Access to this full-text is provided by Springer Nature.
Content available from Molecular Cancer
This content is subject to copyright. Terms and conditions apply.
Dengetal. Molecular Cancer (2025) 24:28
https://doi.org/10.1186/s12943-025-02231-y
RESEARCH Open Access
© The Author(s) 2025. Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0
International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long
as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if
you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or
parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated
otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not
permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To
view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/.
Molecular Cancer
Comprehensive single-cell atlas ofcolorectal
neuroendocrine tumors withliver metastases:
unraveling tumor microenvironment
heterogeneity betweenprimary lesions
andmetastases
Yiqiao Deng1†, Qichen Chen2†, Chengyao Guo1†, Jinghua Chen1, Xin Li1, Zhiyu Li1, Yefan Zhang1, Jianjun Zhao1,
Jianguo Zhou1, Jianqiang Cai1, Tao Yan3*, Xiaobing Wang4*, Xinyu Bi1*, Zhen Huang1* and Hong Zhao1*
Abstract
Background Colorectal neuroendocrine tumors with liver metastases (CRNELM) are associated with a poorer
prognosis compared to their nonmetastatic counterparts. A comprehensive understanding of the tumor microenvi-
ronment (TME) heterogeneity between primary lesions (PL) and liver metastases (LM) could provide crucial insights
for enhancing clinical management strategies for these patients.
Methods We utilized single-cell RNA sequencing to analyze fresh tissue samples from CRNELM patients, aiming
to elucidate the variations in TME between PL and LM. Complementary multidimensional validation was achieved
through spatial transcriptomics, bulk RNA sequencing, and multiplex immunohistochemistry/immunofluorescence.
Results Our single-cell RNA sequencing analysis revealed that LM harboured a higher proportion of CD8 + T cells,
CD4 + T cells, NK cells, NKT cells, and B cells exhibiting a stress-like phenotype compared to PL. RGS5 + pericytes may
play a role in the stress-like phenotype observed in immune cells within LM. MCs in PL (PL_MCs) and LM (LM_MCs)
exhibit distinct activation of tumor-associated signaling pathways. Notably, COLEC11 + matrix cancer-associated
fibroblasts (COLEC11_mCAFs) were found to be significantly associated with LM_MCs. Cell communication analysis
unveiled potential targetable receptor-ligand interactions between COLEC11_mCAFs and LM_MCs. Multidimen-
sional validation confirmed the prominence of the characteristic stress-like phenotypes, including HSPA6_CD8_Tstr,
†Yiqiao Deng, Qichen Chen and Chengyao Guo contributed equally to this
work.
*Correspondence:
Tao Yan
blizzardyt@163.com
Xiaobing Wang
wangxb@cicams.ac.cn
Xinyu Bi
beexy1971@163.com
Zhen Huang
pumchuangzhen@126.com
Hong Zhao
zhaohong@cicams.ac.cn
Full list of author information is available at the end of the article
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
Page 2 of 25
Dengetal. Molecular Cancer (2025) 24:28
HSPA6_NK, and COLEC11_mCAFs in LM. Moreover, a higher abundance of COLEC11_mCAFs correlated with poorer
survival rates in the neuroendocrine tumor patient cohort.
Conclusion Overall, our study provides the first single-cell analysis of the cellular and molecular differences
between PL and LM in CRNELM patients. We identified distinct cell subsets and receptor-ligand interactions that may
drive TME discrepancies and support metastatic tumor growth. These insights highlight potential therapeutic targets
and inform strategies for better managing CRNELM patients.
Keywords Colorectal neuroendocrine tumour, Liver metastases, Tumour microenvironment, Heterogeneity, Single-
cell RNA sequence, Spatial transcriptomics
Introduction
Neuroendocrine tumours (NETs), which arise from spe-
cialized secretory cells scattered throughout the neu-
roendocrine system, present with a wide array of clinical
symptoms and manifestations. e rising number of NET
diagnoses is not solely attributable to heightened aware-
ness or advancements in diagnostic techniques, thus
warranting further exploration [1]. NETs predominantly
arise in the gastrointestinal tract, especially in the colo-
rectum [2], and a considerable proportion of colorectal
neuroendocrine tumour (CRNET) patients show signs
of distant metastasis [3]. In the context of CRNETs, the
liver is the most common site for metastasis. Over half
of CRNET patients present with liver metastases (LM)
at initial diagnosis, and the incidence of LM tends to
increase with time [4, 5]. e prognosis for CRNETs
with LM (CRNELM) is significantly worse than that for
localized CRNETs. CRNELM has relatively high morbid-
ity and mortality rates, with the 5-year survival rate for
untreated patients ranging from approximately 13% to
54%, whereas for CRNETs, it ranges from 75 to 80% [6,
7]. is trend underscores the need for greater attention
and research to improve the prognosis of CRNELM.
e tumour microenvironment (TME) displays nota-
ble heterogeneity, with substantial variability within and
between tumours, thereby influencing tumour biology,
treatment response, and prognosis [8, 9]. e inherent
limitations of conventional therapies, which exclusively
target tumour cells while neglecting the TME, further
highlight the necessity of conducting research to eluci-
date the TME [10]. Breakthroughs in single-cell RNA
sequencing have revolutionized our comprehension
of the TME at the single-cell level, paving the way for
advancements in precision medicine. Previous single-cell
studies have focused primarily on the TME of pancreatic
NETs (pNETs) and lung carcinoid tumours, often with
small sample sizes. Yu etal. [11] identified key molecular
events in the progression from primary tumours to meta-
static lesions in a patient with sporadic pNET. Similarly,
Philip etal. [12] performed single-cell RNA sequencing
on three lung carcinoid tumour samples and matched
normal tissues, revealing a noninflammatory and
vascular-rich TME. However, there is still a lack of rel-
evant research on CRNELM patients. e limited under-
standing of the distinct TME and cellular components of
CRNELM poses a significant barrier to better managing
CRNELM.
is study first explored TME heterogeneity between
primary lesions (PL) and LM to construct a comprehen-
sive single-cell atlas of CRNELM patients, which was
validated through multidimensional analyses, including
spatial transcriptomics (ST), bulk RNA sequencing, and
multiplex immunohistochemistry/immunofluorescence
(mIHC/IF) analysis. is study aims to provide a novel
perspective on the TME of CRNELM, offering promising
insights for future clinical and basic research to improve
CRNELM prognosis.
Methods
Sample collection andprocessing forsingle‑cell andspatial
transcriptomics
e study protocol received approval from the insti-
tutional review board of the Cancer Hospital, Chinese
Academy of Medical Sciences (ID: NCC2021C-515).
For the single-cell RNA sequencing, we obtained 9 fresh
tumor samples from 8 CRNELM patients, including one
paired sample of primary lesion (PL) and liver metastasis
(LM), along with 4 unpaired PL samples and 3 unpaired
LM samples. It’s worth noting that the entire cell popu-
lation was further analysed without any prior sorting
or enrichment of particular cell subtypes. To ensure
the high quality of the cells, stringent filtering crite-
ria were established: 1) cells with fewer than 3 detected
genes were excluded; 2) cells with a total of less than 50
detected genes were omitted; 3) cells exhibiting ≥ 20%
mitochondrial gene content or ≥ 5% haemoglobin gene
content in their transcriptome were removed. Follow-
ing quality control, the dataset was processed using the
Seurat R package (version 4.0.2). Batch effects were miti-
gated across all samples using the harmony [12] function,
and counts were normalized and scaled with SCTrans-
form [13] function, which incorporated regression for
mitochondrial gene content. In this study, integrated
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
Page 3 of 25
Dengetal. Molecular Cancer (2025) 24:28
data of all samples were subjected to subsequent analy-
ses for clustering analysis and subgroup re-clustering
analysis. e optimal number of principal components
(PCs) was determined using the ElbowPlot function in
Seurat. Dimensionality reduction was carried out with
the RunUMAP function, and visualization was achieved
through Uniform Manifold Approximation and Projec-
tion (UMAP). For subgroup cell clustering, cells of differ-
ent types were extracted separately and clustered by their
respective optimal PCs.
To establish a spatial transcriptomic map of CRNELM
tumors, we gathered fresh, paired PL (n = 1) and LM
(n = 1) sample from one extra CRNELM patient utiliz-
ing the Stereo-seq method, adhering to the procedures
detailed by Chen et al. [14]. e raw FASTQ data was
mapped onto the human genome (hg38), and specific
regions were delineated using single-stranded DNA
(ssDNA) staining to minimize background noise for fur-
ther analysis. We designated a 1 × 1 DNA nanoball (DNB)
area as bin 1 and adopted a 100 × 100 DNB grid (bin 100)
as the basic analytical unit, similar to previous study [14].
To obtain better quality for the subsequent analysis, bins
in any section in which genes captured less than 200 were
filtered. Batch effects were mitigated across two sections
using the harmony [12] function, then 2 spatial sections
were normalized and scaled by the function SCTrans-
form [13]. And the integrated data of two sections were
subjected to subsequent analyses. A comprehensive
description of the methodologies employed can be found
in the supplementary materials, while the clinical details
of the patients are summarized in TableS1.
Comparison ofcell type proportions
To explore differences in the cellular compositions of PL
and LM, we initially compared the proportions of distinct
cell subsets across the two sites. e assessment of cell
cluster enrichment or depletion was conducted through
the utilization of the Ro/e methodology [15], which com-
prises the execution of a chi-square test within each clus-
ter, comparing PL and LM. Fundamentally, an Ro/e value
exceeding 1 indicated an increased cell count in PL or
LM, whereas a value below 1 suggested a decrease.
Functional enrichment analysis
To evaluate the significance of the gene expression pat-
terns, we employed the "clusterProfiler" package and per-
formed Gene Ontology (GO) and Kyoto Encyclopedia
of Genes and Genomes (KEGG) enrichment analyses to
elucidate the biological functions associated with various
cell clusters [16]. Further analysis covered a wide range of
cellular phenotypes and focused on calculating the gene
signature scores of T cells [15], NK cells [17], B cells [18],
macrophages [19], dendritic cells [19], and fibroblasts
[20]. ese scores were derived via the AddModuleScore
function in the seurat framework, which incorporates
gene sets (TableS2-7) curated from previous studies [15,
17–20]. Pathway-responsive genes for activity inference
(Progeny) [21] analysis were also utilized to compare and
assess the activation levels of specific tumour signaling
pathways in epithelial cells between PL and LM.
Trajectory analysis
e monocle R package [22] was employed to elucidate
cellular fate decisions and map intricate developmental
trajectories. Furthermore, generalized additive models
were applied to explore the variation in different signa-
ture scores along the pseudotime trajectories between
the PL and LM obtained from the monocle algorithm.
Chromosomal copy number variation estimation
Chromosomal copy number variations (CNVs) were
assessed via the inferCNV package [23]. CNV scores
were calculated by aggregating the CNV data across cells
within each subcluster, which allowed us to differentiate
malignant cells from normal epithelial cells.
Responsiveness analysis withAugur
Augur [24] is a specialized machine learning tool
designed to evaluate the prominence of cell types within
single-cell RNA sequencing data. By analysing transcrip-
tional response intensities across different cell types
exposed to biological perturbations, Augur identified the
most responsive cell types under specific conditions.
Pseudospace construction
To reconstruct the spatial organization of distinct cell
subpopulations in single-cell RNA sequencing data, we
employed the CSOmap algorithm [25]. is method inte-
grates thousands of ligand‒receptor pairs to generate
a cell‒cell affinity matrix, which is then embedded into
three-dimensional space.
Cell communication analysis
To elucidate the intercellular interactions within the TME
and identity significant, unique ligand‒receptor pairs that
are exclusively present in either PL or LM, we harnessed
the capabilities of the CommPath package [26].
Validation using spatial transcriptomics data
For deconvolution and cell type annotation, single-cell
RNA sequencing and ST data were integrated via the
’CARD’ (v 1.1) [27] package with default settings. Ini-
tially, a ’CARD’ object was created with the CreateCAR-
DObject function, and the results were computed using
CARD_deconvolution with default parameters. We then
utilized the AddModuleScore function to calculate the
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
Page 4 of 25
Dengetal. Molecular Cancer (2025) 24:28
representative cell cluster gene score and conducted sta-
tistical analyses to compare the differences between PL
and LM.
Bulk RNA sequencing cohorts andmIHC/IF forvalidation
To validate our key observations, we utilized two inde-
pendent datasets, GSE27162 and GSE211485, both
retrieved from the Gene Expression Omnibus (GEO)
repository, accessible at https:// www. ncbi. nlm. nih. gov/.
e first dataset, GSE211485, consists of a pan-NET
cohort, which includes samples from various NETs of the
small intestine, appendix, colon, pancreas, stomach, and
lung. is dataset incorporates survival data, enabling us
to investigate the potential relationship between specific
cellular subpopulations and patient prognosis, as detailed
in TableS13. e second dataset, GSE27162, comprises
a cohort of colon NETs, including both LM and PL
samples. is dataset served as the primary validation
set for assessing the distribution of cellular subpopula-
tions across different tumor locations, as described in
Table S14. e cutoff values for high and low propor-
tions of specific cell types were determined based on the
median values calculated using CibersortX (https:// ciber
sortx. stanf ord. edu/ index. php). e results were then
validated via mIHC/mIF. irteen samples, consisting of
6 from PL and 7 from LM, were included in the mIHC/
IF analysis. e clinical details of these patients are pro-
vided in TableS7. e detailed procedures for mIHC/IF
can be found in the supplementary materials.
Statistical analysis
Statistical analyses were conducted using R software
(version 4.1.2). Comparisons between groups were made
using the Wilcoxon rank-sum test. Correlations were
assessed using Pearson and Spearman methods. Detailed
descriptions of the statistical approaches can be found in
the Results section and figure legends.
Results
Global single‑cell landscape intheCRNELM
e overall workflow of this study was illustrated in
Fig. 1A. Single-cell RNA sequencing was employed to
explore the variation in the TME between PL (n = 5) and
LM (n = 4) within the CRNELM context. rough rig-
orous quality control of the samples (Fig. S1A-B), batch
effect correction (Fig. S1C-D) and dimensionality reduc-
tion followed by clustering, eight major clusters were
identified (Fig. 1B). Feature plots, which are based on
established marker genes, were used to pinpoint spe-
cific cell types (Fig.1C). e proportions of the different
cellular subsets between PL and LM were subsequently
analysed. Epithelial cells, plasma cells, endothelial cells,
and mast cells were more prevalent in PL, whereas T/
NK cells, myeloid cells, fibroblasts, and B cells were more
abundant in LM (Fig. 1D). ese observed differences
in cell distribution suggested that the anatomical con-
text influenced these cellular subpopulations, potentially
leading to distinct functional phenotypes. ese findings
aligned with previous researches on colorectal cancer
with liver metastases [28, 29], underscoring the signifi-
cance of these cellular variations in LM progression.
T, NK, andNKT cell reprogramming towardsastress‑like
phenotype intheTME ofLMs
Understanding the composition and elucidating the role
of T, natural killer (NK) and natural killer T (NKT) cell
infiltration in the TME is pivotal for accurate prognosis
assessment and the formulation of effective treatment
strategies [9]. We extracted and re-clustered all T cells,
NK cells, and NKT cells from all samples into distinct
subsets including FGFBP2_CD8_Teff, GZMK_CD8_Tem,
GPR183_CD8_Tm, GIMAP7_CD8_Tm, SLC4A10_CD8_
MAIT, HSPA6_CD8_Tstr, CTLA4_CD4_Treg, CCR7_
CD4_Tn, GPR183_CD4_Tm, FGFBP2_NK, HSPA6_NK,
KLRC1_NK, KIT_NKT and TRGC2_NKT (Fig. 2A,
S2A-C & Table1, S8-S9). For gene expression analysis,
GPR183_CD8_Tm and GZMK_CD8_Tem cells pre-
sented increased expression of lymphocyte maturation-
related genes such as CD52, as well as genes associated
with increased antitumour immune responses, such as
CRTAM (TableS9) [30]. HSPA6_CD8_Tstr and HSPA6_
NK exhibited significant upregulation of heat shock pro-
teins such as HSPA6, DNAJB4, and HSPA1A (Fig. S2A-B
& TableS9). CTLA4_CD4_Treg notably upregulated key
inhibitory markers, including BATF, TNFRSF18, LAYN,
and TNFRSF9 (Fig. S2C). e Ro/e index and propor-
tion analysis revealed that certain cell types, namely,
CCR7_CD4_Tn, CTLA4_CD4_Treg, GPR183_CD4_Tm,
GPR183_CD8_Tm, GZMK_CD8_Tem, KIT_NKT and
TRGC2_NKT cells, preferentially accumulated in PL.
Fig. 1 The single-cell landscape of PL and LM of CRNELM patients A Workflow depicting the overall experimental design for scRNA-seq
profiling of CRNELM. B UMAP visualization of eight major cell clusters (top) and corresponding cell density (bottom) from all samples showing
the distribution between LM (left) and PL (right). In the cell density plot, a high relative cell density is indicated by brighter regions. C Heatmap
depicting the expression of marker genes across eight defined major cell clusters. The colour intensity reflects average scaled gene expression. The
expression intensity of markers is shown. D The bar plot illustrates relative proportions of 8 major cell types across PL and LM
(See figure on next page.)
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
Page 5 of 25
Dengetal. Molecular Cancer (2025) 24:28
Fig. 1 (See legend on previous page.)
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
Page 6 of 25
Dengetal. Molecular Cancer (2025) 24:28
Other cell types, including FGFBP2_CD8_Teff, HSPA6_
CD8_Tstr, FGFBP2_NK, HSPA6_NK and KLRC1_NK
cells, exhibited a distribution pattern opposite to that of
the aforementioned groups (Fig.2B, C).
We collected and curated four commonly used gene
signatures (activation, cytotoxicity, stress, and inhibi-
tory) [15] for CD8 + T cells (TableS2), CD4 + T cells
(Table S3), and NK cells (Table S4) from all samples
to more precisely investigate the cellular states of dis-
tinct subclusters, respectively. Intriguingly, within the
CD8 + T cell subclusters (Fig. S2D), HSPA6_CD8_Tstr
exhibited the highest scores for activation, stress, and
inhibitory markers among CD8 + T cell subsets, yet
displayed a relatively low cytotoxicity score. FGFBP2_
CD8_Teff demonstrated the highest cytotoxicity score,
accompanied by activation and stress scores that were
second only to HSPA6_CD8_Tstr, as well as a nota-
ble inhibitory score. In the NK/NKT cell subclusters
(Fig. S2E), HSPA6_NK showed the highest activation
and stress scores, an inhibitory score second only to
KLRC1_NK, and a low cytotoxicity score. FGFBP2_NK
displayed the highest cytotoxicity score, coupled with
stress scores second only to HSPA6_NK, along with
elevated activation and inhibitory scores. Among the
CD4 + T cell subclusters (Fig. S2F), CTLA4_CD4_Treg
cells, which exhibited the highest inhibitory score, also
showed the highest stress score but the lowest activa-
tion score. ese findings underscore the complex phe-
notypes and functional heterogeneity of T cells and NK
cells within the TME of CRNELM.
(See figure on next page.)
Fig. 2 Characterization of T, NK, and NKT cells between PL and LM of CRNELM patients. A UMAP plot illustrating distinct T, NK and NKT cell
subclusters across all samples, color-coded by cell type. At the center are the primary cell types identified: CD4 + T cells, CD8 + T cells, and NK/
NKT cells. The top-left quadrant shows six subclusters of CD8 + T cells. The bottom-left quadrant displays three subclusters of CD4 + T cells. The
top-right quadrant presents five subclusters of NK/NKT cells. B The bar plot illustrates the relative proportions of all T, NK and NKT cell subclusters
between LM and PL. C The lollipop chart depicts the prevalence of all T, NK, and NKT cell subclusters between LM and PL, as quantified
by the Ro/e ratio. Subclusters that favor LM are situated above the baseline, while those that favor PL are situated below. A higher numerical
value indicates a stronger preference, with an Ro/e ratio exceeding 1 (marked by dashed lines above and below) considered indicative
of a significant bias. D Boxplot comparing the activation (top left), cytotoxicity (top right), stress (bottom left), and inhibitory (bottom right)
scores of CD8 + T cells between LM and PL. The position of the CD8 + T cells is annotated below the boxplot. The boxes represent the median
(indicated by the horizontal line with a numerical value), encompass the second to third quartiles (forming the body of the box), and extend
with Tukey-style whiskers beyond the box to depict the data range. Wilcoxon rank-sum test. E Monocle trajectory analysis of CD8 + T cell
differentiation unveiled two primary divergent pathways. The top panel displays the pseudotime map composed of various CD8 + T cell subsets,
while the bottom panel illustrates the directional progression of pseudotime in a smaller, more focused pseudotime trajectory plot. The cells are
color-coded based on their pseudotime. F Correlation analysis between the pseudotime of LM (depicted in dark green) and PL (depicted in light
green), and the activation (top left), cytotoxicity (top right), stress (bottom left), and inhibitory (bottom right) scores of CD8 + T cells. Each plot
highlights the correlation coefficients and P-values for LM (above) and PL (below), utilizing Pearson correlation and a generalized additive model.
G Correlations between the stress score (shown below) and the activation score (top left plot), cytotoxicity score (top right plot), and inhibitory
score (bottom plot) of all CD8 + T cells. Each plot is annotated with the correlation coefficient and P-value at the top, utilizing Pearson correlation
with a linear model. H Boxplot comparing the activation (top left), cytotoxicity (top right), stress (bottom left), and inhibitory (bottom right) scores
of NK and NKT cells between LM and PL. The position of the NK and NKT cells is annotated below the boxplot. The boxes represent the median
(indicated by the horizontal line with a numerical value), encompass the second to third quartiles (forming the body of the box), and extend
with Tukey-style whiskers beyond the box to depict the data range. Wilcoxon rank-sum test. I Monocle trajectory analysis of all NK and NKT cells
differentiation unveiled two primary divergent pathways. The top panel displays the pseudotime map composed of various NK and NKT cells
subsets, while the bottom panel illustrates the directional progression of pseudotime in a smaller, more focused pseudotime trajectory plot. The
cells are color-coded based on their pseudotime. J Correlation analysis between the pseudotime of LM (depicted in dark blue) and PL (depicted
in light blue), and the activation (top left), cytotoxicity (top right), stress (bottom left), and inhibitory (bottom right) scores of NK and NKT cells.
Each plot highlights the correlation coefficients and P-value for LM (above) and PL (below), utilizing Pearson correlation and a generalized
additive model. K Correlations between the stress score (shown below) and the activation score (top left plot), cytotoxicity score (top right plot),
and inhibitory score (bottom plot) of all NK and NKT cells. Each plot is annotated with the correlation coefficient and P-value at the top, utilizing
Pearson correlation with a linear model. L Boxplot comparing the activation (top left), cytotoxicity (top right), stress (bottom left), and inhibitory
(bottom right) scores of CD4 + T cells between LM and PL. The position of the CD4 + T cells is annotated below the boxplot. The boxes represent
the median (indicated by the horizontal line with a numerical value), encompass the second to third quartiles (forming the body of the box),
and extend with Tukey-style whiskers beyond the box to depict the data range. Wilcoxon rank-sum test. M Monocle trajectory analysis of CD4 + T
cell differentiation unveiled one primary divergent pathway. The top panel displays the pseudotime map composed of various CD4 + T cell subsets,
while the bottom panel illustrates the directional progression of pseudotime in a smaller, more focused pseudotime trajectory plot. The cells are
color-coded based on their pseudotime. N Correlation analysis between the pseudotime of LM (depicted in dark purple) and PL (depicted in light
purple), and the activation (top left), cytotoxicity (top right), stress (bottom left), and inhibitory (bottom right) scores of CD4 + T cells. Each plot
highlights the correlation coefficients and P-values for LM (above) and PL (below), utilizing Pearson correlation and a generalized additive model.
O Correlations between the stress score (shown below) and the activation score (top left plot), cytotoxicity score (top right plot), and inhibitory
score (bottom plot) of all CD4 + T cells. Each plot is annotated with the correlation coefficient and P-value at the top, utilizing Pearson correlation
with a linear model
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
Page 7 of 25
Dengetal. Molecular Cancer (2025) 24:28
Fig. 2 (See legend on previous page.)
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
Page 8 of 25
Dengetal. Molecular Cancer (2025) 24:28
Subsequently, we conducted a comparative analy-
sis of the four distinct cellular states of CD8 + T cells
derived from LM and PL (Fig.2D). e results demon-
strated that, in comparison to CD8 + T cells from PL,
those from LM exhibited significantly elevated scores
for activation, stress, and inhibition, while their cyto-
toxicity scores were notably reduced (Wilcox rank-sum
test, P < 0.01), with the stress score being the most sig-
nificantly different between the two groups. To further
investigate the transitions between these cellular states,
we performed pseudotime analysis of all CD8 + T cells
using Monocle, which identified differentiation trajec-
tories initiating from GZMK_CD8_Tem cells, progress-
ing through GIMAP7_CD8_Tm and GPR183_CD8_Tm,
and ultimately converging on either FGFBP2_CD8_Teff
or HSPA6_CD8_Tstr (Fig.2E). e dynamic analysis of
signature score changes along pseudotime trajectory
between PL and LM revealed intriguing differences.
Along the pseudotime trajectory, CD8 + T cells from
both PL and LM showed an increase in activation,
stress, and inhibitory scores. However, based on the
correlation coefficients, the increase in these scores
occurred more rapidly in CD8 + T cells from LM. In
contrast, while CD8 + T cells from PL displayed an
increase in cytotoxicity scores, those from LM showed
a decrease (Pearson correlation, P < 0.01; Fig. 2F). e
subsequent correlation analysis revealed that, among
all CD8 + T cells, the stress score was significantly
positively correlated with the activation and inhibi-
tory scores, while it was significantly negatively corre-
lated with the cytotoxicity score (Pearson correlation,
P < 0.01; Fig.2G).
Likewise, we conducted a comparative assessment of
four distinct cellular states in NK and NKT cells isolated
from LM and PL (Fig.2H). Our findings indicated that
LM-derived NK and NKT cells exhibited notably higher
scores for activation, stress, and inhibition compared to
those from PL, while their cytotoxicity scores were signif-
icantly reduced (Wilcoxon rank-sum test, P < 0.01), with
the stress score showing the most significant difference
between the two groups. Additionally, pseudotime analy-
sis of all NK and NKT cells unveiled differentiation path-
ways originating from FGFBP2_NK cells. ese pathways
either conclude at KLRC1_NK cells or progress through
KIT_NKT cells before converging at HSPA6_NK cells
(Fig. 2I). roughout this differentiation process, both
PL and LM-derived cells demonstrated increased stress
and inhibitory scores, accompanied by decreased cyto-
toxicity scores. However, based on the correlation coef-
ficients, the rise in these scores was more pronounced
in LM-derived cells, and the reduction in cytotoxicity
was more marked (Pearson correlation, P < 0.01). In con-
trast, PL-derived NK and NKT cells showed a decrease
in activation scores, whereas those from LM exhibited
an increase (Pearson correlation, P < 0.01; Fig.2J). Subse-
quent correlation analysis further revealed that, across all
NK and NKT cells, the stress score was significantly posi-
tively associated with activation and inhibitory scores,
while it was significantly negatively correlated with the
cytotoxicity score (Pearson correlation, P < 0.01; Fig.2K).
Likewise, we conducted a comparative assessment
of four distinct cellular states in CD4 + T cells isolated
from LM and PL (Fig. 2L). e results demonstrated
that CD4 + T cells isolated from LM had markedly
higher activation, cytotoxicity, and stress scores, along
with a significantly lower inhibitory score, in com-
parison to those from PL (Wilcoxon rank-sum test,
Table 1 Annotation of All Cell Subclusters
Cell type Representative
genes Respective name
FGFBP2 + CD8 + T effector cell FGFBP2 FGFBP2_CD8_Teff
GZMK + CD8 + T effective-
memory cell GZMK GZMK_CD8_Tem
GPR183 + CD8 + T memory cell GPR183 GPR183_CD8_Tm
GIMAP7 + CD8 + T memory cell GIMAP GIMAP7_CD8_Tm
SLC4A10 + CD8 + T mucosal-
associated invariant cell SLC4A10 SLC4A10_CD8_MAIT
HSPA6 + CD8 + T stress-like cell HSPA6 HSPA6_CD8_Tstr
CTLA4 + CD4 + T regulator y cell CTLA4 CTLA4_CD4_Treg
CCR7 + CD4 + T naive cell CCR7 CCR7_CD4_Tn
GPR183 + CD4 + T memory cell GPR183 GPR183_CD4_Tm
FGFBP2 + NK cell FGFBP2 FGFBP2_NK
HSPA6 + NK cell HSPA6 HSPA6_NK
KLRC1 + NK cell KLRC1 KLRC1_NK
KIT + NKT cell KIT KIT_NKT
TRGC2 + NKT cell TRGC2 TRGC2_NKT
NR4A2 + naive B cell NR4A2 NR4A2_Bn
IGHA1 + plasma cell IGHA1 IgA_PCs
IGHG1 + plasma cell IGHG1 IgG_PCs
STMN1 + macrophage STMN1 STMN1_Macro
APOE + macrophage APOE APOE_Macro
MARCO + macrophage MARCO MARCO_Macro
JUN + macrophage JUN JUN_Macro
FCN1 + monocyte FCN1 FCN1_Mono
CD1C + dendritic cell CD1C CD1C_DC
CLEC9A + dendritic cell CLEC9A CLEC9A_DC
FCGR3B + neutrophil FCGR3B FCGR3B_Neu
CCL5 + neutrophil CCL5 CCL5_Neu
COLEC11 + matrix fibroblast COLEC11 COLEC11_mCAF
CFD + inflammatory fibroblast CFD CFD_iCAF
MYH11 + smooth muscle cell MYH11 MYH11_SMC
RGS5 + pericyte RGS5 RGS5_pericyte
F3 + telocyte F3 F3_telocyte
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
Page 9 of 25
Dengetal. Molecular Cancer (2025) 24:28
P < 0.01), with the stress score showing the most sig-
nificant difference between the two groups. en, pseu-
dotime analysis of all CD4 + T cells revealed a single
major differentiation path starting from CRP183_CD4_
Tm, progressing through CCR7_CD4_Tn, and termi-
nating at CTLA4_CD4_Tregs (Fig. 2M). roughout
this pseudotime progression, CD4 + T cells from both
PL and LM demonstrated an increase in both cyto-
toxicity and inhibitory scores. Notably, as evidenced
by the correlation coefficients, CD4 + T cells from LM
exhibited a more rapid increase in inhibitory scores and
a slower rise in cytotoxicity scores compared to those
from PL (Pearson correlation, P < 0.01). Furthermore,
CD4 + T cells from LM also showed a decline in acti-
vation scores accompanied by an elevation in stress
scores. In contrast, CD4 + T cells from PL displayed
no significant alteration in activation scores (Pearson
correlation, P = 0.42) and a reduction in stress scores
along the pseudotime trajectory (Pearson correlation,
P < 0.01; Fig. 2O). Correlation analysis indicated that
stress scores were positively associated with increased
activation, cytotoxicity and inhibitory scores (Pearson
correlation, P < 0.01; Fig.2O).
Despite certain disparities, a notable consensus
emerged: CD8 + T cells, NK cells, NKT cells, and
CD4 + T cells in LM all demonstrated a more marked
stress phenotype. As stress scores increased, all these
immune cells consistently displayed a notable elevation
in inhibitory scores. Interestingly, with the rise in stress
scores, all CD8 + T cells, NK cells, and NKT cells also
demonstrated a reduction in cytotoxicity scores. Ulti-
mately, the shift towards a stress-like phenotype among
T, NK, and NKT cells within the TME of LM seems to be
a key determinant in the enhancement of immune sup-
pression and exhaustion.
B‑cell reprogramming towardsastress‑like phenotype
intheTME ofLM
We further extracted and re-clustered all B and plasma
cells from all samples into distinct subclusters, includ-
ing NR4A2_Bn, IgA_PCs, and IgG_PCs (Fig.3A, Table1,
and Table S10). NR4A2_Bn exhibited an upregulation
of genes associated with the human leukocyte anti-
gen (HLA) complex, namely, HLA-DRA, HLA-DRB1,
and HLA-DPB1, alongside CD74 (Fig. 3B), indicating
its capacity to mount immune responses against a wide
array of antigens. Furthermore, these cells presented high
expression levels of CXCR4, a pivotal chemokine receptor
involved in cell adhesion that was previously recognized
as a marker for B cells with profibrotic and proinflamma-
tory functions [31]. Additionally, IgA_PCs and IgG_PCs
displayed elevated expression of immunoglobulin-encod-
ing genes (Fig.3B). e presence of IgG_PCs, known for
their antitumour properties [32–34], suggests a height-
ened humoural immune response in PL. Ro/e and pro-
portion analysis indicated a preferential distribution of
NR4A2_Bn cells in LM, whereas IgA_PCs and IgG_PCs
tended to localize predominantly in PL (Fig.3C, D).
Further functional enrichment analysis of GO
revealed that NR4A2_Bn was implicated in the activa-
tion of T-cell receptor-associated pathways (Fig. 3E),
whereas IgA_PCs and IgG_PCs were connected to
pathways involved in the negative regulation of acute
inflammatory responses and the modulation of col-
lagen fibril biogenesis (Fig. 3E). Additionally, results
from KEGG analysis indicated that NR4A2_Bn was
Fig. 3 Characterization of B and plasma cells between PL and LM of CRNELM patients. A UMAP plot illustrating distinct B and plasma cell
subclusters across all samples, color-coded by cell type. B Heatmap depicting the expression of marker genes across three defined B and plasma
cell clusters. The colour intensity reflects average scaled gene expression. The expression intensity of markers is shown. C The bar plot illustrates
the relative proportions of all B and plasma cell subclusters between LM and PL. D The lollipop chart depicts the prevalence of all B and plasma
cell subclusters between LM and PL, as quantified by the Ro/e ratio. Subclusters that favor LM are situated above the baseline, while those
that favor PL are situated below. A higher numerical value indicates a stronger preference, with an Ro/e ratio exceeding 1 (marked by dashed lines
above and below) considered indicative of a significant bias. E Heatmap displaying the expression of functional pathways activated in different
B and plasma cell subclusters via GO analysis. The heatmap is based on scaled gene signature scores. F Heatmap displaying the expression
of functional pathways activated in different B and plasma cell subclusters via KEGG analysis. The heatmap is based on scaled gene signature
scores. G Boxplot comparing the stress scores of three B and plasma cell subclusters. The subcluster is annotated below the boxplot. The boxes
represent the median (indicated by the horizontal line with a numerical value), encompass the second to third quartiles (forming the body
of the box), and extend with Tukey-style whiskers beyond the box to depict the data range. H Boxplot comparing the stress scores of B and plasma
cells between LM and PL. The position of the B and plasma cells is annotated below the boxplot. The boxes represent the median (indicated
by the horizontal line with a numerical value), encompass the second to third quartiles (forming the body of the box), and extend with Tukey-style
whiskers beyond the box to depict the data range. Wilcoxon rank-sum test. I Monocle trajectory analysis of all B and plasma cells differentiation
unveiled one primary divergent pathway. The top panel displays the pseudotime map composed of various subclusters, while the bottom panel
illustrates the directional progression of pseudotime in a smaller, more focused pseudotime trajectory plot. The cells are color-coded based
on their pseudotime. J Correlation analysis between the pseudotime of LM (depicted in dark orange) and PL (depicted in light orange), and stress
scores of B and plasma cells. The highlights the correlation coefficients and P-values for LM (above) and PL (below), utilizing Pearson correlation
and a generalized additive model
(See figure on next page.)
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
Page 10 of 25
Dengetal. Molecular Cancer (2025) 24:28
associated with infectious diseases such as COVID-
19 and tuberculosis, suggesting potential activation of
immune functions (Fig.3F), whereas IgA_PCs and IgG_
PCs were linked to pathways involving cytochrome
P450s, possibly participating in oxidative reactions
(Fig. 3F). Moreover, we compared the stress scores
[18] specifically within all B cells (TableS5) and found
that NR4A2_Bn exhibited the highest score, followed
by IgG_PCs, and IgA_PCs had the lowest (Fig. 3G).
In the overall comparison, B cells derived from LM
demonstrated significantly higher stress scores com-
pared to those from PL (Wilcoxon rank-sum test,
P < 0.01; Fig. 3H). Pseudotime analysis of all B and
plasma cells unveiled a primary differentiation trajec-
tory initiating from NR4A2_Bn cells, transitioning
through IgA_PCs, and ultimately reaching IgG_PCs
(Fig. 3I). roughout this pseudotime progression, B
and plasma cells from LM showed an increase in stress
scores (Pearson correlation, P < 0.01), whereas B and
plasma cells from PL displayed no notable change in
stress scores (Pearson correlation, P = 0.63) (Fig.3J).
Fig. 3 (See legend on previous page.)
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
Page 11 of 25
Dengetal. Molecular Cancer (2025) 24:28
Dissection andclustering ofmyeloid cells inthePL andLM
oftheCRNELM
To gain deeper insights into the myeloid cells present
in the PL and LM of CRNELM patients, we extracted
and re-clustered all myeloid cells within our dataset.
is process resulted in the identification of distinct
subclusters, namely, STMN1_Macro, APOE_Macro,
MARCO_Macro, JUN_Macro, FCN1_Mono, CD1C_DC,
CLEC9A_DC, FCGR3B_Neu and CCL5_Neu (Fig. 4A,
Table 1 and Table S10). APOE_Macro exhibited pro-
nounced expression of gene involved in lipid metabo-
lism, notably APOE, whereas STMN1_Macro presented
upregulated expression of proliferative marker, such as
STMN1 (Fig.4B & TableS11). FCGR3B_Neu increased
the expression of neutrophil maturation markers, includ-
ing FCGR3B, along with S100A8 and S100A9 (Fig. 4B
& Table S11). Additionally, CCL5_Neu exhibited high
expression of chemokines such as CCL5, similar to
N2-type (pro-tumour phenotype) neutrophils (Fig. 4B
& Table S11). Ro/e analysis and proportion analysis
revealed significant enrichment of CCL5_Neu, FCGR3B_
Neu, CLEC9A_DC, JUN_Macro, and MARCO_Macro in
LM. In contrast, differing trends were noted for APOE_
Macro STMN1_Macro and FCN1_Mono (Fig.4C, D).
e functional analysis of GO terms (Fig. S3A-B)
showed that both FCGR3B_Neu and CCL5_Neu were
associated with the upregulation of bacterial defense-
related pathways, such as "Response to molecules of
bacterial origin," and neutrophil chemotaxis-related
pathways, such as "Neutrophil chemotaxis." Additionally,
the KEGG term functional analysis (Fig. S3C-D) revealed
that FCGR3B_Neu was linked to the formation of neu-
trophil extracellular traps, while CCL5_Neu is primarily
involved in the IL-17 signaling pathway. By comparing
multiple gene signature scores indicative of cellular states,
we evaluated the functional phenotypes of macrophages
and DCs (TableS6). For macrophages, we incorporated
scores for M1 polarization, M2 polarization, angiogen-
esis, and phagocytosis. e M1 and M2 polarizations,
initially categorized as a basic distinction in vitro, are
linked to inflammatory and anti-inflammatory responses,
respectively. As an alternative, the scores for angiogen-
esis and phagocytosis may offer a more holistic depic-
tion of TAM attributes [19]. For DCs, we considered
scores indicative of activation, migration, and tolerance,
which mirror the characteristic functional phenotypes
of these cells. Our results demonstrated the highest M1
polarization, M2 polarization, and phagocytosis scores
for MARCO_Macro, whereas STMN1_Macro presented
the highest angiogenesis scores (Fig. 4E). Furthermore,
CLEC9A_DC presented the highest activation score,
whereas CD1C_DC presented the highest migration and
tolerance scores (Fig.4F).
Subsequently, we performed a comparative analysis
of the distinct cellular states of macrophages derived
from LM and PL (Fig. 4G). e results revealed that
(See figure on next page.)
Fig. 4 Characterization of myeloid cells between PL and LM of CRNELM patients. A UMAP plot illustrating distinct myeloid cell subclusters
across all samples, color-coded by cell type. B Heatmap depicting the expression of marker genes across 9 defined myeloid cell clusters. The colour
intensity reflects average scaled gene expression. The expression intensity of markers is shown. C The bar plot illustrates the relative proportions
of all myeloid cell subclusters between LM and PL. D The lollipop chart depicts the prevalence of all myeloid cell subclusters between LM and PL,
as quantified by the Ro/e ratio. Subclusters that favor LM are situated above the baseline, while those that favor PL are situated below. A higher
numerical value indicates a stronger preference, with an Ro/e ratio exceeding 1 (marked by dashed lines above and below) considered indicative
of a significant bias. E Heatmap displaying the expression of curated gene signatures across 4 macrophage subclusters. The heatmap is based
on scaled gene signature scores. F Heatmap displaying the expression of curated gene signatures across 2 dendritic cell subclusters. The heatmap
is based on scaled gene signature scores. G Boxplot comparing the M1 polarization (top left), M2 polarization (top right), angiogenesis (bottom left),
and phagocytosis (bottom right) scores of macrophages between LM and PL. The position of the macrophages is annotated below the boxplot.
The boxes represent the median (indicated by the horizontal line with a numerical value), encompass the second to third quartiles (forming
the body of the box), and extend with Tukey-style whiskers beyond the box to depict the data range. Wilcoxon rank-sum test. H Monocle trajectory
analysis of macrophages differentiation unveiled one primary divergent pathway. The top panel displays the pseudotime map composed of various
macrophage subsets, while the bottom panel illustrates the directional progression of pseudotime in a smaller, more focused pseudotime trajectory
plot. The cells are color-coded based on their pseudotime. (I) Correlation analysis between the pseudotime of LM (depicted in dark purple) and PL
(depicted in light purple), and the M1 polarization (top left), M2 polarization (top right), angiogenesis (bottom left), and phagocytosis (bottom right)
scores of macrophages. Each plot highlights the correlation coefficients and P-values for LM (above) and PL (below), utilizing Pearson correlation
and a generalized additive model. J Boxplot comparing the activation (left), migration (middle), and tolerance (right) scores of DCs between LM
and PL. The position of the DCs is annotated below the boxplot. The boxes represent the median (indicated by the horizontal line with a numerical
value), encompass the second to third quartiles (forming the body of the box), and extend with Tukey-style whiskers beyond the box to depict
the data range. Wilcoxon rank-sum test. K Monocle trajectory analysis of DCs differentiation unveiled one primary divergent pathway. The
top panel displays the pseudotime map composed of various DCs subsets, while the bottom panel illustrates the directional progression
of pseudotime in a smaller, more focused pseudotime trajectory plot. The cells are color-coded based on their pseudotime. L Correlation
analysis between the pseudotime of LM (depicted in dark orange) and PL (depicted in light orange), and the activation (left), migration (middle),
and tolerance (right) scores of DCs. Each plot highlights the correlation coefficients and P-values for LM (above) and PL (below), utilizing Pearson
correlation and a generalized additive model
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
Page 12 of 25
Dengetal. Molecular Cancer (2025) 24:28
Fig. 4 (See legend on previous page.)
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
Page 13 of 25
Dengetal. Molecular Cancer (2025) 24:28
macrophages from LM exhibited significantly higher
M2 polarization and phagocytosis scores compared
to those from PL, while M1 polarization and angio-
genesis scores were notably elevated in PL (Wilcoxon
rank-sum test, P < 0.01). e trajectory of macrophage
development was revealed by Monocle, commencing
with the emergence of APOE_Macro and culminating
in the appearance of STMN1_Macro, MARCO_Macro,
and JUN_Macro (Fig.4H). Across this pseudotime pro-
gression, macrophages originating from both LM and
PL exhibited declines in M1 Polarization score, M2
Polarization score, and angiogenesis score (Pearson
correlation, P < 0.01). However, based on correlation
coefficients, the decline appeared to be more rapid in
LM (Fig.4I). Additionally, macrophages from LM dem-
onstrated an increase in the phagocytosis score (Pear-
son correlation, P < 0.01), while those from PL showed
no significant change (Pearson correlation, P = 0.11)
(Fig.4I).
Similarly, we conducted a comparative analysis of the
distinct cellular states of DCs derived from LM and PL
(Fig.4J). e results indicated that DCs from PL exhib-
ited significantly higher scores for activation, migration,
and tolerance compared to those from LM (Wilcoxon
rank-sum test, P < 0.01). Pseudotime analysis further
demonstrated that CD1C_DC occupied the initial posi-
tions along the developmental trajectory, culminating in
CLEC9A_DC (Fig.4K). roughout this pseudotime pro-
gression, DCs from both PL and LM showed an increase
in activation and tolerance scores, with a more rapid
increase observed in PL based on correlation coefficient
support (Pearson correlation, P < 0.01; Fig. 4L). Addi-
tionally, DCs from PL exhibited an increase in migration
score, whereas those from LM showed a decrease (Pear-
son correlation, P < 0.01; Fig.4L).
Overall, these findings suggest that the TME of LM is
characterized by M2 polarization macrophages, whereas
the TME of PL harbour more activated DC cells, poten-
tially indicating a more pronounced antigen presentation
function.
LM characterized bythepresence ofactivated
andECM‑remodelling broblasts
To explore the possible link between fibroblasts and
malignant transformation, we performed a cluster-
ing analysis on the entire fibroblast population in our
dataset. is analysis revealed five distinct cell clusters,
namely, COLEC11_mCAFs, CFD_iCAFs, MYH11_
SMCs, RGS5_pericytes, and F3_telocytes (Fig. 5A,
Table 1 and TableS11). By evaluating the expression
of specific marker genes across these clusters, a clear
demarcation between specialized cells and fibroblasts
was observed. Specialized cells, namely, RGS5_peri-
cytes, presented prominent RGS5 expression, whereas
MYH11_SMCs presented diverse gene expression pat-
terns, including those of MYH11, among others (Fig. S4
and TableS12). COLEC11_mCAFs presented elevated
expression of genes encoding type I collagen (COL1A1
and COL1A2) and type III collagen (COL3A1), indicat-
ing the potential of these genes to alter the extracellu-
lar matrix (ECM). Additionally, high levels of CTHRC1,
a factor associated with T-cell infiltration inhibition
and immune escape of tumour cells, were detected
[35] (Fig. S4 and TableS12). Subsequent Ro/e and pro-
portion analyses revealed significant enrichment of
COLEC11_mCAFs and RGS5_pericytes in LM, whereas
Fig. 5 Characterization of fibroblasts between PL and LM of CRNELM patients. A UMAP plot illustrating distinct fibroblast subclusters across all
samples, color-coded by cell type. B The bar plot illustrates the relative proportions of all fibroblast subclusters between LM and PL. C The lollipop
chart depicts the prevalence of all fibroblast subclusters between LM and PL, as quantified by the Ro/e ratio. Subclusters that favor LM are situated
above the baseline, while those that favor PL are situated below. A higher numerical value indicates a stronger preference, with an Ro/e ratio
exceeding 1 (marked by dashed lines above and below) considered indicative of a significant bias. D, E Bubble plot showing gene set enrichment
analysis via GO_BP (D) and KEGG (E) term in distinct fibroblast subclusters. The color scale represents the normalized enrichment score (NES) value,
and the bubble size represents the gene enrichment ratio of each pathway. F Monocle trajectory analysis of all fibroblasts differentiation unveiled
two primary divergent pathways. The top panel displays the pseudotime map composed of various fibroblast subclusters, while the bottom panel
illustrates the directional progression of pseudotime in a smaller, more focused pseudotime trajectory plot. The cells are color-coded based on their
pseudotime. G Heatmap showing the dynamic DEGs and their enriched pathways along the pseudotime trajectory transition to COLEC11_mCAF.
These DEGs were divided into four main clusters. H Changes in common matrix genes (including COL1A1, COL1A2, FN1, VCAN, DCN and LUM)
during the pseudotime transition to COLEC11_mCAFs are shown. I Correlation analysis between the pseudotime of LM (depicted in dark blue)
and PL (depicted in light blue), and the inflammatory (top left), vascular development (top middle), ECM (top right), lipid process (bottom left),
antigen presentation (bottom middle) and proliferative (bottom right) scores of fibroblasts. Each plot highlights the correlation coefficients
and P-values for LM (above) and PL (below), utilizing Pearson correlation and a generalized additive model. J Boxplot comparing the inflammatory
(top left), vascular development (top middle), ECM (top right), lipid process (bottom left), antigen presentation (bottom middle) and proliferative
(bottom right) scores of fibroblasts between LM and PL. The position of the fibroblasts is annotated below the boxplot. The boxes represent
the median (indicated by the horizontal line with a numerical value), encompass the second to third quartiles (forming the body of the box),
and extend with Tukey-style whiskers beyond the box to depict the data range. Wilcoxon rank-sum test
(See figure on next page.)
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
Page 14 of 25
Dengetal. Molecular Cancer (2025) 24:28
Fig. 5 (See legend on previous page.)
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
Page 15 of 25
Dengetal. Molecular Cancer (2025) 24:28
the opposite trend was observed for CFD_iCAFs and
F3_telocytes in PL, no significant difference in the dis-
tribution of MYH11_SMC was observed between the
two groups. (Fig. 5B, C). Further enrichment analysis
revealed that COLEC11_mCAFs was involved in ECM
regulation through pathways such as "extracellular
matrix organization" and "ECM-receptor interaction."
Moreover, RGS5_pericytes were involved in pathways
affecting the tumour blood supply, such as "endothelial
development" and "vascular smooth muscle contrac-
tion." In contrast, CFD_iCAFs exhibited strong activity
in pathways related to immune activation and immune
response regulation, including the "immune response"
and "inflammatory response" (Fig.5D, E).
Subsequent analysis of dynamic immunological states
and cellular transformations revealed that the pseudo-
time trajectory started with MYH11_SMCs and sub-
sequently continued through various other cell types,
ultimately ending with CFD_iCAFs or COLEC11_
mCAFs (Fig.5F). To understand the transcriptional pro-
grams driving the formation of COLEC11_mCAFs, we
examined the differential gene expression profiles along
the developmental trajectory of these genes. Initially,
genes associated with ECM remodelling were highly
expressed, followed by those related to muscle contrac-
tion pathways. e final stage was characterized by
increased collagen metabolism and biogenesis, indicating
the acquisition of a profibrotic role during the develop-
ment trajectory (Fig.5G). Notably, matrix genes such as
COL1A1, COL1A2, FN1, VACN, DCN and LUM were
upregulated during the transition to COLEC11_mCAFs
(Fig.5H).
Further analysis of signature scores (TableS7) associ-
ated with fibroblast function along the pseudotime tra-
jectory revealed consistent trends for fibroblasts from
both PL and LM in terms of inflammatory score, vascu-
lar development score, ECM score, lipid process score,
antigen presentation score, and proliferation score [20]
(Pearson correlation, P < 0.01). However, based on the
magnitude of the correlation coefficients, it appears that
the ECM score of fibroblasts in LM increased more rap-
idly than that in PL; the vascular development score and
proliferation score decreased more swiftly in LM com-
pared to PL; whereas the inflammatory score, lipid pro-
cess score, and antigen presentation score of fibroblasts
in PL rose more quickly than those in LM (Fig.5I). Sub-
sequent comparisons confirmed that LM fibroblasts
exhibited significantly higher ECM scores compared to
PL fibroblasts, while their inflammatory score, vascular
development score, lipid process score, and antigen pres-
entation score were notably lower (Wilcoxon rank-sum
test, P < 0.01) (Fig. 5J). ese findings suggest that the
TME of LM is characterized by a prominent presence of
activated fibroblasts engaged in ECM remodelling, which
may drive tumour growth and progression.
Malignant epithelial cells withamore activated
andinvasive phenotype atLM sites
Chromosomal copy number variations (CNV) analysis
was conducted to deduce chromosomal CNV across all
epithelial cells (Fig. S5A). A randomized subsample com-
prising 500T or NK cells, 500 plasma cells, and 500 B
cells served as the reference. Upon assessing CNV data,
it was evident that a significant portion of epithelial cells
exhibited substantial deviations in their copy numbers
compared with those in the reference cells (Fig. S5A-C).
Consequently, all epithelial cells were categorized into
normal epithelial cells (NCs), LM malignant epithelial
cells (LM_MCs), and PL malignant epithelial cells (PL_
MCs) (Fig. 6A). PROGENy scores were subsequently
calculated to assess the activity of tumor-associated path-
ways in both PL_MCs and LM_MCs (Fig.6B), in order
to compare the differences between malignant epithelial
cells (MCs) originating from these two sites. LM_MCs
exhibited a strong positive correlation with the activation
of several tumor-related pathways, including JAK-STAT,
VEGF, TGFβ, and TRAIL pathways, indicating their
potential involvement in tumor invasion and metasta-
sis. In contrast, PL_MCs showed activation of pathways
such as PI3K, NFκB, p53, and Estrogen, suggesting that
MCs in PL and LM may activate distinct cancer-related
pathways. ese differences could underlie variations in
proliferation and invasion characteristics, highlighting
the need for more precise target identification for tai-
lored therapies. Furthermore, we compared the signature
scores associated with malignant biological behaviours
between PL_MCs and LM_MCs (Fig. 6C). e analysis
revealed that most of these signature scores, including
markers of malignant tumor progression such as angio-
genesis, apoptosis, differentiation, DNA damage, EMT,
hypoxia, inflammation, invasion, metastasis, prolifera-
tion, quiescence, and stemness, were significantly ele-
vated in LM_MCs, compared to PL_MCs. is suggests a
greater malignant potential in LM_MCs than in PL_MCs,
with the exceptions of DNA repair and the cell cycle.
Using Monocle, we investigated the dynamic cell tran-
sitions occurring in tumour states and identified two
primary differentiation trajectories, both originating
from NCs and then terminating at LM_MCs or PL_MCs
(Fig. 6D). To gain insight into the transcriptional pro-
grams driving the formation of LM_MCs, we analysed
the differential gene expression profiles along the tra-
jectory (Fig. 6E). Initially, genes involved in filament
organization signaling were highly expressed, followed
by those involved in stress-like response pathways. e
intermediate state was characterized by increased tissue
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
Page 16 of 25
Dengetal. Molecular Cancer (2025) 24:28
morphogenesis and epithelial-to-mesenchymal transi-
tion, followed by increased coagulation, haemostasis, and
wound healing, indicating the acquisition of metastatic
properties during tumour progression (Fig.6E).
Pivotal role ofCOLEC11_mCAFs insupporting LM_MCs
intheTME attheLM site
To investigate the underlying causes of differences
in the TME between PL and LM, we initially applied
the Augur algorithm to compare the transcriptional
responses of various cell subtypes in these regions. e
analysis revealed that COLEC11_mCAFs had the highest
response score in LM, followed by CD1C_DC, TRGC2_
NKT, and RGS5_pericytes (Fig. 7A). Furthermore, we
conducted a Spearman correlation analysis to explore
the relationships between the percentages of various cell
clusters in both the PL and LM (Fig. S6A-B). Intriguingly,
our findings revealed that while there was no significant
positive correlation between COLEC11_mCAFs and PL_
MCs in PL, a striking positive correlation was observed
between COLEC11_mCAFs and LM_MCs in LM
(Fig.7B, C). Likewise, the presence of RGS5_pericytes,
MYH11_SMCs, and stress-associated T or NK cells, spe-
cifically HSPA6_CD8_Tstr cells, FGFBP2_CD8_Teff cells,
FGFBP2_NK cells, and HSPA6_NK cells, also demon-
strated a shift from non-significant correlations in the PL
to significant positive correlations in the LM (Fig.7B, C).
Using CSOmap analysis, which inferred spatial informa-
tion from single-cell transcriptomics, we reconstructed
three-dimensional spatial maps for both the PL and LM
(Fig.7D, E & Figs. S7, S8) and assessed the connection
strength and significance between different cell subtypes.
e results indicated that in LM, the correlation between
the percentages of COLEC11_mCAFs and LM_MCs
was more pronounced than the correlation between the
percentages of COLEC11_mCAFs and PL_MCs in PL
(Fig.7D, E). Additionally, the association between stress-
like immune cells (HSPA6_CD8_Tstr, FGFBP2_CD8_
Teff, FGFBP2_NK, and HSPA6_NK) and major matrix
cells was different for CFD_iCAFs and RGS5_pericytes
(Figure S7-S8). ese findings provide compelling evi-
dence that COLEC11_mCAFs may serve a pivotal role
Fig. 6 Characterization of epithelial cells between PL and LM of CRNELM patients. A UMAP plot depicting all epithelial cells, coloured according
to cell type (left) and sample (right). B Heatmap illustrating the diverse PROGENy activity scores for LM_MCs (left) and PL_MCs (right), based
on normalized gene signature scores. Deeper tangerine hues signify higher enrichment levels. C Heatmap displaying the expression of curated
gene signatures across LM_MCs (left) and PL_MCs (right). The heatmap is based on scaled gene signature scores. D Monocle trajectory analysis
of epithelial cells differentiation unveiled one primary divergent pathway. The top panel displays the pseudotime map composed of various
epithelial cells subsets, while the bottom panel illustrates the directional progression of pseudotime in a smaller, more focused pseudotime
trajectory plot. The cells are color-coded based on their pseudotime. E Heatmap showing the dynamic DEGs and their enriched pathways
along the pseudotime trajectory transition from NCs to LM_MCs. These DEGs were divided into four main clusters
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
Page 17 of 25
Dengetal. Molecular Cancer (2025) 24:28
in supporting LM_MCs within the context of LM, while
RGS5_pericytes could potentially induce a stress-like
phenotype in immune cells in LM.
Consequently, we endeavour to explore potentially
unique receptor-ligand interactions in LM through the
analysis of communication among these cells. To study
intercellular communication between COLEC11_mCAF
and PL_MCs or LM_MCs, we employed Commpath
to delve into the biologically significant interactions
between PL and LM (Fig. S9A-B). Initially, we identi-
fied notable receptor-ligand interactions and activation
pathways in cell communication between COLEC11_
mCAF and LM_MCs (Fig.7F-G). Our findings revealed
that COLEC11_mCAF stimulates GLRA2 and FZD3
receptors on LM_MCs, activating membrane-associ-
ated pathways such as “Apicolateral plasma membrane”
and “Lateral plasma membrane” (Fig. 7F). Conversely,
LM_MCs primarily stimulate the GPC1 receptor on
COLEC11_mCAF, triggering pathways related to the
cellular matrix or defensin, like “extracellular matrix
binding” and “Defective B4GALT7 causing EDS, prog-
eroid type” (Fig.7G). In PL, COLEC11_mCAF activates
distinct receptors on PL_MCs—SDC4, FZD6, MET, and
F11R—which differ from those on LM_MCs, leading
to the activation of pathways involved in cellular tight
junctions and estrogen signaling in PL_MCs (Fig. 7H).
PL_MCs, in turn, predominantly stimulate NPR1, PDG-
FRA, EGFR, and GPC1 receptors on COLEC11_mCAF,
activating extracellular matrix-related pathways such as
“extracellular matrix binding” and “extracellular matrix”
(Fig.7I). Although GPC1 has been identified as a recep-
tor on COLEC11_mCAF stimulated by malignant epithe-
lial cells in both LM and PL, the presence of additional
receptors such as NPR1, PDGFRA, and EGFR in PL sug-
gests that therapies targeting GPC1 may be more effec-
tive specifically against LM.
Furthermore, we have pinpointed distinct receptor-
ligand pairs within the cell communication network. In
this chain, COLEC11_mCAFs and either LM_MCs (Fig.
S9C) or PL_MCs (Fig. S9D) function as upstream cel-
lular subsets. RGS5_pericytes serve as intermediate cel-
lular connectors, while immune cells act as downstream
components. In the context of LM, COLEC11_mCAFs
or LM_MCs activate RGS5_pericytes by stimulating
Fig. 7 Pivotal role of COLEC11_mCAFs in supporting LM_MCs in the TME at the LM site. A Bar chart displaying the Augur scores of the cell types
across all the cell clusters. The length of each bar indicates the Augur score, with longer bars indicating a stronger association with the LM site. B,
C Correlations between the proportions of selected cell clusters in LM (B) and PL (C); redder hues indicate stronger correlations between the two
cell subpopulations; Spearman correlation coefficient was used. D CSOmap unveiling the spatial distribution of COLEC11_mCAF and LM_
MCs in LM. Each dot represents a cell, and its color represents the corresponding cell state. The upper panel displays their spatial locations,
while the lower table summarizes the connection strength between COLEC11_mCAF and LM_MCs calculated using the CSOmap algorithm; ***
indicates P < 0.001. E CSOmap unveiling the spatial distribution of COLEC11_mCAF and PL_MCs in PL. Each dot represents a cell, and its color
represents the corresponding cell state. The upper panel displays their spatial locations, while the lower table summarizes the connection strength
between COLEC11_mCAF and PL_MCs calculated using the CSOmap algorithm; *** indicates P < 0.001. F Pathway-mediated communication
from LM_MCs to COLEC11_mCAFs in LM via CommPath. The figure displays the top 4 significantly enriched pathways in COLEC11_mCAFs
compared to other cells within LM, along with the corresponding ligand-receptor interactions. Thicker lines between ligands and receptors
indicate stronger signal intensity. The dots at the receptors represent the average log2 fold change (log2FC) and -log10(P) value of receptor
expression in COLEC11_mCAFs relative to all other cells in LM, with larger and darker dots indicating higher expression and significance. The bars
in the Pathway annotation column represent the average difference in pathway scores and the -log10(P) value of pathway scores for COLEC11_
mCAFs in LM compared to all other cells, with longer and darker bars indicating greater differences and significance. G Pathway-mediated
communication from COLEC11_mCAFs to LM_MCs in LM via CommPath. The figure displays the top 4 significantly enriched pathways in LM_MCs
compared to other cells within LM, along with the corresponding ligand-receptor interactions. Thicker lines between ligands and receptors indicate
stronger signal intensity. The dots at the receptors represent the average log2 fold change (log2FC) and -log10(P) value of receptor expression
in LM_MCs relative to all other cells in LM, with larger and darker dots indicating higher expression and significance. The bars in the Pathway
annotation column represent the average difference in pathway scores and the -log10(P) value of pathway scores for LM_MCs in LM compared
to all other cells, with longer and darker bars indicating greater differences and significance. H Pathway-mediated communication from PL_MCs
to COLEC11_mCAFs in PL via CommPath. The figure displays the top 4 significantly enriched pathways in COLEC11_mCAFs compared to other
cells within PL, along with the corresponding ligand-receptor interactions. Thicker lines between ligands and receptors indicate stronger signal
intensity. The dots at the receptors represent the average log2 fold change (log2FC) and -log10(P) value of receptor expression in COLEC11_mCAFs
relative to all other cells in PL, with larger and darker dots indicating higher expression and significance. The bars in the Pathway annotation column
represent the average difference in pathway scores and the -log10(P) value of pathway scores for COLEC11_mCAFs in PL compared to all other cells,
with longer and darker bars indicating greater differences and significance. I Pathway-mediated communication from COLEC11_mCAFs to PL_MCs
in PL via CommPath. The figure displays the top 4 significantly enriched pathways in PL_MCs compared to other cells within PL, along with the
corresponding ligand-receptor interactions. Thicker lines between ligands and receptors indicate stronger signal intensity. The dots at the receptors
represent the average log2 fold change (log2FC) and -log10(P) value of receptor expression in PL_MCs relative to all other cells in PL, with larger
and darker dots indicating higher expression and significance. The bars in the Pathway annotation column represent the average difference
in pathway scores and the -log10(P) value of pathway scores for PL_MCs in PL compared to all other cells, with longer and darker bars indicating
greater differences and significance
(See figure on next page.)
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
Page 18 of 25
Dengetal. Molecular Cancer (2025) 24:28
Fig. 7 (See legend on previous page.)
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
Page 19 of 25
Dengetal. Molecular Cancer (2025) 24:28
receptors such as SDC2, DDR2, and FGFR1. Subse-
quently, these activated RGS5_pericytes influence
downstream immune cells through ligands like APOE,
THBS1, LAMB2, TGFB2, and FGF2. Conversely, in PL,
COLEC11_mCAFs or PL_MCs activate RGS5_pericytes
via receptors including PDGFRB, CD36, and NRP1. e
activated RGS5_pericytes then primarily affect down-
stream immune cells through ligands such as LAMB2,
THBS1, TGFB3, and LAMA2. Notably, targeting unique
receptors in LM, such as SDC2, DDR2, and FGFR1,
or specific ligands like APOE, TGFB2, and FGF2, may
potentially reverse the impact of RGS5_pericytes on
downstream immune cells. is suggests that these
receptors and ligands could serve as promising thera-
peutic targets for modulating cell communication and
immune responses in the respective contexts.
Validation analyses viaST analysis, bulk RNA sequencing
andmIHC/IF
By integrating single-cell RNA sequencing with ST
data (Fig. 8 and Fig. S10), we have precisely deline-
ated the spatial distribution patterns of diverse cellu-
lar subpopulations within the TME. We first presented
the single-stranded DNA (ssDNA) imaging of the LM
(Fig.8A) and PL (Fig.8B) based on cell nucleus loca-
tions, as previously described [14]. Utilizing CARD
deconvolution, we observed that COLEC11_mCAFs
appear to be more abundant in LM compared to PL
(Fig.8C-D). To further substantiate these findings, we
generated scores via AddModuleScore function based
on gene sets specific to COLEC11_mCAFs (TableS12)
and found that the COLEC11_mCAF scores were sig-
nificantly higher in LM than in PL (Wilcoxon rank-
sum test, P < 0.01; Fig. 8E), corroborating our previous
conclusions. To further extend and validate our con-
clusions, we obtained two larger datasets (GSE211485
and GSE27162) of bulk sequencing from the GEO.
rough deconvolution of bulk RNA sequencing data
via CIBERSORTX, we revealed a significant associa-
tion between high proportions of COLEC11_mCAFs
and poor prognosis in a pan-NET cohort (Fig. 8F &
TableS13; GSE211485), underscoring its clinical prog-
nostic relevance. Concurrently, the proportion of
COLEC11_mCAFs was also elevated in LM (Fig.8G &
TableS14; GSE27162). Additionally, independent vali-
dation via mIHC/IF confirmed increased COLEC11_
mCAF infiltration in LM (Fig. 8H-I). ese findings
further substantiate our initial observations, suggest-
ing that COLEC11_mCAFs may play a critical role in
tumor metastasis and progression.
e stress score employed in this study is derived
from characteristic gene sets that have been previ-
ously reported or are widely utilized for CD8, CD4,
NK/NKT, and B cells, with each cell type possessing a
unique set of genes. As a result, relying solely on a sin-
gle stress score to assess overall differences between PL
and LM may yield inaccurate results. To address this,
we have identified and selected HSPA6_CD8_Tstr and
HSPA6_NK as representative stress-like phenotypic
cells from our study for subsequent spatial analysis and
bulk sequencing validation (Fig. S10). Utilizing CARD
deconvolution, we observed that both HSPA6_CD8_
Tstr and HSPA6_NK appeared to be more abundant
in LM compared to PL (Fig. S10A-F). To further cor-
roborate these findings, we derived scores using the
AddModuleScore function with gene sets specific to
HSPA6_CD8_Tstr (Fig.8G & TableS9) and HSPA6_NK
(Fig.8H & TableS9). e results revealed that the cor-
responding scores were significantly elevated in LM
compared to PL (Wilcoxon rank-sum test, P < 0.01)
(Fig. S10G-H), thereby reinforcing our earlier conclu-
sions. rough deconvolution of bulk RNA sequenc-
ing data (GSE27162) via CIBERSORTX, we revealed a
significant association between higher proportions of
both HSPA6_CD8_Tstr and HSPA6_NK in LM com-
pared to PL (Fig. S10I). However, there were no sig-
nificant differences in survival outcomes associated
with the proportions of HSPA6_CD8_Tstr (Fig. S10J)
and HSPA6_NK (Fig. S10K) in this pan-NET cohort
(GSE211485).
(See figure on next page.)
Fig. 8 Multidimensional validation of COLEC11_mCAF through spatial transcriptomics analysis, bulk RNA sequencing datasets, and multiplex
immunohistochemistry/immunofluorescence (mIHC/IF) analysis. A, B ssDNA images of LM (A) and PL (B). C, D Spatial distributions
of the proportions of COLEC11_mCAFs in LM (C) and PL (D), as estimated by CARD. E Boxplot comparing the representative scores of COLEC11_
mCAF between LM and PL. The position is annotated below the boxplot. The boxes represent the median (indicated by the horizontal
line with a numerical value), encompass the second to third quartiles (forming the body of the box), and extend with Tukey-style whiskers
beyond the box to depict the data range. Wilcoxon rank-sum test. F Kaplan–Meier (KM) survival curve stratified by the proportion of COLEC11_
mCAFs in the GSE211485 cohort. The cutoff value for stratification was determined using the survdiff method. A log-rank test was employed
to assess whether the difference in survival between the two groups was statistically significant. If the P-value was less than 0.01, it was considered
that the survival difference between the two groups was significant. G The distribution of distinct fibroblast subclusters in the GSE27162 cohort
was compared between two groups using the Wilcoxon rank-sum test. H mIHC/IF images showing positive expression of COLEC11_mCAFs in LM
(right) and PL (left). Scale bar, 50 µm. I Quantification of the density of COLEC11_mCAFs between LM and PL via the Wilcoxon rank-sum test
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
Page 20 of 25
Dengetal. Molecular Cancer (2025) 24:28
Fig. 8 (See legend on previous page.)
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
Page 21 of 25
Dengetal. Molecular Cancer (2025) 24:28
Discussion
CRNELM are rare but aggressive cancers, exhibiting sig-
nificant variability in behaviour between primary tumors
and liver metastases [36, 37]. is research constructs a
comprehensive single-cell atlas that illuminates the heter-
ogeneity of the TME between LM and PL, revealing dis-
tinct cellular profiles and interactions critical for tumor
progression and liver metastasis. By examining immune
cells, fibroblasts and MCs at the single-cell level, the
study offers essential insights into tumor heterogeneity.
Understanding these differences is crucial for identifying
therapeutic targets and developing personalized treat-
ment strategies, ultimately leading to improved clinical
outcomes for patients with CRNELM.
Immunotherapy has emerged as a significant advance-
ment in cancer treatment, showing promising results
across various solid tumors [38, 39]. However, its effi-
cacy in treating CRNELM remains disappointing [40,
41]. Response rates to monoclonal antibodies targeting
programmed cell death protein 1 (PD-1) and its ligand
(PD-L1) in metastatic neuroendocrine neoplasms are
even below 10% [41]. Consequently, enhancing the effi-
cacy of immunotherapy for CRNELM has become a
clinical focal point of interest. Recently, stress-like phe-
notype immune cells have garnered significant attention,
as they are considered a potential cell subpopulation that
may influence immunotherapy efficacy. Previous studies
have explored the relationships between this stress-like
phenotype and the response to immunotherapy as well as
prognosis across various tumors [15, 17, 18]. For exam-
ple, research by Chu etal. indicated that CD8 + /CD4 + T
cells exhibiting a stress-like phenotype were significantly
enriched among non-responders to immunotherapy,
accompanied by notable upregulation of heat shock
protein-related genes [15]. Similarly, Tang etal. reported
that DNAJB1 + NK cells with a stress-like phenotype
were particularly enriched in tumor tissues, correlating
with impaired antitumor function, poor prognosis, and
resistance to immunotherapy [17]. Additionally, Yang
etal. found that stress-like memory B cells are prevalent
in tumors, significantly associated with low survival rates
across various cancer types [18]. ese findings indicate
that the stress-like phenotype immune cells could be a
widespread biological phenomenon rather than confined
to specific tumor types. However, none of these stud-
ies investigated stress-like phenotype immune cells in
CRNELM. In our study, we first observed that CD8, CD4,
NK/NKT, and B cells within the TME of LM exhibited
more pronounced stress-like phenotypic characteristics
compared to those in PL. Conversely, the TME of PL was
characterized by an increased presence of IgG-positive
plasma cells, known for their antitumor properties [32–
34], suggesting a heightened humoral immune response
in PL. e prevalence of stress-like phenotype immune
cells in LM may be a crucial factor contributing to the
poor response to immunotherapy in CRNELM.
e mechanistic underpinnings of stress-like pheno-
type immune cells and their limited efficacy for immu-
notherapy in these tumors remain unclear. One potential
explanation is that the stress-like phenotype alters the
immune landscape within the TME [15, 17, 18]. ese
immune cells may exhibit enhanced immunosuppres-
sive functions, such as increased production of immu-
nosuppressive regulatory cytokines [15, 17, 18], which
can inhibit the activity of cytotoxic T cells and NK cells.
is allows tumor cells to evade immune surveillance,
promoting tumor growth and metastasis [42, 43]. Addi-
tionally, related studies suggest that stress-like phe-
notype cell subpopulations typically arise early in cell
development and are relatively conserved and immature,
potentially leading to restricted immune cell develop-
ment and impaired functionality. In the current study,
we also found that, particularly in CD8 + T cells and NK
cells, which have traditionally been considered cyto-
toxic, an increase in the stress score was significantly
correlated with an elevation in the inhibitory score and
a decrease in the cytotoxicity score. Moreover, these cells
exhibiting stress-like characteristics tended to show an
enhanced activation score. is result may suggest that
prolonged and continuous stimulation could induce a
state of exhaustion in these cells, leading to a loss of their
tumor-killing ability. Targeting these stress-like immune
cells may enhance the effectiveness of immunotherapy,
potentially leading to improved clinical outcomes. is
study offers a new direction for targeting these immune
cells, specifically focusing on RGS5-positive pericytes
and associated receptors/ligands that might regulate
this stress-like phenotype. Our findings provide new
insights and potential therapeutic targets for addressing
this immunosuppressive state in patients with CRNELM.
However, further investigation of these findings is essen-
tial, particularly regarding their application in clinical
settings for patients with CRNELM.
Recent research has increasingly underscored the
role of stromal cells in promoting tumor progres-
sion and metastasis [44–46]. In this study, we identi-
fied COLEC11_mCAFs and observed their significant
accumulation in LM through ST data, additional GEO
cohorts, and mIHC/mIF analyses. COLEC11_mCAFs
exhibited elevated expression of ECM-related genes,
such as LUM, COL1A1, and COL1A2, indicating their
activation and involvement in ECM remodelling. Previ-
ous studies have demonstrated that cancer-associated
fibroblast activation of ECM-related signaling pathways
can lead to stroma-associated fibrosis, promoting tumor
cell migration and increasing vascular permeability [47,
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
Page 22 of 25
Dengetal. Molecular Cancer (2025) 24:28
48]. Notably, COLEC11_mCAFs showed the greatest
transcriptomic responsiveness to the LM phenotype
and exhibited the strongest correlation with LM_MCs,
suggesting that these fibroblasts may facilitate tumor
progression and be activated by LM_MCs in the meta-
static environment. Additionally, a high proportion of
COLEC11_mCAFs was associated with reduced over-
all survival. Our cell communication analysis identified
potential receptors involved in interactions between
COLEC11_mCAFs and LM_MCs, including GLRA2, a
known predictor of recurrence and prognosis in colon
cancer, and FZD3, which is upregulated in various
tumors and may influence tumor progression through
the Wnt signaling pathway [49–52]. PROGENy analy-
sis confirmed an increase in Wnt signaling activity in
LM_MCs compared to PL_LMs. Furthermore, GPC1, an
extracellular matrix component, has been implicated in
fibroblast activation [53, 54]. Targeting FZD3, GLRA2,
or GPC1 disrupted the interactions between COLEC11_
mCAFs and LM_MCs, potentially inhibiting tumor pro-
liferation and growth in LM.
e single-cell analysis of MCs in CRNELM presented
in this study provides valuable insights into the hetero-
geneity and potential molecular drivers of epithelial cell
populations during tumor progression. e distinct CNV
patterns identified among epithelial cells, which classify
them into NCs, LM_MCs, and PL_MCs, underscore the
complexity of tumor evolution and metastatic spread.
Signaling pathways play a central role in tumor metasta-
sis, and their complexity and interconnectivity warrant
further investigation. Understanding the mechanisms of
these pathways can aid in the development of new com-
bination therapies, offering more precise guidance for
personalized treatment strategies. Strong positive cor-
relations between LM_MCs and the activation of key
tumor cell pathways, such as JAK-STAT [55] and VEGF
[56], highlight their critical role in tumor invasion and
metastasis. Furthermore, a comparative analysis of signa-
ture scores associated with malignant behavior between
PL and LM illustrates the heightened malignant potential
of LM_MCs, reflecting molecular differences that drive
metastatic aggressiveness. In summary, these results
suggest that targeting the JAK-STAT, VEGF, TGFβ, and
TRAIL signaling pathways may make metastatic tumor
cells more sensitive to treatment, and further exploration
of these pathways is warranted.
is study faced several limitations. A primary con-
straint was the limited availability of suitable single-cell
specimens, which could contribute to biases in cell clas-
sification and pathway analysis. Additionally, the rar-
ity of CRNELM has resulted in a lack of extensive bulk
sequencing data and large clinical cohorts to validate
our findings. We aim to increase our research efforts by
expanding sample collection and establishing a robust
clinical cohort, thereby facilitating more comprehensive
insights into our research area.
Overall, our study, for the first time, reveals the cellular
and molecular characteristic differences between the PL
and LM of CRNELM patients from a single-cell perspec-
tive. We identified specific cell subsets and associated
receptor-ligand interactions that might contribute to the
discrepancies of TME and support the metastatic tumor
cells. ese findings offer insights into potential thera-
peutic targets and improved management strategies for
CRNELM patients.
Supplementary Information
The online version contains supplementary material available at https:// doi.
org/ 10. 1186/ s12943- 025- 02231-y.
Supplementary Material 1: Figure S1. Quality control and processing of
samples. (A) Parameters of single-cell RNA-seq data before quality control.
(B) Parameters of single-cell RNA-seq data after quality control. (C) UMAP
plot of samples before the removal of the batch effect, coloured by sam-
ple. (D) UMAP plot of samples after the removal of the batch effect using
Harmony, coloured according to sample type.
Supplementary Material 2: Figure S2. Additional scRNA profiles of T and
NK cell subclusters in CRNELM tumor microenvironment. (A-C) Heatmap
depicting the expression of marker genes across 6 defined CD8+ T cell
clusters (A), 5 defined NK and NKT cell clusters (B) and 3 defined CD4+
T cell clusters (C). The colour intensity reflects average scaled gene
expression. The expression intensity of markers is shown. (D-F) Violin plots
displaying the scores for four common functional states—namely activa-
tion score, cytotoxicity score, stress score, and inhibitory score—across
different CD8+ T cell clusters (D), NK and NKT cell clusters (E), and CD4+
T cell clusters (F), respectively. The median scores for each cell subpopula-
tion are shown, with subpopulation labels provided at the bottom.
Supplementary Material 3: Figure S3. Additional characteristics of neu-
trophils in CRNELM tumor microenvironment. (A) Bubble plot of GO_BP
pathways for DEGs between FCGR3B_Neu clusters and other myeloid cell
clusters. The intensity represents the adjusted p-value of each pathway.
Dot size shows gene count for each pathway. Wilcoxon rank-sum test was
used to assess the difference. (B) Bubble plot of GO_BP pathways for DEGs
between CCL5_Neu clusters and other myeloid cell clusters. The intensity
represents the adjusted p-value of each pathway. Dot size shows gene
count for each pathway. Wilcoxon rank-sum test was used to assess the
difference. (C) Bubble plot of KEGG pathways for DEGs between FCGR3B_
Neu clusters and other myeloid cell clusters. The intensity represents the
adjusted p-value of each pathway. Dot size shows gene count for each
pathway. Wilcoxon rank-sum test was used to assess the difference. (D)
Bubble plot of KEGG pathways for DEGs between CCL5_Neu clusters and
other myeloid cell clusters. The intensity represents the adjusted p-value
of each pathway. Dot size shows gene count for each pathway. Wilcoxon
rank-sum test was used to assess the difference.
Supplementary Material 4: Figure S4. Additional analysis of fibroblasts in
CRNELM tumor microenvironment. Heatmap depicting the expression
of marker genes across 5 defined fibroblast clusters. The colour intensity
reflects average scaled gene expression. The expression intensity of mark-
ers is shown.
Supplementary Material 5: Figure S5. Additional characteristics of epithe-
lial cells in the CRNELM tumor microenvironment. (A-C) Process for select-
ing malignant epithelial cells from normal epithelial cells in this study: (A)
Heatmap depicting an overall CNV plot, with CNV patterns clustered into
groups 1-9 and labeled on the bottom of the figure; (B) Violin plot show-
ing that the CNV levels of cluster 3 are visibly lower than all other clusters,
and that the reference T, NK, and NKT B cells are all included in cluster
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
Page 23 of 25
Dengetal. Molecular Cancer (2025) 24:28
3. Therefore, cluster 3 is considered to represent normal epithelial
cells, while the other clusters are considered to represent malignant
epithelial cells; (C) Boxplot comparing CNV scores between annotated
normal epithelial cells (NCs), PL malignant epithelial cells (PL_MCs), and
LM malignant epithelial cells (LM_MCs).
Supplementary Material 6: Figure S6. Correlation analysis of all cell
subsets in the CRNELM tumor microenvironment for PL and LM. (A)
Shows the situation in LM, and (B) shows the situation in PL. Specifically,
the color of a square corresponding to two cell subsets indicates the
strength and direction of their correlation: a greener color (+1) repre-
sents a stronger positive correlation, a browner color (-1) represents
a stronger negative correlation, and a whiter color (0) represents a
weaker correlation. Significance was calculated using Spearman corre-
lation, with * indicating P<0.05, ** indicating P<0.01, and *** indicating
P<0.001.
Supplementary Material 7: Figure S7. The cellular compartments of LM
in CRNELM can be recapitulated through ligand-receptor based infer-
ence with CSOmap. (A) 3D visualization of CSOmap’s prediction of cell
subclusters in LM, with annotations for each cell subset provided on
the left. (B) Heatmap depicting the communication strength between
different cell subsets in LM. Specifically, a redder color in a square
corresponding to two cell types indicates stronger communication,
while a yellower color indicates weaker communication. Cell types are
annotated on the left and bottom of the table. (C) Heatmap showing
the communication probability between different cell subsets in LM. In
this context, a redder color in a square corresponding to two cell types
indicates a smaller P-value, with four categories of significance levels: P
> 0.1, P < 0.1, P < 0.05, and P < 0.01. Cell types are annotated on the left
and bottom of the table.
Supplementary Material 8: Figure S8. The cellular compartments of
PL in CRNELM can be recapitulated through ligand-receptor based
inference with CSOmap. (A) 3D visualization of CSOmap’s prediction of
cell subclusters in PL, with annotations for each cell subset provided on
the left. (B) Heatmap depicting the communication strength between
different cell subsets in PL. Specifically, a redder color in a square
corresponding to two cell types indicates stronger communication,
while a yellower color indicates weaker communication. Cell types are
annotated on the left and bottom of the table. (C) Heatmap showing
the communication probability between different cell subsets in PL. In
this context, a redder color in a square corresponding to two cell types
indicates a smaller P-value, with four categories of significance levels: P
> 0.1, P < 0.1, P < 0.05, and P < 0.01. Cell types are annotated on the left
and bottom of the table.
Supplementary Material 9: Figure S9. Single-cell transcriptional analysis
revealed the cell‒cell crosstalk network via Commpath in CRNELM
tumor microenvironment. (A, B) Circos plot showing the counts of
functional LR interactions among distinct cell clusters (types) identi-
fied in LM (A) and PL (B). In this plot, the thickness of the connecting
lines between two cell types indicates the number of interactions,
with thicker lines representing more communication. The length of
each segment corresponding to a specific cell type reflects the total
communication between that cell type and all other types, with longer
segments representing higher overall communication. (C) Pathway-
mediated communication involving COLEC11_mCAFs and LM_MCs as
upstream cells; RGS5_pericytes as median cells; and HSPA6_CD8_Tstr,
FGFBP2_CD8_Teff, FGFBP2_NK, HSPA6_NK, KLRC1_NK, and TRGC2_NKT
cells as downstream cells in liver metastases. The top 4 significantly
enriched pathways in RGS5_pericytes compared with other cells in
LM and the relevant LR interactions are shown. Thicker lines between
ligands and receptors indicate stronger signal intensity. The dots at
the receptors represent the average log2 fold change (log2FC) and
-log10(P) value of receptor expression in RGS5_pericytes relative to all
other cells in LM, with larger and darker dots indicating higher expres-
sion and significance. The bars in the Pathway annotation column
represent the average difference in pathway scores and the -log10(P)
value of pathway scores for RGS5_pericytes in LM compared to all
other cells, with longer and darker bars indicating greater differences
and significance. (D) Pathway-mediated communication involving
COLEC11_mCAFs and PL_MCs as upstream cells; RGS5_pericytes as
median cells; and HSPA6_CD8_Tstr, FGFBP2_CD8_Teff, FGFBP2_NK,
KLRC1_NK, and TRGC2_NKT cells as downstream cells in PL. The top 4 sig-
nificantly enriched pathways in RGS5_pericytes compared with other cells
in PL and the relevant LR interactions are shown. Thicker lines between
ligands and receptors indicate stronger signal intensity. The dots at the
receptors represent the average log2 fold change (log2FC) and -log10(P)
value of receptor expression in RGS5_pericytes relative to all other cells in
PL, with larger and darker dots indicating higher expression and signifi-
cance. The bars in the Pathway annotation column represent the average
difference in pathway scores and the -log10(P) value of pathway scores for
RGS5_pericytes in PL compared to all other cells, with longer and darker
bars indicating greater differences and significance.
Supplementary Material 10: Figure S10. Multidimensional validation of
HSPA6_CD8_Tstr and HSPA6_NK through spatial transcriptomics analysis
and bulk RNA sequencing datasets analysis. (A) ssDNA image of LM. (B-C)
Spatial distributions of the proportions of HSPA6_CD8_Tstr (B) and HSPA6_
NK (C) in LM, as estimated by CARD. (D) ssDNA image of PL. (E-F) Spatial
distributions of the proportions of HSPA6_CD8_Tstr (E) and HSPA6_NK (F)
in LM, as estimated by CARD. (G) Boxplot comparing the representative
scores of HSPA6_CD8_Tstr between LM and PL. The position is annotated
below the boxplot. The boxes represent the median (indicated by the hor-
izontal line with a numerical value), encompass the second to third quar-
tiles (forming the body of the box), and extend with Tukey-style whiskers
beyond the box to depict the data range. Wilcoxon rank-sum test. (H)
Boxplot comparing the representative scores of HSPA6_NK between LM
and PL. The position is annotated below the boxplot. The boxes represent
the median (indicated by the horizontal line with a numerical value),
encompass the second to third quartiles (forming the body of the box),
and extend with Tukey-style whiskers beyond the box to depict the data
range. Wilcoxon rank-sum test. (I) The distribution of all distinct T, NK and
NKT subclusters in the GSE27162 cohort was compared between LM and
PL using the Wilcoxon rank-sum test. (J) KM survival curve stratified by the
proportion of HSPA6_CD8_Tstr in the GSE211485 cohort. The cutoff value
for stratification was determined using the survdiff method. A log-rank
test was employed to assess whether the difference in survival between
the two groups was statistically significant. If the P-value was less than
0.01, it was considered that the survival difference between the two
groups was significant. (K) KM survival curve stratified by the proportion
of HSPA6_NK in the GSE211485 cohort. The cutoff value for stratification
was determined using the survdiff method. A log-rank test was employed
to assess whether the difference in survival between the two groups was
statistically significant. If the P-value was less than 0.01, it was considered
that the survival difference between the two groups was significant.
Supplementary Material 11.
Supplementary Material 12.
Acknowledgements
None.
Authors’ contributions
H.Z., Z.H., X.W., X.B., and T.Y. designed the study; Y.Q., Q.C. and C.G. performed
the research; Y.Q., Q.C. and C.G. analysed the data; Y.Q., Q.C., and C.G. wrote and
revised the paper. All the authors read and approved the final manuscript.
Funding
This study was supported by the National Key Research and Development
Program of China (2023YFC3403800, 2023YFC3403804), the National Natural
Science Foundation of China (82141127), the China Postdoctoral Science
Foundation (no.2024M763807), the CAMS Innovation Fund for Medical Sci-
ences (CIFMS) (2021-I2M-C&T-B-057), the Non-Profit Central Research Institu-
tion Fund of the Chinese Academy of Medical Sciences (No. 2019PT310026)
and the Sanming Project of Medicine in Shenzhen (No. SZSM202011010).
Data availability
The data reported in this study have been deposited in the China National
Center for Bioinformation, Beijing Institute of Genomics, Chinese Academy of
Sciences (https:// www. cncb. ac. cn/ servi ces). Raw data of single-cell sequences
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
Page 24 of 25
Dengetal. Molecular Cancer (2025) 24:28
can be retrieved from the GSA-Human database (https:// ngdc. cncb. ac. cn/
gsa- human/ browse/ HRA00 9693). The published data used for validation or
comparison in this study were retrieved from the NCBI GEO database under
accession codes GSE211485 and GSE27162. The remaining data are available
within the article, Supplementary Information files and Extended Data files.
All other data supporting the findings of this study are available from the cor-
responding author. Source data are provided with this paper.
Declarations
Ethics approval and consent to participate
The study was approved by the Institutional Review Board of the Cancer Hos-
pital, Chinese Academy of Medical Sciences (ID: NCC2021C-515), and informed
consent was obtained from all the patients. The patients/participants provided
written informed consent to participate in this study. The authors are account-
able for all aspects of the work and for ensuring that questions related to the
accuracy or integrity of any part of the work are appropriately investigated
and resolved.
Consent for publication
Not applicable.
Competing interests
The authors declare no competing interests.
Author details
1 Department of Hepatobiliary Surgery, National Cancer Center/National
Clinical Research Center for Cancer/Cancer Hospital, Chinese Academy
of Medical Sciences and Peking Union Medical College, Beijing 100021, China.
2 Department of Colorectal Surgery, State Key Laboratory of Oncology in South
China, Guangdong Provincial Clinical Research Center for Cancer, Sun Yat-Sen
University Cancer Center, Guangzhou 510060, People’s Republic of China.
3 Department of Anesthesiology, National Cancer Center/National Clinical
Research Center for Cancer/Cancer Hospital, Chinese Academy of Medical
Sciences and Peking Union Medical College, Beijing, China. 4 State Key Labora-
tory of Molecular Oncology, National Cancer Center/National Clinical Research
Center for Cancer/Cancer Hospital, Chinese Academy of Medical Sciences
and Peking Union Medical College, Beijing 100021, China.
Received: 10 September 2024 Accepted: 9 January 2025
References
1. Yao JC, Hassan M, Phan A, et al. One hundred years after “carcinoid”:
epidemiology of and prognostic factors for neuroendocrine tumors in
35,825 cases in the United States. J Clin Oncol. 2008;26(18):3063–72.
2. Sahani DV, Bonaffini PA, Fernández-Del Castillo C, et al. Gastroentero-
pancreatic neuroendocrine tumors: role of imaging in diagnosis and
management. Radiology. 2013;266(1):38–61.
3. Bagante F, Spolverato G, Merath K, et al. Neuroendocrine liver
metastasis: The chance to be cured after liver surgery. J Surg Oncol.
2017;115(6):687–95.
4. Riihimäki M, Hemminki A, Sundquist K, et al. The epidemiology of metas-
tases in neuroendocrine tumors. Int J Cancer. 2016;139(12):2679–86.
5. Pavel M, Grossman A, Arnold R, et al. ENETsS consensus guidelines for the
management of brain, cardiac and ovarian metastases from neuroendo-
crine tumors. Neuroendocrinology. 2010;91(4):326–32.
6. Jagannath P, Chhabra D, Shrikhande S, et al. Surgical treatment of liver
metastases in neuroendocrine neoplasms. Int J Hepatol. 2012;2012:
782672.
7. Touzios JG, Kiely JM, Pitt SC, et al. Neuroendocrine hepatic metasta-
ses: does aggressive management improve survival? Ann Surg. 2005
May;241(5):776–83; discussion 783–5.
8. Hinshaw DC, Shevde LA. The Tumor Microenvironment Innately Modu-
lates Cancer Progression. Cancer Res. 2019;79(18):4557–66.
9. de Visser KE, Joyce JA. The evolving tumor microenvironment: From can-
cer initiation to metastatic outgrowth. Cancer Cell. 2023;41(3):374–403.
10. Zhou Y, Liu S, Liu C, et al. Single-cell RNA sequencing reveals spatiotem-
poral heterogeneity and malignant progression in pancreatic neuroen-
docrine tumour. Int J Biol Sci. 2021;17(14):3760–75.
11. Bischoff P, Trinks A, Wiederspahn J, et al. The single-cell transcriptional
landscape of lung carcinoid tumours. Int J Cancer. 2022;150(12):2058–71.
12. Korsunsky I, Millard N, Fan J, et al. Fast, sensitive and accurate integration
of single-cell data with Harmony. Nat Methods. 2019;16(12):1289–96.
13. Hafemeister C, Satija R. Normalization and variance stabilization of
single-cell RNA-seq data using regularized negative binomial regression.
Genome Biol. 2019;20(1):296.
14. Chen A, Liao S, Cheng M, et al. Spatiotemporal transcriptomic atlas
of mouse organogenesis using DNA nanoball-patterned arrays. Cell.
2022;185(10):1777-1792.e21.
15. Chu Y, Dai E, Li Y, et al. Pancancer T-cell atlas links a cellular stress response
state to immunotherapy resistance. Nat Med. 2023;29(6):1550–62.
16. Yu G, Wang LG, Han Y, He QY. clusterProfiler: an R package for comparing
biological themes among gene clusters. OMICS. 2012;16(5):284–7.
17. Tang F, Li J, Qi L, et al. A pan-cancer single-cell panorama of human
natural killer cells. Cell. 2023;186(19):4235-4251.e20.
18. Yang Y, Chen X, Pan J, et al. Pan-cancer single-cell dissection reveals
phenotypically distinct B cell subtypes. Cell. 2024;187(17):4790-4811.e22.
19. Kang K, Wu Y, Han C, et al. Homologous recombination deficiency in
triple-negative breast cancer: Multi-scale transcriptomics reveals distinct
tumor microenvironments and limitations in predicting immunotherapy
response. Comput Biol Med. 2023;158: 106836.
20. Wang H, Liang Y, Liu Z, et al. POSTN+ cancer-associated fibroblasts
determine the efficacy of immunotherapy in hepatocellular carcinoma. J
Immunother Cancer. 2024;12(7): e008721.
21. Schubert M, Klinger B, Klünemann M, et al. Perturbation-response genes
reveal signaling footprints in cancer gene expression. Nat Commun.
2018;9(1):20.
22. Trapnell C, Cacchiarelli D, Grimsby J, et al. The dynamics and regulators
of cell fate decisions are revealed by pseudotemporal ordering of single
cells. Nat Biotechnol. 2014;32(4):381–6.
23. Patel AP, Tirosh I, Trombetta JJ, et al. Single-cell RNA-seq highlights
intratumoral heterogeneity in primary glioblastoma. Science.
2014;344(6190):1396–401.
24. Skinnider MA, Squair JW, Kathe C, et al. Cell type prioritization in single-
cell data. Nat Biotechnol. 2021;39(1):30–4.
25. Ren X, Zhong G, Zhang Q, et al. Reconstruction of cell spatial organization
from single-cell RNA sequencing data based on ligand-receptor medi-
ated self-assembly. Cell Res. 2020;30(9):763–78.
26. Lu H, Ping J, Zhou G, et al. CommPath: An R package for inference and
analysis of pathway-mediated cell-cell communication chain from single-
cell transcriptomics. Comput Struct Biotechnol J. 2022;26(20):5978–83.
27. Ma Y, Zhou X. Spatially informed cell-type deconvolution for spatial
transcriptomics. Nat Biotechnol. 2022;40(9):1349–59.
28. Xu Y, Wei Z, Feng M, et al. Tumour-infiltrated activated B cells suppress
liver metastasis of colorectal cancers. Cell Rep. 2022;40(9): 111295.
29. Chen Z, Zhang G, Ren X, et al. Cross-talk between Myeloid and B Cells
Shapes the Distinct Microenvironments of Primary and Secondary Liver
Cancer. Cancer Res. 2023;83(21):3544–61.
30. Takeuchi A, Itoh Y, Takumi A, et al. CRTAM confers late-stage activation
of CD8+ T cells to regulate retention within lymph node. J Immunol.
2009;183(7):4220–8.
31. Zhang L, Zhao C, Dai W, et al. Disruption of cholangiocyte-B cell crosstalk
by blocking the CXCL12-CXCR4 axis alleviates liver fibrosis. Cell Mol Life
Sci. 2023;80(12):379.
32. Meylan M, Petitprez F, Becht E, et al. Tertiary lymphoid structures generate
and propagate anti-tumor antibody-producing plasma cells in renal cell
cancer. Immunity. 2022;55(3):527-541.e5.
33. Patil NS, Nabet BY, Müller S, et al. Intratumoral plasma cells predict
outcomes to PD-L1 blockade in non-small cell lung cancer. Cancer Cell.
2022;40(3):289-300.e4.
34. Dyugay IA, Lukyanov DK, Turchaninova MA, et al. Accounting for B-cell
Behavior and Sampling Bias Predicts Anti-PD-L1 Response in Bladder
Cancer. Cancer Immunol Res. 2022;10(3):343–53.
35. Hu Y, Huang L, Zhao K, et al. CTHRC1 Is Associated With Immune Escape
and Poor Prognosis in Gastric Cancer. Anticancer Res. 2023;43(1):115–26.
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
Page 25 of 25
Dengetal. Molecular Cancer (2025) 24:28
36. Ramage JK, Valle JW, Nieveen van Dijkum EJM, et al. Colorectal Neu-
roendocrine Neoplasms: Areas of Unmet Need. Neuroendocrinology.
2019;108(1):45–53.
37. Zhang XB, Fan YB, Jing R, et al. Gastroenteropancreatic neuroendocrine
neoplasms: current development, challenges, and clinical perspectives.
Mil Med Res. 2024;11(1):35.
38. Antonia SJ, López-Martin JA, Bendell J, et al. Nivolumab alone and
nivolumab plus ipilimumab in recurrent small-cell lung cancer (Check-
Mate 032): a multicentre, open-label, phase 1/2 trial. Lancet Oncol.
2016;17(7):883–95.
39. Nghiem PT, Bhatia S, Lipson EJ, et al. PD-1 Blockade with Pembrolizumab
in Advanced Merkel-Cell Carcinoma. N Engl J Med. 2016;374(26):2542–52.
40. Albertelli M, Dotto A, Nista F, et al. Present and future of immunotherapy
in Neuroendocrine Tumors. Rev Endocr Metab Disord. 2021;22(3):615–36.
41. Lu M, Zhang P, Zhang Y, et al. Efficacy, Safety, and Biomarkers of
Toripalimab in Patients with Recurrent or Metastatic Neuroendo-
crine Neoplasms: A Multiple-Center Phase Ib Trial. Clin Cancer Res.
2020;26(10):2337–45.
42. Jiang X, Wang J, Deng X, et al. Role of the tumor microenvironment in
PD-L1/PD-1-mediated tumor immune escape. Mol Cancer. 2019;18(1):10.
43. Bhat AA, Nisar S, Singh M, et al. Cytokine- and chemokine-induced
inflammatory colorectal tumor microenvironment: Emerging avenue for
targeted therapy. Cancer Commun (Lond). 2022;42(8):689–715.
44. Zhou Y, Bian S, Zhou X, et al. Single-Cell Multiomics Sequencing Reveals
Prevalent Genomic Alterations in Tumour Stromal Cells of Human Colo-
rectal Cancer. Cancer Cell. 2020;38(6):818-828.e5.
45. Wang R, Song S, Qin J, et al. Evolution of immune and stromal cell states
and ecotypes during gastric adenocarcinoma progression. Cancer Cell.
2023;41(8):1407-1426.e9.
46. Nee K, Ma D, Nguyen QH, et al. Preneoplastic stromal cells promote
BRCA1-mediated breast tumorigenesis. Nat Genet. 2023;55(4):595–606.
47. Di X, Chen J, Li Y, et al. Crosstalk between fibroblasts and immunocytes
in fibrosis: From molecular mechanisms to clinical trials. Clin Transl Med.
2024;14(1): e1545.
48. Mao X, Xu J, Wang W, et al. Crosstalk between cancer-associated fibro-
blasts and immune cells in the tumor microenvironment: new findings
and future perspectives. Mol Cancer. 2021;20(1):131.
49. Wong SC, He CW, Chan CM, et al. Clinical significance of frizzled homolog
3 protein in colorectal cancer patients. PLoS ONE. 2013;8(11): e79481.
50. Smith AJ, Sompel KM, Elango A, et al. Non-Coding RNA and Frizzled
Receptors in Cancer. Front Mol Biosci. 2021;4(8): 712546.
51. Mo D, Jiang P, Yang Y, et al. A tRNA fragment, 5’-tiRNAVal, suppresses the
Wnt/β-catenin signaling pathway by targeting FZD3 in breast cancer.
Cancer Lett. 2019;10(457):60–73.
52. Li C, Nguyen V, Clark KN, et al. Down-regulation of FZD3 receptor sup-
presses growth and metastasis of human melanoma independently of
canonical WNT signaling. Proc Natl Acad Sci U S A. 2019;116(10):4548–57.
53. Kaur SP, Verma A, Lee HK, et al. Inhibition of glypican-1 expression
induces an activated fibroblast phenotype in a human bone marrow-
derived stromal cell-line. Sci Rep. 2021;11(1):9262.
54. Shah H, Hacker A, Langburt D, et al. Myocardial Infarction Induces Cardiac
Fibroblast Transformation within Injured and Noninjured Regions of the
Mouse Heart. J Proteome Res. 2021;20(5):2867–81.
55. Philips RL, Wang Y, Cheon H, et al. The JAK-STAT pathway at 30: Much
learned, much more to do. Cell. 2022;185(21):3857–76.
56. Patel SA, Nilsson MB, Le X, et al. Molecular Mechanisms and Future
Implications of VEGF/VEGFR in Cancer Therapy. Clin Cancer Res.
2023;29(1):30–9.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in pub-
lished maps and institutional affiliations.
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
1.
2.
3.
4.
5.
6.
Terms and Conditions
Springer Nature journal content, brought to you courtesy of Springer Nature Customer Service Center GmbH (“Springer Nature”).
Springer Nature supports a reasonable amount of sharing of research papers by authors, subscribers and authorised users (“Users”), for small-
scale personal, non-commercial use provided that all copyright, trade and service marks and other proprietary notices are maintained. By
accessing, sharing, receiving or otherwise using the Springer Nature journal content you agree to these terms of use (“Terms”). For these
purposes, Springer Nature considers academic use (by researchers and students) to be non-commercial.
These Terms are supplementary and will apply in addition to any applicable website terms and conditions, a relevant site licence or a personal
subscription. These Terms will prevail over any conflict or ambiguity with regards to the relevant terms, a site licence or a personal subscription
(to the extent of the conflict or ambiguity only). For Creative Commons-licensed articles, the terms of the Creative Commons license used will
apply.
We collect and use personal data to provide access to the Springer Nature journal content. We may also use these personal data internally within
ResearchGate and Springer Nature and as agreed share it, in an anonymised way, for purposes of tracking, analysis and reporting. We will not
otherwise disclose your personal data outside the ResearchGate or the Springer Nature group of companies unless we have your permission as
detailed in the Privacy Policy.
While Users may use the Springer Nature journal content for small scale, personal non-commercial use, it is important to note that Users may
not:
use such content for the purpose of providing other users with access on a regular or large scale basis or as a means to circumvent access
control;
use such content where to do so would be considered a criminal or statutory offence in any jurisdiction, or gives rise to civil liability, or is
otherwise unlawful;
falsely or misleadingly imply or suggest endorsement, approval , sponsorship, or association unless explicitly agreed to by Springer Nature in
writing;
use bots or other automated methods to access the content or redirect messages
override any security feature or exclusionary protocol; or
share the content in order to create substitute for Springer Nature products or services or a systematic database of Springer Nature journal
content.
In line with the restriction against commercial use, Springer Nature does not permit the creation of a product or service that creates revenue,
royalties, rent or income from our content or its inclusion as part of a paid for service or for other commercial gain. Springer Nature journal
content cannot be used for inter-library loans and librarians may not upload Springer Nature journal content on a large scale into their, or any
other, institutional repository.
These terms of use are reviewed regularly and may be amended at any time. Springer Nature is not obligated to publish any information or
content on this website and may remove it or features or functionality at our sole discretion, at any time with or without notice. Springer Nature
may revoke this licence to you at any time and remove access to any copies of the Springer Nature journal content which have been saved.
To the fullest extent permitted by law, Springer Nature makes no warranties, representations or guarantees to Users, either express or implied
with respect to the Springer nature journal content and all parties disclaim and waive any implied warranties or warranties imposed by law,
including merchantability or fitness for any particular purpose.
Please note that these rights do not automatically extend to content, data or other material published by Springer Nature that may be licensed
from third parties.
If you would like to use or distribute our Springer Nature journal content to a wider audience or on a regular basis or in any other manner not
expressly permitted by these Terms, please contact Springer Nature at
onlineservice@springernature.com
Available via license: CC BY-NC-ND 4.0
Content may be subject to copyright.