analysed at a more detailed level in HCC progress. Notably,
exhausted T cells have previously been observed as a unique
dysfunctional cell lineage that develops during persistent in-
fections and malignancies [19]. The exhausted T cells exhibit
several distinct characteristics, including decreased effector
cytotoxicity, decreased cytokine production, and upregulated
multiple inhibitory molecular receptors, such as PDCD‐1,
LAG3, CTLA4 [20]. Moreover, many researchers suggested
that the exhausted T cells may be a crucial factor in the success
of immunotherapy and patient outcomes [21, 22]. In addition
to the classical markers to dene exhausted T cells, some
scholars have proposed that in a different eld, exhausted T
cells show many phenomena beyond the traditional denition
[23]. Seo, Hyungseok et al. reported that by changing the
phenotype and transcriptional prole of chimaeric antigen re-
ceptor (CAR) T cells, basic leucine zipper ATF‐like transcrip-
tion factor (BATF) could improve their anti‐tumour response
[24]. Recently, Grebinoski, Stephanie et al. indicated that LAG3
might be a core gene regulating the function of CD8
+
T cells,
and the deciency of LAG3 can reduce their exhaustion
phenotype, which reveals the possibility of LAG3 as a target
for immunotherapy [25]. Therefore, a thorough understanding
of exhausted T cells and further exploration of their new
crucial markers in oncology is critical for developing novel
immunotherapies for HCC patients.
In this study, we obtained both single‐cell RNA‐seq data
and bulk RNA‐seq data from public databases and analysed
them through various computational biology techniques. The
single‐cell RNA‐seq data was performed by Su, Xianbin et al.
[26], which contained four HCC patients. We further described
the tumour heterogeneity and explored the characterisation of
exhausted T cells. By correlating the single‐cell RNA‐seq
data with the ICGC database, we found that a high level of
exhausted T cells was signicantly associated with a low clinical
outcome. Moreover, in the TCGA database, we formed a new
prognostic model through the crucial genes in T cells evolu-
tion. The model is a stable measure to estimate the prognosis
and an accurate indicator for evaluating the exhausted T cells'
level in HCC patients. This investigation offers a fresh view on
evaluating the HCC patients' outcome from an exhausted T
cells perspective and might help clinicians develop therapeutic
systems. The ow chart of the study design and analysis is
shown in Figure 1.
2
|
MATERIALS AND METHODS
2.1
|
Data acquisition
The single‐cell RNA‐seq dataset of HCC (GSE146115) was
obtained from the Gene Expression Omnibus database (GEO,
https://www.ncbi.nlm.nih.gov/geo/) and has been described
by Su, Xianbin et al. [26] in their supplemental data. An
additional single‐cell RNA‐seq dataset of HCC (GSE151530)
described by Ma, Lichun et al. [27] was utilised for validation.
Bulk RNA‐seq data and the relevant patients' information of
HCC were obtained from The Cancer Genome Atlas database
(TCGA, https://portal.gdc.cancer.gov/, v29.0) and Interna-
tional Cancer Genome Consortium database (ICGC, https://
dcc.icgc.org/). The matrix les of RNA‐seq data for each
sample were collated and annotated onto the genome by R
software (version 4.0.2).
2.2
|
Quality control and single‐cell RNA‐
seq analysis
To explore the tumour cell heterogeneity in HCC cells, we
enrolled single‐cell RNA‐seq data from GSE146115. The
quality control (QC) process was conducted through R‐
package “Seurat” (version 3.0.1) [28]. It was considered low‐
quality and removed with less than 50 unique molecular
identiers (UMIs) or more than 5% mitochondrion‐derived
UMI count for single cells. The differences among the pa-
tients' batch were normalised with the Integrate function of
“Seurat”. The top 1500 variable genes were selected for further
analysis. Subsequently, principal component analysis (PCA) and
t‐distributed stochastic neighbour embedding (TSNE) were
enrolled to process the data, and the major cell clusters were
visualised through 2D TSNE plots [29]. The TSNE analysis
represents a non‐linear dimensionality reduction technique,
conceived by Laurens van der Maaten et al. in 2008. This
method is specically designed for the visualisation of high‐
dimensional data [30]. We employed the FindAllMarkers
TANG ET AL.
-
229

function to identify the markers of each cell cluster. The pri-
mary cell characters were then recognised through the markers
from the CellMarker database (http://biocc.hrbmu.edu.cn/
CellMarker/index.jsp) [31].
2.3
|
Pseudo‐time trajectory analysis
To discover the evolutionary process of tumour cells and T
cells during the tumour advancement, we visualised the tra-
jectories through 2D TSNE plots by the R‐package “monocle”
[32]. Then, we analysed the distribution of time, branching, and
cell clustering along these trajectories.
2.4
|
Functional enrichment analysis
To understand how the function of exhausted T cells changes
during tumour progression, we performed Gene Ontology
(GO) and Kyoto Encyclopaedia of Genes and Genomes
(KEGG) analysis based on the T cell evolution‐associated
genes through R‐package “clusterProler”, “enrichplot”,
“ggplot2”, and “org.Hs.for example,.db” [33, 34]. Then, we
used the Bioplanet model in Enrichr (https://maayanlab.
cloud/Enrichr/) [35] and ClueGO's immunoassay model in
Cytoscape (http://apps.cytoscape.org/apps/cluego) [36] to
explore the immunity function further.
2.5
|
Exploration of the relationship between
subtypes of T cells and prognosis of HCC
patients
To further explore the relationship between various subtypes
of T cells and outcomes of HCC patients, we clustered the
HCC patients in the ICGC database through T cell evolution‐
associated genes by R‐package “ConsensusClusterPlus” [37].
Then, Kaplan–Meier (K‐M) survival analysis, TME score
analysis, and immune checkpoints analysis were used to reveal
the clinicopathologic feature of each cluster of HCC patients
FIGURE 1 The ow chart of the study design and analysis.
230
-
TANG ET AL.

through R‐package “survival”, “survminer” [38], “estimate”,
and “limma” [39]. The TME score encompasses four distinct
metrics, namely Stromal Score, Immune Score, ESTIMATE
Score, and Tumour Purity. This algorithm primarily relies on
the “estimate” R package and employs ESTIMATE (Estima-
tion of Stromal and Immune cells in MAlignant Tumour tis-
sues using Expression data), a methodology predicated on
gene expression data, for the purpose of calculating cellular
composition and purity within tumour tissues [40].
2.6
|
Screening for the core genes in T cell
evolution and building the prognostic model
Weighted gene co‐expression network analysis (WGCNA)
was used on the basis of T cell evolution‐associated genes to
screen the core genes in T cells evolution. We constructed a
scale‐free co‐expression network in the TCGA database and
built some crucial modules associated with clinical features
through R‐package “WGCNA” [41]. An appropriate power
of β=6 was set to ensure a signed scale‐free co‐expression
gene network. Then, based on the result of WGCNA, we
further selected the differentially expressed genes (DEGs)
between tumour and adjacent tissues using the Wilcoxon
Test with R‐package “limma”. We set the criteria as |Log
Fold Change| ≥1 and false discovery rate (FDR)‐adjusted
p‐value ≤0.05. Subsequently, we enrolled univariate COX
analysis and least absolute shrinkage and selection operator
(LASSO) regression analysis in screening the core genes with
overall survival (OS). With the coefcients from the multi-
variate regression analysis and the expression level, we
calculated the signature of T cell evolution (TCESig) of
each patient by the following formula: TCESig (pa-
tient) =Pn
j¼1Coef j ∗Xj, TCESig represents the prognostic
risk score for each HCC patient, Coefj represents the coef-
cient, and Xj represents the expression level of every gene
in the signature. We treated the TCGA database as a train set
and treated the ICGC database as a test set. We adopted the
median TCESig in TCGA as a cutoff value and separated
HCC patients into the high‐risk and low‐risk groups in the
TCGA and ICGC database. Subsequently, we arrange pa-
tients based on their risk scores, positioning them from left
to right in ascending order of their scores. In accordance
with this sequence, we generate a heatmap of TCESig's gene,
risk curve, and survival state. By superimposing these three
graphical representations, we aim to more effectively eluci-
date the relationship between TCESig and patient prognosis
in both the training and testing sets. The R‐package “time-
ROC” was used to draw the Receiver Operating Character-
istic (ROC) curve for investigating the sensitivity and
specicity of the survival prediction by the TCESig [42].
Area Under Curve (AUC) delivered as an index of prognostic
accuracy. K‐M analysis was used to assess the survival rate
for each group. Through univariate and multivariate Cox
analysis, we veried whether TCESig was an independent
prognosis factor in HCC.
2.7
|
Validation of the relationship between
the TCESig and immunity characteristics
To explore the relationship between the TCESig and subtypes
of T cells, we examined the expression of TCESig's genes in
different T cell groups. TSNE analysis was used to show the
high expression region of TCESig's genes. TME score analysis,
immune cell inltration analysis, and immune checkpoints
analysis were used to reveal the immunity characteristics of
TCESig.
2.8
|
Exploration of the TCESig’ function
and prognostic value
Several exhausted T cells‐related gene sets were enrolled to
illustrate the TCESig's function through Gene Set Enrichment
Analysis (GSEA) (version 4.1.0) analysis [43]. Then, we exam-
ined the distribution of the TCESig in different clusters in the
ICGC database. Since the CD8A expression can reect the
CD8T cells' level, and the TCESig may reect the degree of
accumulation of exhausted T cells. The combination of CD8A
expression and TCESig may better reect the degree of effective
cytotoxic T cell deciency in HCC patients. Hence, we per-
formed survival analysis combining CD8A expression and the
TCESig through the R‐package “survival” and “survminer”.
2.9
|
Statistical analysis
We used the R software (version 4.0.2) and strawberry‐Perl
(version 5.30.0) to perform data processing. In all hypothesis
tests, p‐values ≤0.05 were considered signicant.
3
|
RESULTS
3.1
|
Tumour cell heterogeneity in HCC
cells
Based on GSE146115, a total of 16 samples from four HCC
patients were involved in this study, which contained 3200
single cells. Details of the patients are shown in Supplementary
Table 1 (Table S1). The QC standards are described in Mate-
rials and Methods, and 3198 single cells matched the QC
standards (Figure S1A‐C). According to the CellMarker data-
base, these cells were divided into 14 subgroups (Figure S1D).
In addition, the 14 subgroups were mainly annotated into
four main types, including hepatocytes, T cells, NK cells, and
monocyte (Figure 2a–b). Then, through the function of
monocle, we performed pseudo‐time analysis to infer the co‐
expression module and visualise the cell differentiation tra-
jectory. We found that the tumour heterogeneity gradually
increased with the advance of tumour, and various subtypes of
tumour cells and immune cells gradually appeared (Figure 2c–
e). The results indicated that tumour cells and related immune
TANG ET AL.
-
231

cells are constantly evolving during tumour progression.
However, the differences in functionality between the various
divisions of single‐cell subtypes remain unclear.
3.2
|
T cells tend to be exhausted in the
TME in HCC
As we all know, T cells play an essential role in tumour
progression. Since we found that T cells were divided into
various subtypes (Figure 2a–b), we isolated and analysed
them again. Through the QC process, 354 single T cells
matched the QC standards (Figure S2A‐C). Based on the
known markers mentioned in the CellMarker database, these
T cells were divided into four subgroups (Figure 3a;
Figure S2D). We found that the expression of PDCD1, an
important inhibitory immune checkpoint, was signicantly
increased in group 3 (Figure 3b). Therefore, we suspected
that the T cells of group 3 might be involved in exhausted T
cells. Subsequently, we investigated the crucial immune
checkpoints in the four cell groups (Figure 3c). Notably, we
found that some vital inhibitory checkpoints were also
upregulated in group 3, such as CTLA4, TIGIT, JAK1,
LAG3, YTHDF1, LDHA, and IFNG. Since these genes are
the markers of T cell exhaustion, the result indicated that T
cells from group 3 were exhausted in the TME. Then, we
used monocle to perform pseudo‐time analysis to infer the
co‐expression module and visualise the cell differentiation
trajectory (Figure 3d–f). We found that the various subtypes
of T cells gradually appeared with the advance of time, and
the T cells from group 3, which represent the exhausted T
cells, were distributed at the end of branch 2. These results
suggested that in TME of HCC, Exhausted T cells gradually
appeared with tumour progression.
3.3
|
Functional enrichment analysis
In order to understand how the function of exhausted T cells
changes during tumour progression, we performed a series of
functional enrichment analyses by the DEGs of branch 2 in T
cells differentiation, which contained the exhausted T cells.
The results of GO analysis contained cadherin binding, cell
adhesion molecule binding, and actin binding (Figure 4a). The
results of KEGG analysis contained regulation of actin cyto-
skeleton, proteasome, and cell cycle (Figure 4b). In addition,
the results of the Bioplanet model of Enrichr mainly contained
cell cycle and T cell receptor regulation of apoptosis
(Figure 4c). The results of Clue GO's immunoassay module in
cytoscape mainly contained positive regulation of leucocyte
FIGURE 2 Overview of single‐cell analysis in HCC tissues. (a) TSNE clustered all single tumour cells into four main types: hepatocytes, T cells, NK cells,
and monocytes. (b) TSNE clustered all single tumour cells into 14 subgroups. (c‐e) Pseudo‐time analysis revealed the differentiation trajectory of all single cells in
HCC, with each colour‐coded for pseudo‐time, cell types and subgroups. TSNE, t‐distributed stochastic neighbour embedding.
232
-
TANG ET AL.

chemotaxis, positive regulation of neutrophil chemotaxis, and
antigen processing and presentation of exogenous peptide
antigen via MHC class I (Figure 4d). These results shed light
on the potential biological behaviours that T cells may expe-
rience during differentiation in tumour progression. During the
tumour progression, antigen presentation and immune func-
tion were gradually enhanced with the increase of tumour‐
associated antigens. However, with the increase of tumour
mutation burden, T cells were gradually overwhelmed, leading
to the emergence of exhausted T cells, and nally, T cell
apoptosis receptors were gradually activated, increasing T cell
apoptosis.
3.4
|
Exhausted T cells are associated with
poor survival in HCC patients
To further investigate the relationship between various sub-
types of T cells and outcomes of HCC patients, we clustered
the HCC patients in the ICGC database using T cell evolution‐
associated genes. The patients tted into three clusters since
the results showed that when the number of sets is three, the
difference between clusters is obvious, and the difference
within clusters is inconspicuous (Figure 5a). K‐M survival
analysis revealed statistically signicant differences among the
three clusters (Figure 5b). The analysis suggested that the
FIGURE 3 T cells tend to be exhausted in the TME of HCC. (a) TSNE clustered T cells into four groups. (b) Violin plot showed that PDCD1 expression
was signicantly increased in group 3. (c) Bubble plot revealed that the vital inhibitory checkpoints were upregulated in group 3. The red square represents group
3. (d‐f) Pseudo‐time analysis revealed the differentiation trajectory of T cells in HCC, with each colour‐coded for branches, pseudo‐time and subgroups. The T
cells from group 3 were distributed at the end of branch 2.
TANG ET AL.
-
233

patients from cluster 1 (C1) had a worse prognosis than cluster
2 (C2) and cluster 3 (C3), and C2 had the best prognosis. We
further analysed the TME score of the three clusters. Using an
analysis of variance (ANOVA), we validated the differences in
the TME scores between the clusters. The result showed that
the ESTIMATE score, Immune score, and Stromal score of
C2 were the highest, and the tumour purity of C2 was the
lowest (Figure 5c–f). The variation in overall survival time
across distinct clusters reects the disparate prognosis associ-
ated with each cluster. As we all know, a favourable TME score
predicts a better outcome. According to the C2's best prog-
nosis, this result of C2's TME score analysis was consistent
with our expectations. Interestingly, the ESTIMATE score,
Immune score, and Stromal score of C1 were signicantly
higher than C3, and the tumour purity of C1 was signicantly
less than C3 (Figure S3A‐D). However, the prognosis of C1
was signicantly worse than C3 (Figure 5g). In order to further
explore the reasons behind this unusual phenomenon, we
conducted survival analysis on ESTIMATE score, Immune
score, Stromal score, and tumour purity. We found that the
Stromal score was signicantly related to prognosis, and a
higher Stromal score presents a better prognosis (Figure 5h).
Although there was no statistical difference in Immune score,
ESTIMATE score, and tumour purity, it could be seen that the
prognosis with high Immune and ESTIMATE score was better
than that with a low score, and the prognosis with low tumour
purity was better than that with high tumour purity
(Figure S3E‐G). In addition, CD8A is a crucial molecule of T
cell activation signal transduction and is considered a
favourable prognosis predictor. Likewise, we found that high
expression of CD8A was signicantly associated with the
favourable prognosis of HCC patients in the ICGC database
(Figure S3I). However, we found that CD8A was signicantly
higher in C1 than in C3 (Figure S3H). Therefore, we suspected
that other vital factors are affecting patient outcomes of C1.
Then, we further analysed the immune checkpoints of C1 and
C3 (Figure 5i), and it can be seen that many critical inhibitory
immune checkpoints, such as PDCD1, TIGIT, CTLA4, and
HAVCR2, were higher in C1 than C3, indicating that there
were more exhausted T cells in C1 than in C3, which may be
the reason for the poor prognosis of C1. Besides, through
GSEA analysis, we found that the CTLA4 pathway and PD 1
signalling were signicantly enriched in C1 (Figure 5j), further
suggesting a more accumulation of exhausted T cells in C1
than in C3.
3.5
|
Searching the core genes in T cell
evolution and building the prognostic model
We design the TCGA database as a train set and the ICGC
database as a test set. In the TCGA database, we enrolled
WGCNA to process the T cell evolution‐associated genes and
constructed some modules associated with clinical features
through a scale‐free co‐expression network (Figure S4). A total
of four modules were generated (Figure 6a). Both blue and
turquoise modules were signicantly associated with tumour
grade and stage. Hence, we chose the genes of blue and
FIGURE 4 Functional enrichment analysis of branch 2. (a) GO enrichment analysis revealed the primary biological function and processes (BP), cellular
component (CC), and molecular function (MF). (b) KEGG enrichment analysis revealed the primary pathways. (c) The Bioplanet model of Enrichr showed the
primary pathways. (d) Clue GO's immunoassay module revealed the main pathways of immunity.
234
-
TANG ET AL.

turquoise modules for further analysis. Then, using the Wil-
coxon Test, we screened out the DEGs in tumour tissue and
adjacent tissue from the two modules (Figure 6b–c). By uni-
variate COX analysis, we further identied the genes associated
with prognosis (Figure 6d). To further validate the accuracy of
our data, we veried these genes using an additional inde-
pendent single‐cell RNA‐seq dataset of HCC (GSE151530). As
anticipated, we observed that many of these genes were highly
expressed in exhausted T cells, reinforcing the reliability of our
initial results (Figure S5). Subsequently, according to Lasso
regression analysis, we nally found 19 genes. Then, using the
multivariate Cox analysis, the coefcients of the 19 genes were
calculated (Table S2). Finally, a T cell evolution signature
(TCESig) was established based on the 19 genes' expression
level and coefcients to calculate each patient's risk. We found
that there was a signicant association between the TCESig
and the clinical stage in the TCGA database (Table S3) and in
the ICGC database (Table S1). Then, we used the risk score of
FIGURE 5 Exhausted T cells are associated with poor survival in HCC patients from the ICGC database. (a) The consensus matrix showed that patients
tted into three clusters through T cell evolution‐associated genes. (b) K‐M survival curve revealed that the three clusters' prognosis were signicantly different.
(c‐f) Violin plot of tumour micro‐environment (TME) score analysis showed that the three clusters' ESTIMATE score, Immune score, Stromal score, and
Tumour purity were signicantly different. (g) K‐M survival curve revealed that the prognosis of patients in cluster 1 was signicantly worse than that of cluster
3. (h) K‐M survival curve revealed that the prognosis of patients with a high Stromal score was signicantly better than patients with a low Stromal score. (i) The
bar graph revealed that many critical inhibitory immune checkpoints were higher in cluster 1 than in cluster 3. (j) GSEA analysis displayed that the CTLA4
pathway and PD 1 signaling were signicantly enriched in cluster 1. *p<0.05, **p<0.01, ***p<0.001.
TANG ET AL.
-
235

TCESig to predict prognosis in both TCGA and ICGC data-
bases. The distribution of risk scores and survival time sug-
gested that patients' survival time decreased as the TCESig
increased (Figure 7a–c; Figure 7f–h). In the TCGA database,
the time‐dependent ROC of TCESig showed that the AUC of
1‐, 3‐, and 5‐year were 0.818, 0.775, and 0.761, respectively
(Figure 7d). In the ICGC database, the time‐dependent ROC
of TCESig showed that the AUC of 1‐, 3‐, and 5‐year were
0.695, 0.718, and 0.278, respectively (Figure 7i). K‐M cumu-
lative curves demonstrated that the overall survival time of
patients with the high‐risk score was signicantly shorter than
that of patients with low‐risk score (Figure 7e; Figure 7j). In
addition, compared with clinicopathology, univariate and
multivariate Cox analysis showed that the TCESig was an in-
dependent prognostic predictor in the TCGA database
(Figure 8a–b) and in the ICGC database (Figure 8c–d). These
results suggested that the TCESig may be a reliable prognostic
indicator in HCC.
3.6
|
Validation of the relationship between
the TCESig and exhausted T cells
We hypothesised that the TCESig is closely related to
exhausted T cells. Hence, we analysed the expression level of
the TCESig's genes in the four subtypes of T cells. The
results showed that the TCESig's genes were highly
expressed in group 3, which represent the exhausted T cells
(Figure 9a), suggesting a close relationship between TCESig
and exhausted T cells. In addition, through TSNE analysis,
we identied the location of exhausted T cells (Figure 3a)
and found that TCESig's genes were mainly highly expressed
in the region of exhausted T cells (Figure 9b–c). Immune cell
inltration analysis showed that most immune cell in-
ltrations did not differ statistically between high and low‐
risk groups (Figure 9d). TME score analysis also showed
no signicant difference between the high and low‐risk
groups (Figure S6A‐D). Then we analysed the relationship
FIGURE 6 Searching the essential genes in T cell evolution and building the prognostic model in the TCGA database. (a) WGCNA analysis showed that
four modules were generated, and blue and turquoise modules were signicantly associated with tumour grade and stage. (b‐c) Heatmap and volcano plot
showed the DEGs in tumour tissue and adjacent tissue from blue and turquoise modules. (d) Univariate COX analysis further identied the genes potentially
associated with OS.
236
-
TANG ET AL.

between the TCESig and the vital inhibitory immune
checkpoints. The vital inhibitory immune checkpoints refer
to essential molecular mechanisms that play a crucial role in
regulating the immune response. These checkpoints act as
negative regulators or “brakes” to prevent excessive or un-
intended activation of immune cells, ensuring the mainte-
nance of self‐tolerance and preventing autoimmunity. By
inhibiting immune cell activation, these checkpoints can also
contribute to immune evasion by cancer cells. We conducted
a Wilcoxon Test to analyse the differences between high and
low‐risk groups and used Spearman's correlation to
investigate the association between patient risk scores and
the expression of inhibitory immune checkpoints. The results
showed that the vital inhibitory immune checkpoints were
almost signicantly higher in the high‐risk group than in the
low‐risk group (Figure 10a–g) and positively correlated with
the risk scores (Figure 10h–n). These results indicated that
the TCESig was also an accurate indicator for evaluating the
exhausted T cells' level in hepatocellular carcinoma. These
results suggested that the genes in TCESig that led to poor
outcomes may be primarily associated with exhausted T cells,
not immune cell inltration.
FIGURE 7 survival analysis of TCESig in the TCGA and ICGC database. Heatmap, distribution of risk scores, survival time and status revealed that the
patients' survival time decreased as the TCESig increased in the TCGA database (a‐c) and the ICGC database (f‐h). The dashed lines present the cutoff value,
which divided HCC patients into low‐risk and high‐risk groups. Time‐dependent ROC curve showed that the AUC of TCESig of 1‐, 3‐, and 5‐year were 0.818,
0.775, 0.761 in the TCGA database (d) and were 0.695, 0.718, 0.278 in the ICGA database (i). The K‐M survival curve revealed that the prognosis of patients
with high risk was signicantly worse than those with low risk in the TCGA database (e) and ICGC database (j).
TANG ET AL.
-
237

FIGURE 8 Univariate and multivariate Cox analysis of TCESig in the TCGA and ICGC database. Compared with clinicopathology, such as age, gender,
and stage, univariate and multivariate Cox analysis showed that the TCESig was an independent prognostic predictor for HCC patients in the TCGA database (a,
b) and the ICGC database (c, d).
FIGURE 9 Validation of the relationship between the TCESig and exhausted T cells. (a) Bubble chart revealed that the TCESig's genes were highly
expressed in group 3, which represent the exhausted T cells. The red square indicates group 3 of T cells. (b, c) TSNE plot of T cells revealed that the TCESig's
genes were mainly highly expressed in the region of exhausted T cells. The red colour indicates a high expression of the relevant gene in the single T cell. The
dashed squares indicate the region of exhausted T cells. (d) The bar graph of immune cell inltration analysis revealed that most immune cell inltrations did not
differ statistically between high and low‐risk groups. *p<0.05, **p<0.01, ***p<0.001.
238
-
TANG ET AL.

3.7
|
Exploration of the TCESig’ function
and prognostic value
To further validate the relationship between TCESig and
exhausted T cells, we illustrated several exhausted T cells‐
related gene sets via GSEA analysis. The analysis showed
that the exhausted T cells‐related gene sets were signicantly
enriched in the high‐risk score group of TCESig (Figure 11a).
Consisting with the previous survival analysis result
(Figure 5b), an ANOVA analysis of the ICGC database
revealed that the risk‐score in C1 was signicantly higher than
in C2 and C3 (Figure 11b). CD8+T cells represent the primary
subtype of cytotoxic T cells. CD8A, as a marker for CD8+T
cells, provides a reliable measure of the quantity of cytotoxic T
cells. Because the combination of CD8A expression and
TCESig may better reect the degree of effective cytotoxic T
cell deciency in HCC patients, we performed survival analysis
combining CD8A expression and the TCESig. K‐M survival
analysis revealed CD8A Low/High risk group, which may have
the least effective cytotoxic T cells, and has a signicantly
worse prognosis with a median survival rate of fewer than
2 years in the TCGA database (Figure 11c). These results
further veried the relationship between the TCESig and
exhausted T cells and revealed the potential of TCESig in
predicting patient outcomes.
4
|
DISCUSSION
With the advancement of cell sequencing technology, more
information regarding the differentiation map and transcrip-
tional heterogeneity has been revealed [44]. Here, we found
that exhausted T cells are gradually present during HCC
progression and may be associated with poor patient
FIGURE 10 The relationship between the TCESig and the vital inhibitory immune checkpoints. (a‐g) Bar graphs indicated that except for LAG3, almost
vital inhibitory immune checkpoints were signicantly higher in the high‐risk group than in the low‐risk group. (h‐n) Correlation analysis revealed that except for
LAG3, almost vital inhibitory immune checkpoints also positively correlated with the risk scores of TCESig.
TANG ET AL.
-
239

outcomes through single‐cell RNA‐seq. We collected single‐
cell and bulk RNA‐seq data from public databases and
analysed them using various computational biology methods.
We investigated the heterogeneity of tumour and charac-
terised the exhausted T cells in HCC. The presence of
exhausted T cells was shown to be strongly linked with poor
clinical outcomes. Additionally, we developed a novel prog-
nostic model using critical genes involved in T cell evolution.
The model provides a consistent predictor of prognosis for
patients with HCC. Interestingly, all model genes are signif-
icantly expressed in exhausted T cells, further indicating that
exhausted T cells are associated with a poor prognosis for
HCC patients. At last, we revealed that the model might also
be an accurate indicator for evaluating the exhausted T cells'
level. Collectively, this study provides a novel viewpoint on
assessing the outcome of HCC patients from an exhausted T
cell perspective, which may aid physicians in developing
treatment approaches.
Through TSNE and pseudo‐time plot, we found that the
tumour heterogeneity gradually increased with HCC progres-
sion, and various subtypes of immune cells gradually appeared.
The results indicated that tumour cells and related immune
cells are constantly evolving during tumour progression. These
ndings corroborated with recent research that demonstrated
the utility of single‐cell data analysis in dissecting tissue into
inltration and tumour cells, elucidating the critical roles of
humoral immunity inltration in the landscape of HCC prog-
ress [45]. The pseudo‐time analysis provided the trajectory of T
cell differentiation to the exhaustion phenotype in HCC and
revealed the genes and related pathways involved in this pro-
cess. Blocking these genes or pathways may provide the pos-
sibility to interrupt T cell dysfunction and restore T cell activity
and may contribute to an opportunity for nding new immu-
notherapies for HCC. We found that exhausted T cells grad-
ually appeared during tumour progression and performed
different gene‐expression proling. These results agree with
FIGURE 11 Exploration of the TCESig’ function and prognostic value. (a) GSEA analysis suggested that TCESig was associated with the gene set of
exhausted T cells elevation. (b) The risk score of TCESig in C1 was signicantly higher than C2 and C3 in the ICGC database. (c) K‐M survival analysis revealed
CD8A Low/High risk group has a signicantly worse prognosis with a median survival rate of fewer than 2 years in the TCGA database.
240
-
TANG ET AL.

previous ndings which showed that accumulation of
exhausted T cells increased in tumour progression, and
expression of various inhibitory immune checkpoints was
enhanced in exhausted T cells [46]. Likewise, Zheng, Bo et al.
indicated that tumour‐associated T cells performed a robust
function in HCC, and the existence of T cells with high
PDCD‐1 was conrmed by multiplex immunouorescence
tissue staining, and its enrichment showed denite prognostic
value [47]. These results contribute to our understanding of the
dynamic framework of T cells' function in HCC and may help
develop rational cancer immunotherapies.
The GO and KEGG analysis explored the primary func-
tion and pathways of the evolutionary branch of exhausted T
cells, which contains cadherin binding, proteasome, and cell
cycle. These ndings may reveal the potential functional
changes during the T cells differentiation. Consistent with the
literature, Banh, CIndy et al. conrmed that endogenous
cadherin can inhibit T cell functions [48]. These results also
corroborate the ideas of Widjaja, Christella E et al., who
indicated that proteasome activity plays a critical role in
determining the fate of T cells [49]. In addition, He, Tiansheng
et al. suggested that proteasome activity regulates T cell acti-
vation and apoptosis [50]. The analysis of the Bioplanet model
of Enrichr mainly contained cell cycle and T cell receptor
regulation of apoptosis, suggesting that the cell cycle of T cells
in exhausted T branch was signicantly changed, and the cell
receptors regulating apoptosis were elevated. The results of
Clue GO's immunoassay module mainly contained positive
regulation of leucocyte chemotaxis, positive regulation of
neutrophil chemotaxis, and antigen processing and presenta-
tion of exogenous peptide antigen via MHC class I, revealing
the progress that during the tumour advance, with the addition
of tumour‐associated antigens, antigen presentation and
immunological response are steadily increased. However, as the
tumour mutation burden increased, T cells became increasingly
overwhelmed, resulting in the appearance of exhausted T cells.
Finally, T cell apoptosis receptors were gradually activated,
resulting in increased T cell apoptosis. These results are
consistent with the commonly accepted denition of exhaus-
ted T cells [23], conrming the accuracy of our analysis.
All HCC patients in the ICGC database were divided into
three clusters through the T cell evolution‐associated genes.
Comparing prognosis and TME score in each cluster, we were
surprised to nd a contradiction in these results that C1 had a
worse prognosis than C3, while C1 had a better score of TME
and a higher CD8A level than C3. Many studies have proved
that a low TME score or CD8A level is associated with poor
prognosis in cancer patients [51–54]. Likewise, we also found
that patients with a high TME score or a high CD8A level have
a better prognosis than patients with a low TME score or a low
CD8A level in the ICGC database. Through detailed analysis,
numerous crucial inhibitory immunological checkpoints were
discovered to be higher in C1 than in C3, and the CTLA4
pathway and PD 1 signaling were signicantly enriched in C1.
Many studies have proved that exhausted T cells are signi-
cantly associated with the poor prognosis of patients in a va-
riety of tumours [55–57]. Recently, Barsch, Maryam et al.
reported that in HCC patients, exhausted CD8+T cells indi-
cate a worse prognosis than tissue‐resident memory T cells
[58]. Likewise, Hung, Man Hsin et al. also reported that
exhausted T cells are related to the poor prognosis of HCC
patients, and the exhausted T cells are driven by tumour
methionine metabolism [59]. These ndings revealed that C1
had a higher proportion of exhausted T cells than C3, which
may account for C1's poor prognosis. Our data are consistent
with the report of Yang, Yanying et al. that the accumulation of
exhausted T cells leads to a worse prognosis in HCC patients
[46]. Besides, consistency with other studies ensured the reli-
ability of subsequent analyses.
Through WGCNA analysis, differential expression analysis
and COX analysis in the TCGA database, we found 19 core
genes in T cell evolution and constructed a prognostic model
as TCESig. We design the TCGA database as a train set and the
ICGC database as a test set. TCGA and ICGC are two inde-
pendent datasets without any common samples between them.
The result of survival analysis showed that TCESig was a
robust prognostic indicator in both training and test sets. In
addition, our model's 1‐, 3‐, and 5‐year AUC values were more
accurate than the model of Zhang, Fapeng et al. built by eight
TME‐related genes in HCC [60]. Recently, Qin, Ge et al.
demonstrated that NPM1, a gene of our model, selectively
binds to the PD‐L1 promoter in triple‐negative breast cancer
cells and activates its transcription, thus suppressing the T cell
function in vitro and in vivo [61]. Ternette, Nicola et al. re-
ported that CFL1's potential as a target in breast cancer
immunotherapy should be further investigated [62]. Song,
Kwon Ho et al. found that HSP90AA1 inhibition sensitised
immune‐refractory tumours to adoptive T cell transfer and
PD‐1 blocking, hence re‐energising tumour‐reactive T cells'
immunological cycle [63]. Zhao, Ende et al. reported that
EZH2, a methyltransferase, activated the Notch pathway by
inhibiting the Notch repressors Numb and Fbxw7 by trime-
thylation of histone H3 at Lys27, which resulted in increased
polyfunctional cytokine production and survival of T cells via
Bcl‐2 signaling. Furthermore, he suggested that EZH2 (+)CD8
(+) T cells were correlated with increased survival of patients
[64]. Although the results of Zhao, Ende et al. differ from the
study of Zhou, Liye et al. that EZH2 may be a therapeutic
target since it has been shown to increase tumour cell antigen
presentation and sensitize refractory tumours to anti‐PD‐1
treatment [65]. They are consistent with that EZH2 is an
important gene for patients' outcomes through affecting the
function of T cells. However, the relationship between some
other genes in TCESig and T cells has not been fully discov-
ered yet. These genes may also be involved in the differenti-
ation progress of exhausted T cells. Further analysis of these
genes may provide new targets for cancer immunotherapy.
Our results showed that the TCESig's genes were highly
expressed in subgroup 3, which represents the exhausted T
cells, indicating that the genes of TCESig are closely related to
exhausted T cells. Tumour‐inltrating cytotoxic T cells can
inhibit tumour development specically. However, they often
go into a condition of “exhaustion” or “dysfunction” [66].
Hence, in recent years, immuno‐checkpoint blocking treatment
TANG ET AL.
-
241

to correct T cell malfunction and exhaustion has gotten much
interest [67]. Unfortunately, only a tiny percentage of cancer
patients benet from it in clinical trials [68]. Only 15%–20% of
individuals with HCC responded to PD1 blocking [69]. One
possible explanation is that anti‐PD1 therapy alone only helps
the less exhausted T cells [70], indicating that inhibiting a single
immune checkpoint is inefcient in practice. Therefore, it is
essential to identify more potential therapeutic targets and
conduct comprehensive evaluation and treatment. The genes in
our model may provide a new chance to get some novel
therapeutic targets.
Our results showed that the TCESig was also an accurate
indicator for evaluating the exhausted T cells' level in hepa-
tocellular carcinoma. As we all know, the therapeutic window
has been further widened by immunotherapy, including nivo-
lumab alone or in combination with ipilimumab or pem-
brolizumab. Immunotherapy based on the use of immune
checkpoint inhibitors, as single agents or in combination with
kinase inhibitors, anti‐angiogenic drugs, chemotherapeutic
agents, and locoregional therapies, offers great promise in the
treatment of HCC paralleling a lack of patients' selection
criteria [71, 72]. Previous study indicated that preexisting
exhausted T cells may have limited the ability to reinvigorate
post‐therapy [73]. The response of T cells to checkpoint
blockade is from a specic subset of T cell clones that recently
entered the tumour, and T cells clonal replacement was
shown to be more common in exhausted T cells [73]. Given
that our model performed well in predicting the accumulation
of exhausted T cells and considering that exhausted T cells
often show an inadequate response to checkpoint bloc-
kade, The model might suggest potential avenues for those
involving the use of immune checkpoint inhibitors in HCC
patients.
There are certain aws in the current study that should be
mentioned. The results of this study may have an extraordinary
divergence due to the limited number of patients included in
the analysis. Another vulnerability source assumes that we
should manage more functional experiments to indicate the
potential molecular mechanisms for predicting the effect of the
genes in TCESig.
5
|
CONCLUSIONS
In conclusion, we used single‐cell RNA‐seq and bulk RNA‐
seq to show that exhausted T cells are present during HCC
progression and are correlated with poor patient outcomes.
We investigated the heterogeneity of tumour and charac-
terised the exhausted T cells in HCC. What is more, we
created a new prediction model based on crucial genes
involved in T cell differentiation. For patients with HCC, the
model is a reliable predictor of prognosis and accumulation
of exhausted T cells. This work offers a unique perspective
on analysing the outcome of HCC patients from the stand-
point of exhausted T cells, which may help physicians create
treatment strategies and provide new therapeutic targets for
further research.
AUTHOR CONTRIBUTION
Xiaolong Tang contributed to the conceptualization, data
curation, formal analysis, software, visualization, and writing‐
original draft. Yandong Miao and Wuhua Ha performed the
validation of this study. Yandong Miao wrote the writing –
review & editing. Lixia Yang and Denghai Mi conducted the
resources and supervision. Zheng Li contributed to the
funding acquisition, investigation, and methodology. Denghai
Mi conducted the project administration.
ACKNOWLEDGEMENTS
We thank the GEO, TCGA, ICGC, Ensembl, David, KEGG,
and CellMarker databases for the availability of the data. This
work was supported by the Special Plan for Condition Con-
struction of [Gansu Provincial Scientic Research Institutes]
(Grants No. 20JR10RA432) and the China Postdoctoral Sci-
ence Foundation (Grant No. 2019M663860).
CONFLICT OF INTEREST STATEMENT
No potential conicts of interest were disclosed.
DATA AVAILABILITY STATEMENT
The data used to support the ndings of this study can be
found in Single‐cell RNA‐seq data and were collected from the
GEO database (GSE146115 and GSE151530) (https://www.
ncbi.nlm.nih.gov/geo/). Bulk RNA‐seq data and clinical data
were collected from TCGA (https://portal.gdc.cancer.gov/
v29.0) and ICGC (https://dcc.icgc.org/) databases.
ORCID
Denghai Mi
https://orcid.org/0000-0002-8643-4496
REFERENCES
1. Sung, H., et al.: Global cancer statistics 2020: GLOBOCAN estimates of
incidence and mortality worldwide for 36 cancers in 185 countries. CA A
Cancer J. Clin. 71(3), 209–249 (2021). https://doi.org/10.3322/caac.
21660
2. Kulik, L., El‐Serag, H.B.: ‘Epidemiology and management of hepato-
cellular carcinoma. Gastroenterology 156(2), 477–491.e1 (2019). https://
doi.org/10.1053/j.gastro.2018.08.065
3. Wang, Y., et al.: Hepatic resection versus transarterial chemoembolization
in inltrative hepatocellular carcinoma: a multicenter study. J. Gastro-
enterol. Hepatol. 35(12), 2220–2228 (2020). https://doi.org/10.1111/
jgh.15060
4. Samuel, D., Coilly, A.: Management of patients with liver diseases on the
waiting list for transplantation: a major impact to the success of liver
transplantation. BMC Med. 16(1), 113 (2018). https://doi.org/10.1186/
s12916‐018‐1110‐y
5. Kim, N., et al.: Stereotactic body radiation therapy vs. radiofrequency
ablation in Asian patients with hepatocellular carcinoma. J. Hepatol.
73(1), 121–129 (2020). https://doi.org/10.1016/j.jhep.2020.03.005
6. Xu, F., et al.: Immune checkpoint therapy in liver cancer. J. Exp. Clin.
Cancer Res. 37(1), 110 (2018). https://doi.org/10.1186/s13046‐018‐
0777‐4
7. Gordan, J.D., et al.: Systemic therapy for advanced hepatocellular carci-
noma: ASCO guideline. J. Clin. Oncol. 38(36), 4317–4345 (2020).
https://doi.org/10.1200/jco.20.02672
8. Wei, L., et al.: ‘Genome‐wide CRISPR/Cas9 library screening identied
PHGDH as a critical driver for Sorafenib resistance in HCC. Nat.
Commun. 10(1), 4681 (2019). https://doi.org/10.1038/s41467‐019‐
12606‐7
242
-
TANG ET AL.

9. Tsilimigras, D.I., et al.: Hepatocellular carcinoma tumour burden score to
stratify prognosis after resection. Br. J. Surg. 107(7), 854–864 (2020).
https://doi.org/10.1002/bjs.11464
10. Khan, A.R., Wei, X., Xu, X.: Portal vein tumor thrombosis and hepa-
tocellular carcinoma ‐the changing tides. J. Hepatocell. Carcinoma 8,
1089–1115 (2021). https://doi.org/10.2147/jhc.s318070
11. Dagogo‐Jack, I., Shaw, A.T.: Tumour heterogeneity and resistance to
cancer therapies. Nat. Rev. Clin. Oncol. 15(2), 81–94 (2018). https://doi.
org/10.1038/nrclinonc.2017.166
12. Feo, F., Pascale, R.M.: Multifocal hepatocellular carcinoma: intrahepatic
metastasis or multicentric carcinogenesis? Ann. Transl. Med. 3(1), 4
(2015)
13. Zhang, Q., et al.: Integrated multiomic analysis reveals comprehensive
tumour heterogeneity and novel immunophenotypic classication in
hepatocellular carcinomas. Gut 68(11), 2019–2031 (2019). https://doi.
org/10.1136/gutjnl‐2019‐318912
14. Liu, F., et al.: Identication of FABP5 as an immunometabolic marker in
human hepatocellular carcinoma. J Immunother Cancer 8(2), e000501
(2020). https://doi.org/10.1136/jitc‐2019‐000501
15. Patel, A.P., et al.: ‘Single‐cell RNA‐seq highlights intratumoral hetero-
geneity in primary glioblastoma. Science 344(6190), 1396–1401 (2014).
https://doi.org/10.1126/science.1254257
16. Papalexi, E., Satija, R.: Single‐cell RNA sequencing to explore immune
cell heterogeneity. Nat. Rev. Immunol. 18(1), 35–45 (2018). https://doi.
org/10.1038/nri.2017.76
17. Roma‐Rodrigues, C., et al.: Targeting tumor microenvironment for
cancer therapy. Int. J. Mol. Sci. 20(4), E840 (2019). https://doi.org/10.
3390/ijms20040840
18. O’Donnell, J.S., Teng, M.W.L., Smyth, M.J.: Cancer immunoediting and
resistance to T cell‐based immunotherapy. Nat. Rev. Clin. Oncol. 16(3),
151–167 (2019). https://doi.org/10.1038/s41571‐018‐0142‐8
19. Alfei, F., et al.: ‘TOX reinforces the phenotype and longevity of
exhausted T cells in chronic viral infection. Nature 571(7764), 265–269
(2019). https://doi.org/10.1038/s41586‐019‐1326‐9
20. Zeng, Z., Wei, F., Ren, X.: Exhausted T cells and epigenetic status.
Cancer Biol Med 17(4), 923–936 (2020). https://doi.org/10.20892/j.issn.
2095‐3941.2020.0338
21. Kallies, A., Zehn, D., Utzschneider, D.T.: Precursor exhausted T cells: key
to successful immunotherapy? Nat. Rev. Immunol. 20(2), 128–136
(2020). https://doi.org/10.1038/s41577‐019‐0223‐7
22. Hsu, C.‐L., et al.: Exploring markers of exhausted CD8 T cells to predict
response to immune checkpoint inhibitor therapy for hepatocellular
carcinoma. Liver Cancer 10(4), 346–359 (2021). https://doi.org/10.
1159/000515305
23. Blank, C.U., et al.: ‘Dening “T cell exhaustion. Nat. Rev. Immunol.
19(11), 665–674 (2019). https://doi.org/10.1038/s41577‐019‐0221‐9
24. Seo, H., et al.: BATF and IRF4 cooperate to counter exhaustion in
tumor‐inltrating CAR T cells. Nat. Immunol. 22(8), 983–995 (2021).
https://doi.org/10.1038/s41590‐021‐00964‐8
25. Grebinoski, S., et al.: Autoreactive CD8+T cells are restrained by an
exhaustion‐like program that is maintained by LAG3. Nat. Immunol.
23(6), 868–877 (2022). https://doi.org/10.1038/s41590‐022‐01210‐5
26. Su, X., et al.: Clonal evolution in liver cancer at single‐cell and single‐
variant resolution. J. Hematol. Oncol. 14(1), 22 (2021). https://doi.
org/10.1186/s13045‐021‐01036‐y
27. Ma, L., et al.: Single‐cell atlas of tumor cell evolution in response to
therapy in hepatocellular carcinoma and intrahepatic chol-
angiocarcinoma. J. Hepatol. 75(6), 1397–1408 (2021). https://doi.org/10.
1016/j.jhep.2021.06.028
28. Butler, A., et al.: Integrating single‐cell transcriptomic data across
different conditions, technologies, and species. Nat. Biotechnol. 36(5),
411–420 (2018). https://doi.org/10.1038/nbt.4096
29. Kobak, D., Berens, P.: ‘The art of using t‐SNE for single‐cell tran-
scriptomics. Nat. Commun. 10(1), 5416 (2019). https://doi.org/10.1038/
s41467‐019‐13056‐x
30. Maaten, L.V.D., Hinton, G.: ‘Visualizing data using t‐SNE. J. Mach.
Learn. Res. 9, 2579–2625 (2008)
31. Zhang, X., et al.: CellMarker: a manually curated resource of cell markers
in human and mouse. Nucleic Acids Res. 47(D1), D721–D728 (2019).
https://doi.org/10.1093/nar/gky900
32. Qiu, X., et al.: ‘Reversed graph embedding resolves complex single‐cell
trajectories. Nat. Methods 14(10), 979–982 (2017). https://doi.org/10.
1038/nmeth.4402
33. Walter, W., Sánchez‐Cabo, F., Ricote, M.: ‘GOplot: an R package for
visually combining expression data with functional analysis. Bioinfor-
matics 31(17), 2912–2914 (2015). https://doi.org/10.1093/bioinform
atics/btv300
34. Kanehisa, M., et al.: KEGG: new perspectives on genomes, pathways,
diseases and drugs. Nucleic Acids Res. 45(D1), D353–D361 (2017).
https://doi.org/10.1093/nar/gkw1092
35. Kuleshov, M.V., et al.: Enrichr: a comprehensive gene set enrichment
analysis web server 2016 update. Nucleic Acids Res. 44(W1), W90–97
(2016). https://doi.org/10.1093/nar/gkw377
36. Bindea, G., et al.: ClueGO: a Cytoscape plug‐in to decipher functionally
grouped gene ontology and pathway annotation networks. Bioinfor-
matics 25(8), 1091–1093 (2009). https://doi.org/10.1093/bioinfor
matics/btp101
37. Wilkerson, M.D., Hayes, D.N.: ‘ConsensusClusterPlus: a class discovery
tool with condence assessments and item tracking. Bioinformatics
26(12), 1572–1573 (2010). https://doi.org/10.1093/bioinformatics/
btq170
38. Blaser, N., et al.: Gems: an R package for simulating from disease pro-
gression models. J. Stat. Software 64(10), 1–22 (2015). https://doi.org/
10.18637/jss.v064.i10
39. Ritchie, M.E., et al.: Limma powers differential expression analyses for
RNA‐sequencing and microarray studies. Nucleic Acids Res. 43(7), e47
(2015). https://doi.org/10.1093/nar/gkv007
40. Yoshihara, K., et al.: Inferring tumour purity and stromal and immune
cell admixture from expression data. Nat. Commun. 4(1), 2612 (2013).
https://doi.org/10.1038/ncomms3612
41. Langfelder, P., Horvath, S.: ‘WGCNA: an R package for weighted cor-
relation network analysis. BMC Bioinf. 9(1), 559 (2008). https://doi.org/
10.1186/1471‐2105‐9‐559
42. Blanche, P., Dartigues, J.‐F., Jacqmin‐Gadda, H.: Estimating and
comparing time‐dependent areas under receiver operating characteristic
curves for censored event times with competing risks. Stat. Med. 32(30),
5381–5397 (2013). https://doi.org/10.1002/sim.5958
43. Subramanian, A., et al.: GSEA‐P: a desktop application for gene set
enrichment analysis. Bioinformatics, 23, (23), pp, 3251–3253, (2007).
https://doi.org/10.1093/bioinformatics/btm369
44. Kiselev, V.Y., Andrews, T.S., Hemberg, M.: Challenges in unsupervised
clustering of single‐cell RNA‐seq data. Nat. Rev. Genet. 20(5), 273–282
(2019). https://doi.org/10.1038/s41576‐018‐0088‐9
45. Zhang, S., et al.: Single‐cell RNA‐seq analysis reveals microenvironmental
inltration of plasma cells and hepatocytic prognostic markers in HCC
with cirrhosis. Front. Oncol. 10, 596318 (2020). https://doi.org/10.
3389/fonc.2020.596318
46. Yang, Y., et al.: Analysis of single‐cell RNAseq identies transitional
states of T cells associated with hepatocellular carcinoma. Clin. Transl.
Med. 10(3), e133 (2020). https://doi.org/10.1002/ctm2.133
47. Zheng, B., et al.: Trajectory and functional analysis of PD‐1high CD4
+CD8+T cells in hepatocellular carcinoma by single‐cell cytometry and
transcriptome sequencing. Adv. Sci. 7(13), 2000224 (2020). https://doi.
org/10.1002/advs.202000224
48. Banh, C., Fugère, C., Brossay, L.: ‘Immunoregulatory functions of
KLRG1 cadherin interactions are dependent on forward and reverse
signaling. Blood 114(26), 5299–5306 (2009). https://doi.org/10.1182/
blood‐2009‐06‐228353
49. Widjaja, C.E., et al.: Proteasome activity regulates CD8+T lymphocyte
metabolism and fate specication. J. Clin. Invest. 127(10), 3609–3623
(2017). https://doi.org/10.1172/jci90895
50. He, T.‐S., et al.: ALG‐2 couples T cell activation and apoptosis by
regulating proteasome activity and inuencing MCL1 stability. Cell Death
Dis. 11(1), 5 (2020). https://doi.org/10.1038/s41419‐019‐2199‐4
TANG ET AL.
-
243

51. Zeng, D., et al.: Tumor microenvironment characterization in gastric
cancer identies prognostic and immunotherapeutically relevant gene
signatures. Cancer Immunol Res 7(5), 737–750 (2019). https://doi.org/
10.1158/2326‐6066.cir‐18‐0436
52. Li, Y., et al.: ‘Identication of new therapeutic targets for gastric cancer
with bioinformatics. Front. Genet. 11, 865 (2020). https://doi.org/10.
3389/fgene.2020.00865
53. Ling, B., et al.: Microenvironment analysis of prognosis and molecular
signature of immune‐related genes in lung adenocarcinoma. Oncol. Res.
28(6), 561–578 (2021). https://doi.org/10.3727/096504020x159074282
81601
54. Xu, D., et al.: Identication of immune subtypes and prognosis of he-
patocellular carcinoma based on immune checkpoint gene expression
prole. Biomed. Pharmacother. 126, 109903 (2020). https://doi.org/10.
1016/j.biopha.2020.109903
55. Wagner, J., et al.: ‘A single‐cell atlas of the tumor and immune ecosystem
of human breast cancer. Cell 177(5), 1330–1345.e18 (2019). https://doi.
org/10.1016/j.cell.2019.03.005
56. Liu, F., et al.: Heterogeneity of exhausted T cells in the tumor micro-
environment is linked to patient survival following resection in hepato-
cellular carcinoma. OncoImmunology 9(1), 1746573 (2020). https://doi.
org/10.1080/2162402x.2020.1746573
57. Nakano, M., et al.: PD‐1+TIM‐3+T cells in malignant ascites predict
prognosis of gastrointestinal cancer. Cancer Sci. 109(9), 2986–2992
(2018). https://doi.org/10.1111/cas.13723
58. Barsch, M., et al.: ‘T‐cell exhaustion and residency dynamics inform
clinical outcomes in hepatocellular carcinoma. J. Hepatol. 77(2), 397–409
(2022). https://doi.org/10.1016/j.jhep.2022.02.032
59. Hung, M.H., et al.: Tumor methionine metabolism drives T‐cell
exhaustion in hepatocellular carcinoma. Nat. Commun. 12(1), 1455
(2021). https://doi.org/10.1038/s41467‐021‐21804‐1
60. Zhang, F.‐P., et al.: Construction of a risk score prognosis model based
on hepatocellular carcinoma microenvironment. World J. Gastroenterol.
26(2), 134–153 (2020). https://doi.org/10.3748/wjg.v26.i2.134
61. Qin, G., et al.: NPM1 upregulates the transcription of PD‐L1 and sup-
presses T cell activity in triple‐negative breast cancer. Nat. Commun.
11(1), 1669 (2020). https://doi.org/10.1038/s41467‐020‐15364‐z
62. Ternette, N., et al.: Immunopeptidomic proling of HLA‐A2‐positive
triple negative breast cancer identies potential immunotherapy target
antigens. Proteomics 18(12), e1700465 (2018). https://doi.org/10.1002/
pmic.201700465
63. Song, K.‐H., et al.: HSP90A inhibition promotes anti‐tumor immunity by
reversing multi‐modal resistance and stem‐like property of immune‐
refractory tumors. Nat. Commun. 11(1), 562 (2020). https://doi.org/
10.1038/s41467‐019‐14259‐y
64. Zhao, E., et al.: Cancer mediates effector T cell dysfunction by targeting
microRNAs and EZH2 via glycolysis restriction. Nat. Immunol. 17(1),
95–103 (2016). https://doi.org/10.1038/ni.3313
65. Zhou, L., et al.: Targeting EZH2 enhances antigen presentation, anti-
tumor immunity, and circumvents anti‐PD‐1 resistance in head and neck
cancer. Clin. Cancer Res. 26(1), 290–300 (2020). https://doi.org/10.
1158/1078‐0432.ccr‐19‐1351
66. Ma, J., et al.: PD1Hi CD8+T cells correlate with exhausted signature and
poor clinical outcome in hepatocellular carcinoma. J Immunother Cancer
7(1), 331 (2019). https://doi.org/10.1186/s40425‐019‐0814‐7
67. Nishino, M., et al.: Monitoring immune‐checkpoint blockade: response
evaluation and biomarker development. Nat. Rev. Clin. Oncol. 14(11),
655–668 (2017). https://doi.org/10.1038/nrclinonc.2017.88
68. Chan, T.A., et al.: Development of tumor mutation burden as an
immunotherapy biomarker: utility for the oncology clinic. Ann. Oncol.
30(1), 44–56 (2019). https://doi.org/10.1093/annonc/mdy495
69. El‐Khoueiry, A.B., et al.: Nivolumab in patients with advanced
hepatocellular carcinoma (CheckMate 040): an open‐label, non‐
comparative, phase 1/2 dose escalation and expansion trial. Lancet
389(10088), 2492–2502 (2017). https://doi.org/10.1016/s0140‐6736(17)
31046‐2
70. Blackburn, S.D., et al.: Selective expansion of a subset of exhausted CD8
T cells by alphaPD‐L1 blockade. Proc. Natl. Acad. Sci. U. S. A. 105(39),
15016–15021 (2008). https://doi.org/10.1073/pnas.0801497105
71. Solimando, A.G., et al.: ‘Second‐line Treatments for Advanced Hepato-
cellular Carcinoma: A Systematic Review and Bayesian Network Meta‐
analysis’Clin Exp Med (2021)
72. Leone, P., et al.: The evolving role of immune checkpoint inhibitors in
hepatocellular carcinoma treatment. Vaccines 9(5), 532 (2021). https://
doi.org/10.3390/vaccines9050532
73. Yost, K.E., et al.: Clonal replacement of tumor‐specic T cells following
PD‐1 blockade. Nat. Med. 25(8), 1251–1259 (2019). https://doi.org/10.
1038/s41591‐019‐0522‐3
SUPPORTING INFORMATION
Additional supporting information can be found online in the
Supporting Information section at the end of this article.
How to cite this article: Tang, X., et al.: Single‐cell
RNA‐seq and bulk RNA‐seq explore the prognostic
value of exhausted T cells in hepatocellular carcinoma.
IET Syst. Biol. 17(4), 228–244 (2023). https://doi.org/
10.1049/syb2.12072
244
-
TANG ET AL.