Abstract
Distal cholangiocarcinoma (dCCA) is a highly lethal malignancy that
accounts for approximately 40% of patients with primary
cholangiocarcinoma. Remarkable cellular heterogeneity and perineural
invasion (PNI) are two typical features of dCCA. Deciphering the
complex interplay between neoplastic and neural cells is crucial for
understanding the mechanisms propelling PNI-positive dCCA progression.
Herein, we conduct single-cell RNA sequencing on 24,715 cells from two
pairs of PNI-positive dCCA tumors and adjacent tissues, identifying
eight unique cell types. Malignant cells exhibit significant inter- and
intra-tumor heterogeneity. We delineate the compositional and
functional phenotypes of five Schwann cell (SC) subsets in PNI-positive
dCCA. Moreover, our analyses reveal two potential cell subtypes
critical to forming PNI: NEAT1^+ malignant cells characterized by
hypoxic propensity and GFAP^+ dedifferentiated SCs featuring
hypermetabolism. Further bioinformatics uncover extensive cellular
interactions between these two subpopulations. Functional experiments
confirm that lactate in the hypoxic tumor microenvironment can induce
GFAP-dedifferentiation in SCs, which promotes cancer cell invasion and
progression through upregulating HMGB1. Taken together, our findings
offer a thorough characterization of the transcriptional profile in
PNI-positive dCCA and unveil potential therapeutic targets for dCCA
PNI.
graphic file with name 41419_2025_7543_Figa_HTML.jpg
Subject terms: Cancer microenvironment
Introduction
Cholangiocarcinoma (CCA) represents the most prevalent malignancy of
the biliary system [[44]1]. Depending on its anatomical location, CCA
is typically categorized into intrahepatic CCA (iCCA), perihilar CCA
(pCCA), and distal CCA (dCCA) [[45]2, [46]3], with the latter localized
to the common bile duct below the cystic duct insertion [[47]2]. Over
nearly three decades, the diagnostic morbidity and mortality of dCCA
have continued to increase [[48]4]. Due to early asymptomatic or
nonspecific symptoms, many cases of dCCA are diagnosed in advanced
stages, leading to limited available therapeutic options and an
extremely poor prognosis. Surgical resection followed by adjuvant
therapy may improve survival outcomes in patients with dCCA, but the
high recurrence rate after dCCA surgery remains a challenge in clinical
management [[49]5]. Perineural invasion (PNI), refers to the
characteristic biological process by which cancer cells invade nerves
and spread along the perineurium [[50]6]. In a recent cohort study, PNI
was a common pathological phenomenon, present in 81.8% of dCCA cases
[[51]7]. Importantly, PNI has been recognized as a significant
prognostic indicator affecting patients with resectable dCCA [[52]8,
[53]9].
Peripheral nerves partake in the constitution of a complicated tumor
ecosystem comprising diverse cell populations, including Schwann cells
(SCs). Physical contact between malignant cells and SCs has been found
to promote directed movement and invasion of cancer cells [[54]10].
Fuji-Nishimura et al. demonstrated that SCs facilitate colonization of
pancreatic cancer in nerves by activating the epithelial-mesenchymal
transition (EMT) pathway in tumor cells [[55]11]. Recent research has
illuminated that SCs could contribute to tumor progression by
transitioning to a dedifferentiated state, analogous to their response
to neurotrauma [[56]10]. This reprogramming of SCs leads to the
re-expression of glial fibrillary acidic protein (GFAP), neural cell
adhesion molecule 1 (NCAM1), and L1 cell adhesion molecule (L1CAM),
which can drive the development of PNI [[57]10, [58]12, [59]13].
Presently, the initiator and tumor-promoting effect of dedifferentiated
SCs (dSCs) in PNI-positive dCCA remains unclear. Consequently, a
thorough comprehension of the cellular and molecular mechanism
underlying neuromodulation of cancer progression is crucial for
developing strategies for inhibiting tumor progression [[60]6].
Herein, we employed the powerful technique of single-cell RNA
sequencing (scRNA-seq) to profile PNI-positive dCCA and adjacent
tissues, and identified two PNI-associated cellular components: NEAT1^+
malignant cells and GFAP^+ dSCs. We provided hitherto undocumented
evidence that lactate in hypoxic tumor microenvironment (TME) could
initiate GFAP-dedifferentiation of SCs, and the latter enhanced dCCA
progression through upregulating high mobility group box 1 (HMGB1).
Taken together, our findings offer an exhaustive transcriptomic
overview and elucidate the intercellular interaction between malignant
cells and SCs in PNI-positive dCCA, revealing potential therapeutic
vulnerabilities in dCCA PNI.
Results
Single-cell transcriptomic profiling uncovered the spectrum of cell
populations in human PNI-positive dCCAs
To comprehensively understand the tumor ecosystem in dCCA with PNI, we
conducted scRNA-seq on tumor and paired adjacent non-neoplastic tissues
from two untreated PNI-positive dCCAs (Fig. [61]1A). Detailed
clinicopathological features of the study population are listed in
Table [62]S1. Following quality control and filtering, single-cell
transcriptome profiles were obtained for 24,715 cells. Eight primary
cell types were determined informed by established marker genes,
including epithelial cells (2696, 10.9%), myeloid cells (4212, 17.0%),
lymphoid cells (13,161, 53.3%), endothelial cells (1946, 7.9%), SCs
(472, 1.9%), fibroblast (1919, 45.6%), MKI67^+ cells (227, 0.9%), and
smooth muscle cells (SMCs, 82, 0.3%, Fig. [63]1B). Subsequently, we
extracted all epithelial cells and identified 13 subclusters through
reclustering analysis. Clusters 2 and 4 were considered normal
epithelium and served as a normal reference for copy number variation
(CNV) analysis due to their predominant distribution in adjacent
noncancerous tissues (Fig. [64]1C, Fig. [65]S1A). A total of 1203
malignant cells expressing high levels of KRT19 were inferred and
further reclustered (Fig. [66]S1B). Figure [67]1D illustrates the
original 21 cell clusters for all cells. Consistent with previous dCCA
studies [[68]14, [69]15], non-malignant cells (excluding SMCs)
exhibited inter- and intratumoral heterogeneity across different
tissues. For instance, endothelial cells, epithelial cells, myeloid
cells, and fibroblasts were heavily infiltrated in tumors, whereas
lymphoid cells and SCs were predominantly found in adjacent biliary
ductal tissues (Fig. [70]1E). Moreover, to validate our findings, we
used CIBERSORTx [[71]16] to deconvolute bulk RNA-seq data from a
broader cohort of CCA and normal samples. The relative abundance of
endothelial cells, fibroblasts, and SCs in our samples conformed with
estimates from the TCGA-CHOL dataset. However, epithelial cells and
immune cells displayed discrepant patterns (Fig. [72]S1C).
Fig. 1. scRNA-seq profiling of 2 dCCAs.
[73]Fig. 1
[74]Open in a new tab
A Schematic representation of the experimental strategy. Part of the
pictures were adapted from Servier Medical Art
([75]http://smart.servier.com). B Heatmap showing the expression of
marker genes in the indicated cell types. C Chromosomal landscape of
inferred large-scale CNVs in normal epithelial cells (top) and
potentially malignant cells (bottom) from 2 dCCA samples. Rows
represent individual cells and columns represent chromosomal positions.
Amplifications (red) or deletions (blue) were inferred by averaging
expression over 100-gene stretches on the respective chromosomes. D
Uniform manifold approximation and projection (UMAP) plot of malignant
and non-malignant cells from 2 dCCA samples. E Boxplot showing the
fraction of non-malignant cells in tumor and peri-tumor tissues.
Highly heterogeneous hypoxic patterns of malignant cells and their
contribution to the PNI-positive dCCA microenvironment
To characterize the tumor cell landscape in PNI-positive dCCA,
malignant cells were subsequently clustered and divided into three
primary subclusters (Fig. [76]2A, B). In alignment with previous
findings in dCCA [[77]14], malignant cells exhibited significant intra-
and inter-tumor heterogeneity (Fig. [78]2A). The distinctive expression
patterns within these three subpopulations are illustrated in Fig.
[79]2B. Cluster 0 was enriched for cells that highly expressed genes in
the S100 family, such as S100A4, S100A10, and S100A11. S100 protein
family members have been commonly observed to be dysregulated in
various tumors, including iCCA, and are critically implicated in
carcinogenesis and cancer progression [[80]17–[81]19]. Cluster 1 was
characterized by a prominent upregulation of NEAT1 and MALAT1. These
two adjacent long non-coding RNA genes have been extensively documented
to be involved not only in activating multiple oncogenic mechanisms but
also in conferring resistance to chemotherapeutics [[82]20, [83]21].
TOP2A and TK1, both of which were substantially expressed in cluster 2,
have been previously recognized as proliferative markers in many
studies [[84]22–[85]24]. Hypoxia is a ubiquitous property of most solid
cancers and is strongly linked to tumor metastasis and invasion
[[86]25]. We subsequently visualized the hypoxia statuses of malignant
cells using the cellular hypoxia predicting framework (CHPF) [[87]26].
Among these, most hypoxic cells were concentrated in cluster 1
(NEAT1^+) malignant cells, with fewer found in cluster 0 (S100A4^+) and
cluster 2 (TOP2A^+) (Fig. [88]2C). Additionally, to explore the
influence of hypoxia on the evolutionary dynamics of malignant cells in
PNI-positive dCCA, Monocle2 and CytoTRACE were employed to perform
unsupervised cell trajectory analysis, both of which revealed a similar
differentiation pathway of malignant cells originating from hypoxic
cells (Fig. [89]2D, Fig. [90]S2A), consistent with the conclusion drawn
by Zhang et al. in glioblastoma [[91]26]. Three cell states (S1–S3)
were defined for pseudotime trajectory analysis based on Monocle2 (Fig.
[92]2D). In terms of cellular status, hypoxic cells were primarily
confined to S1 and S2 at the initial stage of differentiation, whereas
normoxic cells were predominantly concentrated in S3. Regarding cell
clusters, NEAT1^+ malignant cells (cluster 1) dominated the S1 state,
appearing at the earliest stage of pseudotime and exhibiting
significantly higher stemness scores. We postulated that the high
stemness of cluster 1 might be related to its deduced hypoxic state,
according to previous studies [[93]27–[94]29]. Correspondingly,
S100A4^+ malignant cells (cluster 0) constituted the primary subcluster
of the S3 state and were exclusively observed in the final stage of
cell differentiation. Notably, TOP2A^+ malignant cells (cluster 2)
spanned across both S2 and S3 states, suggesting the presence of two
distinct cell substrates within cluster 2 (Fig. [95]2D). Taken
together, these findings indicated an orchestrated differentiation
process of dCCA cells during PNI. PEAK1, a novel human pseudokinase,
has recently been implicated in cancer pathogenesis [[96]30]. We
observed that PEAK1^+ malignant cells were positioned at the beginning
of the major branch and aligned well with cluster 1. Similarly,
metastasis scores and hepatic vascular invasion scores were
predominantly observed at the onset of differentiation. These findings
suggested that cluster 1 might represent a key cell type with high
invasiveness in PNI-positive dCCA.
Fig. 2. Transcriptional signatures and hypoxia heterogeneity of malignant
cells.
[97]Fig. 2
[98]Open in a new tab
A UMAP plot of three malignant subtypes. Pie charts for each subtype
showing the contributing percentage of cells from each patient. B Heat
map showing the top differentially expressed genes (DEGs) in each
malignant subtype. C UMAP plot of malignant cells colored by hypoxia
status. D Semisupervised pseudotime trajectory of malignant subtypes
inferred by Monocle2. Trajectory is colored by pseudotime (top left),
cell subtypes (top middle), hypoxia status (top right), cell states
(left), CNV levels (middle), the expression dynamics of a selected
marker gene PEAK1 (right), stemness signature scores (bottom left),
metastasis signature scores (bottom middle), signature scores
calculated based on the HO_LIVER_CANCER_VASCULAR_INVASION geneset
(bottom right). E Malignant cells were grouped into different
categories based on the CNV score. Ridgeline Plots show the
distribution of CNV scores across different cell clusters. The red
dashed line indicates the threshold value. F CNV inferred by scRNA-seq
data in patient P1. G The percentage of hypoxic cells is positively
correlated with the proportion of cells with high levels of CNVs. H
Heatmap showing the scaled expression of DEGs across pseudotime
trajectory in (D). Bar plots at the top of the heatmap are scale
diagrams of different cell states, hypoxia status, CNV levels, and cell
subtypes during pseudotime differentiation trajectory. I Association of
relative cell abundance (estimated by CIBERSORTx) and patient survival
using the TCGA-CHOL cohort (n = 36) by COX regression analysis. J
Kaplan–Meier curves of TCGA-CHOL patients (n = 36) showing the survival
rates grouped by the cell abundance in malignant cell cluster 1. The P
value is calculated with two-sided log-rank test. K Violin plots
displaying the cell abundance in malignant cell cluster 1 in non-PNI
and PNI groups. non-PNI, n = 26 samples; PNI, n = 7 samples. The
central mark indicates the median, and the bottom and top edges of the
box indicate the first and third quartiles, respectively. The top and
bottom whiskers extend the boxes to a maximum of 1.5 times the
interquartile range. ns, not significant.
To delve deeper into the relationship between hypoxia and the
aggressive phenotype of PNI-positive dCCA, hypoxia-related signature
genes from the CancerSEA database [[99]31] and several hallmark gene
sets, including EMT, IL2/STAT5, PI3K/AKT/mTOR, and KRAS signaling from
MSigDB, were manually curated (Table [100]S2). Given gene set variation
analysis (GSVA) to determine the activity score of each malignant cell,
the association between hypoxia and tumor invasion activities in
PNI-positive dCCA was evaluated. The results showed that invasion score
was significantly positively correlated with hypoxia (Fig. [101]S2B).
Utilizing the previously inferred single-cell CNV spectrum, we observed
that clusters 1 and 2 exhibited higher CNV levels than cluster 0 (Fig.
[102]2E). In addition, the extent of CNV accumulation was correlated
with the hypoxic status of cells. As exemplified by malignant cells
derived from patient P1, hypoxic malignant cells displayed
significantly higher CNV levels than normoxic malignant cells,
indicative of a more malignant trait. In this respect, high-frequency
CNV events were enriched in certain chromosomes, such as chr6, chr12,
and chr15 (Fig. [103]2F). We categorized all malignant cells into low
and high groups based on CNV levels (Fig. [104]2D, E). The percentage
of hypoxic cells within each malignant cell cluster demonstrated a
positive correlation with the percentage of CNV^high cells in that
cluster (Fig. [105]2G). In addition, pathway enrichment analysis using
GSVA revealed that MTORC1 signaling, MYC targets, E2F targets, and EMT
pathways were enriched in the CNV-high group (Fig. [106]S2C). Overall,
hypoxia and high CNV levels might be essential for preserving the
malignant characteristics of cluster 1. To summarize the transcriptomic
features of malignant cells, we integrated meta-information regarding
cell cluster, hypoxic state, CNV status, and predicted trajectories.
The cellular developmental process was divided into two distinct phases
based on dynamic gene expression patterns (Fig. [107]2H).
Correspondingly, the initial phase primarily comprised states S1 and
S2, and there was a propensity for cluster 1 cells to transition to
cluster 2 during this stage. This alteration was accompanied by
downregulation of the hypoxia-induced gene VEGFA and the oncogenic
driver, AKT, as well as diminished signaling pathways associated with
hypoxia response and epithelial cell migration. In contrast, cluster 1
cells in the second phase exhibited a greater propensity to transform
into cluster 0 and subsequently progress toward the S3 state,
characterized by heightened expression of RPS15 and a shift in energy
metabolism towards aerobic respiration (Fig. [108]2H).
To investigate the clinical implication of the malignant cell subtypes
identified in our study, we estimated the proportion of epithelial cell
subpopulations (including normal epithelial cells) within patient
samples from the TCGA-CHOL cohort using CIBERSORTx (Table [109]S3).
Only the increased abundance of cluster 1 malignant cells showed a
significant correlation with decreased overall survival (OS; Fig.
[110]2I, J). We subsequently obtained similar results using
[111]GSE107943 as a validation dataset (Fig. [112]S2D, Table [113]S4).
Furthermore, utilizing information on samples from the TCGA cohort
containing patient PNI status, we discovered that cluster 1 malignant
cells were significantly more abundant in CCA with PNI than CCA without
PNI (Fig. [114]2K). These findings indicated that cluster 1 (NEAT1^+)
malignant cells, characterized by hypoxia propensity and higher levels
of CNV, may be associated with PNI in dCCA.
dSCs play a significant role in dCCA PNI
SCs have been firmly established as a novel cell type within the TME,
playing a specific and cancer-promoting role in PNI [[115]32]. We
focused our analysis on SCs in dCCA, performing unsupervised clustering
on 472 cells, and identifying five distinct subclusters (Fig. [116]3A).
Utilizing a marker gene list curated from the Tabula Sapiens portal
[[117]33] and previous literature by Kastriti et al. [[118]34], we
observed that clusters 0 and 1 exhibited overexpression of myelinating
SC (mSC) markers like EGR2, MPZ, and PMP22. Cluster 2 displayed
upregulation of well-defined non-myelinating SC (nmSC) markers such as
IGFBP5, TAGLN2, TPM1, and A2M. Notably, cluster 4 preferentially
expressed genes indicative of SC precursors (SCPs): CD69, BTG1, CD52,
CYBA, and LTB (Fig. [119]3A, Fig. [120]S3A). Among these clusters,
MPZ^+ mSCs (cluster 0), PMP22^+ mSCs (cluster 1), and SCPs (cluster 4)
were predominantly located in cancer-adjacent tissues. Conversely,
nmSCs (cluster 2) and cluster 3 had a greater proportion of cells
distributed within cancer tissues (Fig. [121]3B). Figure [122]3C
illustrates the unique transcriptomic signatures of all SC subsets
identified in dCCA.
Fig. 3. Transcription profiling of SCs in the TME of PNI-positive dCCA
tissues.
[123]Fig. 3
[124]Open in a new tab
A UMAP showing the five subtypes of SCs, colored by subclusters. B
Distribution of SCs in different sample groups on the UMAP. Pie chart
showing the proportion of two sample groups in each SC subcluster. C
Violin plots (left) displaying the representative expression pattern
across different subtypes of SCs. Dot plot (right) showing the
expression of the top six subtype-specific gene markers in each
subtype. D Semisupervised pseudotime trajectory of SC subtypes by
Monocle2. Trajectory is colored by pseudotime (top left), cell states
(top middle), cell clusters (top right), sample groups (bottom left),
and expression dynamics of two marker genes GFAP (bottom middle) and
NCAM1 (bottom right). E Dot plot illustrating the expression patterns
of selected dSC gene markers in each SC subtype. F Dot plot showing the
metabolic activity analysis of all SC subclusters by scMetabolism. The
circle size and color darkness both represent the scaled metabolic
score. G Association of relative cell abundance (estimated by
CIBERSORTx) and patient survival using the TCGA-CHOL cohort (n = 36) by
COX regression analysis. H Violin plots displaying the cell abundance
in GFAP^+ dSC in non-PNI and PNI groups. non-PNI, n = 26 samples; PNI,
n = 7 samples. The central mark indicates the median, and the bottom
and top edges of the box indicate the first and third quartiles,
respectively. The top and bottom whiskers extend the boxes to a maximum
of 1.5 times the interquartile range. I Representative images (top
left) of immunohistochemistry (IHC) expression of GFAP and NCAM1 in
patients from the Zhengzhou-dCCA cohort (n = 22). Representative images
(bottom left) of H&E staining assays of PNI and non-PNI patients from
the Zhengzhou-dCCA cohort. The experiment was repeated once with
similar results. Nerves are highlighted with dotted lines and tumor
cells with arrows. Scale bars, 100 μm. Bar plot (right) showing the
positive proportion of IHC staining for GFAP and NCAM1 from PNI and
non-PNI patients from the Zhengzhou-dCCA cohort.
To investigate the developmental pathways and potential roles of these
distinct SC subclusters in dCCA with PNI, we first employed CytoTRACE
to estimate the differentiation degree of each subcluster. As expected,
the SCP cluster, representing multipotent embryonic progenitors for
many neural cells [[125]35, [126]36], possessed the highest
differentiation score (Fig. [127]S3B). Then, we reconstructed the SCs
into a pseudotime trajectory using Monocle2, designating the SCPs as
the starting point. Five distinct cell states (S1–S5) and a primary
trajectory route were identified (Fig. [128]3D). We observed that
cluster 3 was positioned in close proximity to the differentiation
starting site and characteristically expressed the dedifferentiation
markers GFAP and NCAM1 (Fig. [129]3D). Furthermore, we found that
cluster 3 also upregulated the myelin-related gene SOX2 and the
immature genes NGFR and L1CAM (Fig. [130]3E), aligning well with the
reprogramming process of dSCs described by Jessen et al. [[131]37].
Therefore, we classified cluster 3 as dSC. Previous studies have
indicated that both mSCs and nmSCs can contribute to cancer progression
by transitioning to the dSC phenotype characteristic of repair SCs in
cancer [[132]32, [133]37]. Our analysis further revealed that PMP22^+
mSCs and certain nmSCs (specifically cluster S4) initially transitioned
to cluster 0 (MPZ^+ mSC) during the dedifferentiation process. Notably,
this transition occurred with little up-regulation of immature genes,
while JUN expression increased but SOX2 remained relatively unchanged.
Indeed, both JUN and SOX2 are myelin suppressor genes. In contrast,
SOX2 expression became prominent during the dedifferentiation phase
(Fig. [134]3D, E). These findings suggest that the abandonment of
myelin differentiation in dSCs may precede the activation of the
immature phenotype, and different negative regulators of myelination
seem to act asynchronously. Overall, for PNI-positive dCCA, most SCs
transition from SCPs to GFAP^+ dSCs, traversing an intermediate state
(Cluster 0, MPZ^+ mSC). Ultimately, they may develop into PMP22^+ mSCs
or nmSCs. Our analysis provides a comprehensive ecological map and
trajectory evolution of SCs in PNI-positive dCCA.
Gene ontology (GO) analysis revealed that MPZ^+ mSCs were significantly
enriched in neuron apoptosis processes, neuron death, and tumor
necrosis factor-mediated signaling pathways, possibly reflecting the
damage response induced by cancer cell invasion (Fig. [135]S3C).
Conversely, enriched GO terms for PMP22^+ mSCs were associated with
neural support and regeneration, including axonogenesis, axon
development, and regulation of synapse maturation (Fig. [136]S3C).
nmSCs were characterized by a high level of extracellular matrix
similar to fibroblasts, while GO terms of GFAP^+ dSCs were enriched in
cholesterol binding, lipid transfer activity, and phosphatidylcholine
binding, indicating their higher metabolic properties (Fig. [137]S3C).
Finally, GO analysis of SCPs revealed their enrichment in pathways such
as activation of the immune response, regulation of T cell activation,
and neutrophil migration, suggesting a potential role in immune
regulation (Fig. [138]S3C). To further elucidate the metabolic
landscape of SCs in PNI-positive dCCA, scMetabolism was employed
[[139]38] to systematically quantify metabolic activities at
single-cell resolution. We computed metabolic pathway activity scores
for all 63 metabolic pathways annotated in scMetabolism and found that
GFAP^+ dSCs exhibited higher metabolic activity (Fig. [140]3F, Fig.
[141]S3D). Among these pathways, pyruvate metabolism, lactate
metabolism, glycerolipid metabolism, and fatty acid biosynthesis were
markedly activated in GFAP^+ dSCs (Fig. [142]3F).
To explore the influence of each SC cluster on dCCA prognosis,
CIBERSORTx was applied to determine the percentage of diverse SC types
across the TCGA-CHOL samples (Table [143]S5). High infiltration of
GFAP^+ dSCs was associated with an inferior prognosis (Fig. [144]3G,
Fig. [145]S3E). Similar results were obtained in the [146]GSE107943
validation cohort (Fig. [147]S3F, Table [148]S6). To study the
contribution of GFAP^+ dSCs to the occurrence of PNI in dCCA, the
TCGA-CHOL samples were sorted into PNI and non-PNI groups founded on
the presence or absence of concomitant PNI. We observed that the PNI
group displayed significantly higher infiltration of GFAP^+ dSCs (Fig.
[149]3H). Additionally, 22 dCCA patients from the First Affiliated
Hospital of Zhengzhou University were enrolled in our internal cohort
(Zhengzhou-dCCA cohort). Hematoxylin and eosin (H&E) staining confirmed
that all pathological sources were tumor tissues (Fig. [150]S4A).
Immunohistochemistry (IHC) analysis demonstrated that the positive
rates of GFAP and NCAM1 proteins in the neural tissue of PNI samples
were higher than those in non-PNI samples (Fig. [151]3I, Table
[152]S7). Collectively, these data suggest that GFAP^+ dSCs possess
high metabolic characteristics and play crucial roles in the
PNI-positive microenvironment.
Interactome landscape across NEAT1^+ malignant cells and GFAP^+ dSCs in the
PNI-related dCCA microenvironment
To elucidate the crosstalk between NEAT1^+ malignant cells and GFAP^+
dSCs within the TME during PNI progression, we investigated
intercellular communication by simulating ligand-receptor interactions
using CellChat. A total of 116 pairs of interactions were identified
across the four cell types we classified. Notably, NEAT1^+ malignant
cells and GFAP^+ dSCs exhibited the highest number of interactions
(Fig. [153]4A). A similar pattern was observed in terms of the strength
of intercellular interactions (Fig. [154]S5A). These results underscore
the critical roles of NEAT1^+ malignant cells and GFAP^+ dSCs in
PNI-positive dCCA. Subsequently, we utilized CellChat’s pattern
recognition to identify major secretory signaling events of various
cell types (Fig. [155]4B). When NEAT1^+ malignant cells served as the
signal source and GFAP^+ dSCs as the signal input, the CDF15-TGFBR2
interaction exhibited the highest interaction score (Fig. [156]4B).
Previous studies have demonstrated that inactivation of the TGFBR2 gene
leads to uneven and severely underdeveloped dSC invasion in mice (in
vivo), hindering their involvement in the bridge regeneration process
after nerve injury [[157]39]. We also noted that the CDF15 gene was
predominantly expressed in NEAT1^+ malignant cells, while TGFBR2 was
generally distributed across all SC types (Fig. [158]4C). Conversely,
when GFAP^+ dSCs sent ligands to NEAT1^+ malignant cells, the primary
interaction occurred through the BTC-EGFR pathway. The role of EGFR in
cancer progression and as a therapeutic target in various human
malignancies, including cholangiocarcinoma, lung cancer, colon cancer,
and breast cancer, has been well-established [[159]40–[160]43].
Analyzing the receptor-ligand expression distribution, we found that
BTC was almost exclusively expressed in GFAP^+ dSCs, while EGFR was
predominantly expressed by NEAT1^+ malignant cells. Therefore, the
BTC-EGFR interaction pair might represent a characteristic mode of
communication between GFAP^+ dSCs and NEAT1^+ malignant cells (Fig.
[161]4B-D).
Fig. 4. Cell-cell communication between malignant cells and SCs.
[162]Fig. 4
[163]Open in a new tab
A Cell-cell interaction network (top) of NEAT1^+ malignant cells, other
malignant cells, GFAP^+ dSCs, and other SCs. The node size represents
the number of interactions. The width of the edge represents the number
of significant ligand–receptor interactions in two cell types. Bar plot
(bottom) presenting the numbers of putative ligand-receptor pairs
between malignant cells and SCs. B Bubble heatmap showing interaction
strength for different ligand-receptor pairs. Dot size indicates the P
value generated by the permutation test and dot color represents
communication probabilities. Empty space indicates that the
communication probability is zero. C, D UMAP plot showing expression
levels of GDF15-TGFBR2 (C) and BTC-EGFR (D) ligand–receptor pairs in
specific cell types.
To investigate the significance of the NEAT1^+ malignant cell-GFAP^+
dSC interaction within the TME, we utilized SCENIC [[164]44] to
decipher the gene regulatory network (GRN) of these cell types. The GRN
differed among the subtypes of both malignant cells and SCs (Fig.
[165]S5B, C). We identified four key genes in the GRN of NEAT1^+
malignant cells: SREBF2, ATF3, RFX2, and JUN (Fig. [166]S5B). These
genes have previously been shown to be upregulated in damaged neurons
and regulate oxidative stress during the dedifferentiation of
neighboring SCs [[167]45, [168]46]. JUN is known to control mSC
dedifferentiation and the activation of repair programs [[169]47].
Conversely, multiple oncogenic transcription factors (TFs), including
ETS1, EP300, SMAD4, and ELK4, were upregulated in GFAP^+ dSCs (Fig.
[170]S5C). Jin et al. reported that tumor-derived extracellular
vesicles promote renal cell carcinoma invasion and metastasis by
transferring MALAT1 facilitating the binding of ETS1 and the TFCP2L1
promoter [[171]48]. Interestingly, MALAT1 is one of the genes that
characterize NEAT1^+ malignant cells. In conclusion, our data highlight
the close communication between NEAT1^+ malignant cells and GFAP^+ dSCs
within the PNI-associated dCCA microenvironment and identify potential
TF candidates for further investigation.
Hypoxia induces lactate secretion from cancer cells and further promotes SC
dedifferentiation
Previous research demonstrated that pancreatic cancer cell supernatants
under hypoxic conditions can induce GFAP activation in human SCs
[[172]49]. Of note, hypoxia is also a predicted hallmark of NEAT1^+
malignant cells. To investigate the mechanism of SC dedifferentiation
induced by hypoxic cells, we initially cultured CCLP1 and HUCCT1 cell
lines under hypoxic conditions in vitro to simulate the in vivo hypoxic
TME. After a 48-h incubation under either normoxic or hypoxic
conditions, HIF-1α levels were detected via western blot analysis to
assess the successful induction of hypoxic stress in the cancer cells.
The results indicated a significant enhancement of HIF-1α expression
under hypoxic conditions (Fig. [173]5A), confirming the effectiveness
of our hypoxia modeling. Subsequently, we stimulated ipNF95.6 (a human
SC line) with the modeled CCA cell supernatants to evaluate the
activation of SCs by the hypoxic microenvironment of cancer cells. A
significant increase in GFAP protein expression was observed when
ipNF95.6 cells were exposed to the supernatant of the hypoxia group
(Fig. [174]5B). Given that cancer cells consume substantial amounts of
oxygen and nutrients, secreting excess lactate [[175]50], and the high
lactate metabolic activity of dSCs described above, we sought to
determine whether SC dedifferentiation was related to lactate within
the hypoxic TME. We first measured lactate levels in the supernatant of
hypoxic cancer cells. Our findings revealed a significant elevation of
lactate levels within the supernatant of hypoxic CCA cells (Fig.
[176]5C). Similarly, we measured lactate levels in 22 dCCA tissues from
the Zhengzhou-dCCA cohort, which were significantly higher in the GFAP
protein-positive nerve group compared to the protein-negative group
(Fig. [177]5D). For NCAM1 protein, there was a trend towards higher
lactate content in the NCAM1 protein-positive nerve group, although
these results were not statistically significant (Fig. [178]5D).
Furthermore, we categorized all SCs from our scRNA-seq data into
hypermetabolism and hypometabolism groups according to the median
lactate metabolic activity score. We found that the expression of
multiple dedifferentiation-related SC markers, including L1CAM, JUN,
NCAM1, GFAP, and NGFR, was increased in the lactate hypermetabolism
group (Fig. [179]5E). To examine the impact of lactate on SC
dedifferentiation, we conducted a series of experiments. Referencing a
previous study [[180]51], we established a gradient lactate
concentration (0, 10, 20, 40, 80, and 160 mM) to determine the optimal
lactate concentration. SCs exposed to different lactate levels were
cultured for 8 h, and their viability was assessed. The results
indicated a dramatic decrease in SC viability at a lactate
concentration of 20 mM. Consequently, we selected a lactate
concentration of 10 mM for subsequent experiments (Fig. [181]5F). Next,
we analyzed multiple representative dedifferentiation-related genes,
among which the mRNA and protein levels of NCMA1, GFAP, and SOX2 were
noticeably upregulated in lactate-treated SCs (Fig. [182]5G-I).
Additionally, we obtained cross-species validation in RSC96 (a rat SC
line, Fig. [183]5I), indicating that the evolutionary process of
lactate-induced dedifferentiation in SCs might be conserved.
Collectively, these findings indicate that lactate produced by hypoxic
cancer cells promotes the dedifferentiation of SCs.
Fig. 5. Hypoxia condition induced lactate secretion of CCA cells, further
facilitated SC dedifferentiation.
[184]Fig. 5
[185]Open in a new tab
A The protein levels of HIF-1α in CCLP1 and HUCCT1 cells under normoxia
and hypoxia conditions. B The protein levels of GFAP in ipNF95.6 cells
receiving supernatants from CCA cells under normoxia and hypoxia
conditions. C Detection of lactate in the supernatants of CCLP1 and
HUCCT1 cells by lactate assay kit. D Detection of lactate in positive
and negative nerve tissues for SC dedifferentiation markers (GFAP and
NCAM1) in the Zhengzhou-dCCA cohort by lactate assay kit. E Expression
levels of 10 dedifferentiation-related markers in SCs with high and low
lactate metabolism. F The cytotoxic activity of lactate was measured
using the MTT cell viability assay in ipNF95.6 cells. G Detection of
mRNA expression levels of seven dedifferentiation-related markers in
ipNF95.6 cells by RT‑qPCR. H, I Western blotting assays detecting
protein levels of SC dedifferentiation members in ipNF95.6 (H) and
RSC96 cells (I). *P < 0.05, **P < 0.01, ***P < 0.001.
Cancer cell-derived lactate upregulates HMGB1 in SCs, which further promotes
the carcinogenic behavior of CCA cells
HMGB1 was initially reported to be released from
lipopolysaccharide-stimulated macrophages and to function as a
pro-inflammatory factor in sepsis [[186]52]. More recent studies have
demonstrated that stromal cells, such as tumor-associated macrophages,
upregulate intracellular HMGB1 expression upon lactate stimulation,
thereby promoting cancer progression [[187]53, [188]54]. Interestingly,
our single-cell data revealed that HMGB1 was generally upregulated in
GFAP^+ dSCs, and the percentage of GFAP^+ cells within each SC
subcluster correlated positively with the percentage of HMGB1^high SCs
in that cluster. (Fig. [189]6A, Fig. [190]S6A). Furthermore, a
protein-protein interaction (PPI) network was constructed utilizing the
STRING database v12.0, linking HMGB1 with 14 SC
dedifferentiation-related genes. The PPI network indicated an
interaction between HMGB1 and dSC markers, such as GFAP and JUN (Fig.
[191]S6B). To further examine the relationship between SC
dedifferentiation and HMGB1 expression, we performed IHC in the
Zhengzhou-dCCA cohort (Fig. [192]S6C). The images revealed higher HMGB1
IHC scores in the neural tissues of GFAP and NCAM1 protein-positive
groups were higher than those of protein-negative groups, although the
latter showed no statistical significance (Fig. [193]6B, Table
[194]S7). Additionally, we found that HMGB1 protein was significantly
upregulated after stimulation of ipNF95.6 cells with cancer cell
supernatants after hypoxia incubation (Fig. [195]6C). To determine if
HMGB1 expression in SCs was similarly linked to lactate secreted in the
hypoxic TME, immunofluorescence (IF) experiments demonstrated an
increase in the cytoplasmic level of HMGB1 in lactate-treated ipNF95.6
cells (Fig. [196]6D). To investigate the role of HMGB1 in SC
dedifferentiation and its impact on tumor progression, we introduced
glycyrrhizin (1 nM) to inhibit HMGB1 expression in subsequent protein
immunoblotting and IF experiments [[197]55]. Our findings revealed a
significant elevation of HMGB1 protein levels in lactate-stimulated
SCs, which was effectively inhibited by glycyrrhizin (Fig. [198]6E,
Fig. [199]S6D). Next, we sought to understand whether SCs stimulated
with lactate promoted tumor progression through HMGB1. Co-culture
experiments with lactate-induced SCs demonstrated accelerated cell
migration and invasion in both CCA cell lines (Fig. [200]6F, G).
Nevertheless, glycyrrhizin reversed the lactate-induced effect (Fig.
[201]6F, G). To further evaluate the functional role of lactate-treated
dSCs in CCA progression, we conducted in vivo experiments using
xenograft mice. Mice injected with a mixture of lactate-stimulated SCs
and CCLP1 cells exhibited larger tumor volumes. Notably, glycyrrhizin
attenuated the tumor growth-promoting effect of lactate-stimulated SCs
through HMGB1 inhibition (Fig. [202]6H, I). Considering that the
nuclear protein HMGB1 is released in response to diverse stimuli,
including lactate [[203]56, [204]57], we focused on the expression
level of HMGB1 within tumor cells after co-culture with SCs.
Lactate-treated SCs elevated the level of HMGB1 within cancer cells,
while glycyrrhizin inhibited this elevation (Fig. [205]S6E).
Collectively, these results suggest that SCs enhance the invasion and
migration of cancer cells through lactate-induced upregulation of
HMGB1.
Fig. 6. Cancer cell-derived lactate upregulated HMGB1 in SCs and HMGB1
further promoted carcinogenic behaviors in CCA cells.
[206]Fig. 6
[207]Open in a new tab
A The percentage of GFAP-positive cells is positively correlated with
the proportion of cells with high levels of HMGB1. B The correlation
between IHC expression of SC dedifferentiation markers (GFAP and NCAM1)
and IHC scores of HMGB1 in the Zhengzhou-dCCA cohort. C The protein
level of HMGB1 in ipNF95.6 cells receiving supernatants from cancer
cells under normoxia and hypoxia conditions. D The observation of HMGB1
protein levels in ipNF95.6 cells using immunofluorescence. E The
protein levels of HMGB1 in CCLP1 and HUCCT1 cells. F, G Wound healing
assay (F) and trans-well invasion assay (G) were performed respectively
to assess the mobility and invasion of cancer cells. H The xenograft
tumor model was established with a mixture of CCLP1 cells and ipNF95.6
cells. The arrows indicate the subcutaneous tumor. I Tumor volumes were
documented every 3 days. *P < 0.05, **P < 0.01, ***P < 0.001.
Discussion
At present, dCCA, a subtype of CCA, remains a highly lethal disease
despite significant advancements in scientific understanding and
clinical management [[208]2]. In recent years, numerous studies have
been conducted on the molecular pathogenesis of CCA especially for
iCCA. Yet the unraveling of the molecular complexity of dCCA remains
limited, and there are no approved targeted therapies with demonstrated
clinical benefit. PNI, a common pathological feature in dCCA, is
strongly associated with postoperative recurrence and poor prognosis
[[209]58]. Several studies have highlighted the pivotal role of dSCs in
PNI and cancer progression [[210]10, [211]59, [212]60]. A deep
appreciation for the cellular ecosystem of PNI-associated dCCA and the
potential molecular mechanisms underlying the contribution of dSCs to
PNI remains an unmet clinical need. In our study, we utilized scRNA-seq
to comprehensively map the transcriptomic landscape of human
PNI-positive dCCA, unveiling novel cell-cell communications between
dCCA cells and dSCs at single-cell resolution.
Through scRNA-seq analysis, we identified multiple distinct cell types
within PNI-positive dCCAs. Lymphoid cells predominated in PNI-positive
dCCAs, accounting for over 30% of all cells, followed by myeloid cells
and epithelial cells (both malignant and normal epithelium). The
distribution of individual cell subsets within a single sample varied
significantly, demonstrating substantial inter-tumor heterogeneity.
scRNA-seq profiling enabled the definition of three distinct malignant
subtypes. These three malignant subtypes exhibited specific
differentially expressed genes (DEGs) and potential TFs. Importantly,
we observed that an accumulation of NEAT1^+ malignant cells was related
to poorer clinical outcomes and the development of PNI in dCCA
patients, suggesting their highly malignant properties. Notably, we
found that NEAT1^+ malignant cells displayed a highly hypoxic profile.
Hypoxia-induced NEAT1 has been reported to be mediated by HIF-2α
transcriptional activity [[213]61, [214]62]. CNV-wise analysis revealed
a significantly higher proportion of CNV-high malignant cells within
NEAT1^+ malignant cells than other malignant cell types, similarly
confirming a malignant nature. Our research employed pseudotemporal
trajectory analysis to identify distinct states of the three malignant
cells and further characterize their developmental dynamics. Among
these states, NEAT1^+ malignant cells might represent an earlier stage
of differentiation in dCCA cells. We identified several NEAT1^+
malignant cell-associated genes, many of which are related to hypoxic
and oncogenic signaling pathways, such as VEGFA, AKT, JUN, and KRAS.
Previous studies have reported that JUN promotes de-differentiation of
SCs after neural injury by inhibiting P0, MBP, and KROX20 [[215]37,
[216]63]. Interestingly, JUN was also a predicted regulator of NEAT1^+
malignant cells. This suggests the possibility of JUN-mediated cellular
communication between these malignant cells and dSCs, which warrants
further investigation. Additionally, upregulated MALAT1 in NEAT1^+
malignant cells was associated with synapse formation and neuronal cell
survival [[217]64], potentially resulting from the close interaction
between NEAT1^+ malignant cells and dSCs predicted at the single-cell
level. In conclusion, our findings highlight that NEAT1^+ malignant
cells may represent a class of malignant cells with hypoxia propensity
that dominate the PNI-positive dCCA tissues.
As major constituent cells of nerves, SCs have been demonstrated to
promote tumor growth and play a pivotal role in PNI across multiple
tumor types. In this study, we provided hitherto undocumented evidence
of five distinct SC subtypes in human dCCA and its adjacent tissues.
Notably, nmSCs and GFAP^+ dSCs exhibited a higher proportion of cells
within dCCA tissues, while SCPs, PMP22^+ mSCs, and MPZ^+ mSCs were
predominantly found in the adjacent tissues, highlighting the
heterogeneity of the neurological tissue microenvironment in dCCA. Our
analysis revealed that GFAP^+ dSCs expressed minimal myelinating but
high levels of immature SC genes, suggesting they might be a
biochemically and metabolically active subpopulation of SCs.
Subsequently, we verified that CCA cell-derived lactate is a metabolite
that induces and maintains GFAP^+ dSC dedifferentiation. It has been
reported that monocarboxylate transporter protein (MCT) is highly
expressed in perineuronal cells and facilitates lactate uptake as its
preferred energy metabolite [[218]65, [219]66]. The dependence of
peripheral nerve function on lactate metabolism was further emphasized
in a study by Morrison et al., where MCT1 deficiency impeded nerve
regeneration after peripheral nerve injury in mice [[220]67].
Importantly, survival analysis and pseudotemporal analysis indicated
that GFAP^+ dSCs may represent a harmful SC population within the
PNI-positive dCCA neural microenvironment, potentially originating from
MPZ^+ mSCs. Therefore, the GFAP^+ dSCs subpopulation may serve as a
promising therapeutic target for dCCA patients with concomitant PNI.
HMGB1, a representative injury-associated molecule, has been implicated
in various pathological processes, including neurodegenerative
diseases, autoimmunity, and cancer progression [[221]68, [222]69].
Peripheral nerve injury can induce HMGB1 expression through the
proliferation of SCs and infiltration of macrophages within nerves
[[223]70, [224]71]. HMGB1 expression is significantly elevated in pCCA
tissues [[225]72] and is associated with poor prognosis, lymphatic
invasion, and direct involvement in CCA proliferation and angiogenesis
[[226]72, [227]73]. Recent studies have demonstrated that lactate
stimulates macrophage M2 polarization and secretes HMGB1, thereby
promoting glioma cell invasion [[228]74]. Our IHC staining has
confirmed that GFAP^+ and NCAM1^+ peripheral nerves express high levels
of HMGB1 protein internally. This prompted us to investigate whether
SCs also act as a lactate-induced “HMGB1 reservoir”, contributing to
neural infiltration by dCCA cells. In this study, we found that CCA
cell-derived lactate stimulated the dedifferentiation of SCs and
significantly induced HMGB1 expression in GFAP^+ dSCs, enhancing
malignancy of cancer cells. It is widely believed that HMGB1 may
directly contribute to tumor cell metastasis by modifying extracellular
matrix components and regulating cell adhesion properties [[229]75] or
enhance tumor cell progression by inducing MIA [[230]76]. Cellular
immunofluorescence confirmed a significant upregulation of HMGB1 in CCA
cells following co-culture with SCs that exocrine HMGB1, suggesting its
potential role in exacerbating the oncogenicity of cancer cells.
To sum up, our study presents a uniquely matched set of transcriptomic
landscapes within the TME of PNI-positive dCCA and paracancerous
samples, offering a valuable resource for elucidating SC diversity in
PNI-positive dCCA. This study also highlights the intra-tumor crosstalk
between PNI-associated malignant cells and dSCs. Future research is
warranted to corroborate the molecular mechanisms underlying dCCA PNI,
and our dataset can serve as a valuable tool for designing targeted
therapeutics against PNI-positive tumors.
Methods
Patients and clinical samples collection
Two patients with dCCA who did not receive preoperative chemotherapy or
radiotherapy participated in this study. Informed consent was obtained
from all participants, who were requested to donate their tumor tissues
and corresponding peri-tumor tissues for scientific research. Tissue
samples were transported on ice and processed within 30 min of
acquisition.
scRNA-seq and data analysis of dCCA tissues
Single-cell suspensions were prepared for each sample. Cell viability
was ensured to be above 70%, and the cell concentration was adjusted to
300–600 cells/μL. scRNA-seq was performed using the 10× Genomics
Chromium Single Cell 3’ platform following the manufacturer’s
instructions. The generated count matrices were converted to a Seurat
object using the Seurat package (version 4.4.0) [[231]77]. Cells
expressing fewer than 200 genes or with mitochondrial reads exceeding
40% were excluded from downstream analysis. Batch effect correction was
conducted using the Harmony package [[232]78], and the filtered
gene-barcode matrices were normalized using the LogNormalize method.
The top 3000 highly variable genes for principal component (PC)
analysis were identified using the FindVariableFeatures function. The
top 30 PCs were then selected for uniform manifold approximation and
projection (UMAP) visualization of the cells. For cell clustering, the
FindClusters function was employed at a resolution of 0.3. Subgroup
cell clusters were analyzed by selecting the top 30 PCs and clustering
at various resolutions, which were determined through visual
inspection.
Distinguish malignant and non-malignant epithelial cells based on inferred
CNVs
Initial CNVs were estimated using the inferCNV package (version
1.12.0), as previously described [[233]79]. To minimize the impact of
genes with extreme expression, the expression values were
re-standardized and restricted to the range [−2,2]. For each cell, the
mean of the squared CNVi (CNV of the ith window) across the genome was
calculated as the CNV signal. Additionally, the CNV correlation values
were calculated by correlating the CNV profile of a single cell with
the average CNVi profile of the top 5% of cells with the highest CNV
scores. Epithelial cells with a CNV signal above 0.225 and a CNV
correlation above 0.45 were classified as malignant.
Identification of high-confidence hypoxic cells using the CHPF model
CHPF [[234]26] is an open-source modeling framework designed to predict
cellular hypoxia status. The CHPF script was executed in Python
(version 3.11.5). The single-cell expression profile and seven
pre-selected hypoxia-related gene sets served as input files for the
construction of the prediction model. The final formula was provided
below:
[MATH:
P(x)=∑i=1n
Wi(x)fi(x)
∑i=1nWi(x)(n=100)
:MATH]
Cells with P(x) > 0.5 were considered as hypoxic cells.
Pseudotime analysis by Monocle
Pseudo-time analysis and transcriptome dynamic analysis along the
pseudo-time trajectory were conducted using Monocle2 (version 2.26.0)
[[235]80] with the default parameters recommended by the developer.
Deconvolution analysis
We employed CIBERSORTx ([236]https://cibersortx.stanford.edu/), a
deconvolution analysis tool, to investigate gene expression within the
TME. Our analysis focused on the 10x scRNA-seq data, specifically
looking for DEGs between epithelial cell subtypes and SC subtypes.
These genes were used to create a signature matrix. To deconvolute the
bulk RNA-sequencing data, we employed two separate reference sources:
1) data from the TCGA-CHOL cohort within The Cancer Genome Atlas
(TCGA); and 2) an RNA-sequencing dataset ([237]GSE107943) downloaded
from the Gene Expression Omnibus database. These datasets served as the
mixture files for CIBERSORTx analysis.
H&E staining and IHC assay
22 cases of dCCA tissues that underwent pancreaticoduodenectomy were
obtained from the First Affiliated Hospital of Zhengzhou University
between 2021 and 2024. Tissues were fixed with 4% paraformaldehyde,
embedded in paraffin, and cut into 4-μm thick sections.
For H&E staining, the sections were stained with hematoxylin (BA4097,
BaSo Diagnostics Inc., Zhuhai, China) for 5 min and eosin (BA4098, BaSo
Diagnostics Inc.) for 3 min. For the IHC assay, sections were
deparaffinized, rehydrated, and blocked. Primary antibodies were
incubated at 4 °C overnight, followed by incubation with goat
anti-mouse IgG H&L HRP (1:4000, SA00001-1, Proteintech, Wuhan, China)
at room temperature for 2 h. The primary antibodies used in this study
included GFAP (1:5000, 60190-1-Ig, Proteintech), NCAM1 (1:5000,
14255-1-AP, Proteintech), and HMGB1 (1:400, 66525-1-Ig, Proteintech).
The staining extent score (<25%, score = 1; 25–50%, score = 2; 50–75%,
score = 3; >75%, score = 4) and staining intensity (negative, score =
0; weak, score = 1; moderate, score = 2; strong, score = 3) were
assessed using ImageJ software 1.46r. IHC results were scored by
multiplying the staining extent score by the intensity score. All H&E
stained and IHC sections were scanned with a Pannoramic MIDI II scanner
(3D HISTECH Ltd., Hungary).
Cell culture and treatment
CCLP1 (JNO-H0653) and HUCCT1 (BNCC337995) human CCA cell lines were
obtained from Jennio Biotech Co., Ltd. (Guangzhou, China) and Beina
Chuanglian Biotechnology Institute (Beijing, China), respectively.
Human ipNF95.6 SCs (CTCC-001-0379, Meisen CTCC, Panan, China) and rat
RSC96 SCs (CL-0199, Pricella Biotechnology Co., Ltd., Wuhan, China)
were maintained in our laboratory. All cells were cultured in DMEM
(12100, Solarbio, Beijing, China) medium supplemented with 10% fetal
bovine serum (C04001-500, VivaCell, Shanghai, China). The normoxia
group cells were cultivated at 37 °C in a 5% CO[2] humidified incubator
(Galaxy 170R, Eppendorf, Hamburg, Germany). For hypoxia induction,
cells were transferred to a tri-gas incubator (Galaxy 48R, Eppendorf)
and incubated under 1% O[2], 5% CO[2], and 94% N[2] for 48 h prior to
commencing subsequent experiments.
A non-contact coculture system was established using 24-well plates
containing 0.4-μm polyethylene terephthalate membrane filters (Corning,
NY, USA) to separate the lower and upper chambers. ipNF95.6 cells,
subjected to various treatments, were seeded in the upper chamber at a
density of 1 × 10^5 cells/mL. CCLP1 or HUCCT1 cells were then
inoculated in the lower chamber at a density of 1.5 × 10^5 cells/mL.
Following a 48-h incubation period, CCA cells were harvested for
subsequent experiments.
To stimulate lactate production, SCs were cultured for 8 h with
increasing concentrations of lactate (0, 10, 20, 40, 80, and 160 mM;
L1750, Merck, NJ, USA).
Glycyrrhizin (B20417-20mg, Yuanye Bio-Technology, Shanghai, China), a
direct inhibitor of HMGB1, was added to SCs [[238]55] at a
concentration of 1 nM in conjunction with lactate to suppress HMGB1
expression in these cells.
The supernatants of CCA cells cultured under normoxic or hypoxic
conditions were collected by centrifuging at 500 × g for 10 min,
followed by a second centrifugation at 2000 × g for 20 min. After any
necessary pre-treatments, SC supernatants were collected using the same
procedure. These collected supernatants were stored at −80 °C and used
within one month.
Cell–cell interaction analysis
Intercellular communication between malignant and SC types was
investigated using the CellChat package (version 1.6.1) [[239]81]. The
Seurat-normalized data was transformed into a CellChat object using the
createCellChat function. Subsequently, the computeCommunProbPathway
function was employed to infer intercellular communication at a
signaling pathway level.
SCENIC analysis
SCENIC (version 1.3.1) was utilized to evaluate the transcriptional
activity of malignant cells and SCs [[240]44]. SCENIC was implemented
in R using the motif databases of RcisTarget and GRNboost
(corresponding to GENIE3 1.20.0, AUCell 1.22.0, and RcisTarget 1.18.2).
Raw UMI counts served as input for the analysis.
Western blot assay
The total proteins were extracted with RIPA buffer (R0010, Solarbio).
Proteins were separated by 10% SDS-PAGE and transferred to
nitrocellulose membranes. The membranes were incubated with primary
antibodies at 4 °C overnight, and specific binding of the primary
antibodies was detected with peroxidase-labeled goat anti-mouse
(1:4000, SA00001-1, Proteintech) or goat anti-rabbit (1:4000,
SA00001-2, Proteintech) secondary antibodies. The following primary
antibodies were used: HIF-1α (1:4000, ab51608, Abcam), GFAP (1:5000,
60190-1-Ig, Proteintech), NCAM1 (1:20,000, 14255-1-AP, Proteintech),
SOX2 (1:1000, #2748, CST, Boston, USA) and HMGB1 (1:3000, 66525-1-Ig,
Proteintech) and β-actin (1:40,000, 66009-1-Ig, Proteintech).
Lactate measurement
The lactate concentrations in cell supernatants and tissues were
determined by an L-Lactate Assay Kit (A019-2-1, Nanjing Jiancheng
Bioengineering Institute, Nanjing, China). Samples were prepared
according to the manufacturer’s instruction and the lactate levels of
the samples were calculated by measuring the absorbance at 530 nm.
Cell cytotoxicity experiment
For the thiazolyl blue tetrazolium bromide (MTT) assay, MTT (M5655,
Sigma, Shanghai, China) was dissolved in DMEM at a concentration of
5 mg/mL. Cells were incubated with MTT for 4 h at 37 °C under 5% CO[2].
After removing the MTT, the formed MTT-formazan crystal was dissolved
in DMSO (150 μL/well). Absorbance at 490 nm was measured using a
microplate reader (Spark, Tecan, Switzerland). The results were
expressed as the percentage change in absorbance compared to untreated
control cells, which were set to 100%. Data represent the average of
triplicate measurements from three independent experiments.
Real‑time quantitative polymerase chain reaction (RT‑qPCR)
Total RNA from cells was extracted using TriQuick Reagent (R1100,
Solarbio). cDNA was synthesized using NovoScriptPlus All-in-one 1st
Strand cDNA Synthesis SuperMix (E047-01B, Novoprotein Scientific Inc.,
Shanghai, China). RT-qPCR was conducted in a 20-μL reaction volume
containing forward and reverse primers, cDNA, and NovoStart SYBR qPCR
SuperMix Plus (E096-01A, Novoprotein Scientific Inc.). All primers were
synthesized by Sangon Biotech and normalized to GAPDH. RNA folding
changes were quantified using the 2^−ΔΔCt method. The primer sequences
are listed in Table [241]S8.
Immunofluorescence staining
Cell climbing slices were sterilized with 75% alcohol, air-dried in
24-well plates, and inoculated with CCA cells or SCs. After that, cells
were fixed with 4% paraformaldehyde, permeabilized with 0.2% Triton
X-100 for 15 min at 4 °C, and blocked with 5% bovine serum albumin.
Primary HMGB1 antibody (1:200, 66525-1-Ig, Proteintech) was added
overnight at 4 °C. Cells were then incubated with goat anti-mouse IgG
(H&L) secondary antibody (1:400, Alexa Fluor 488, GB25301, Servicebio,
Wuhan, China) or (1:200, Alexa Fluor 594, AB0152, Abways, Shanghai,
China) for 2 h at room temperature, protected from light, and
counterstained with DAPI for 10 min. Images were captured using a DM4B
microscope system (Leica, Wetzlar, Germany).
Wound healing assay
A coculture system was employed for the scratching experiment.
Following coculture completion, the upper chamber was removed, and the
cells in the lower chamber were rinsed with phosphate-buffered saline
(PBS). Cells were maintained in culture until reaching 90% confluence.
Two lines were then scratched using a 200 μL pipette tip. Nonadherent
cells were washed away twice with PBS and incubated in serum-free DMEM
for 24 h. Microscopic observation and photography were performed at 0
and 24 h. The ImageJ software (version 1.6.0) was utilized to analyze
the wound healing rate by quantifying the wound closure area.
Cell invasion assay
A cell invasion assay was conducted using 24-well plates equipped with
8.0-μm pore polycarbonate membrane inserts (Corning). Firstly, 100 μL
matrigel (diluted with the serum-free DMEM at 1:8, cat#354234, Corning)
was added to the upper chamber and incubated at 37 °C for 2 h. Then,
the upper chamber was added with 100 μL serum-free DMEM and incubated
for another 30 min. Afterward, the liquid in the chamber was removed. A
200 µL cell suspension (serum-free DMEM) was seeded in the upper
chamber, and Medium (600 µL) containing 10% FBS was added to the lower
chamber. Following incubation at 37 °C for 48 h, non-migratory cells on
the top surface of the inserts were gently removed with cotton swabs.
Subsequently, the inserts were fixed in 4% paraformaldehyde for 15 min,
and the cells were stained with 0.2% crystal violet. The number of
cells that invaded the membrane was quantified under a light
microscope.
Tumorigenicity assay
Tumor xenografts were established in 5-week-old female BALB/c nude mice
obtained from Beijing Vital River Laboratory Animal Technology Co., Ltd
(Beijing, China). The mice were maintained in specific pathogen-free
conditions with a 12-h light/dark cycle. All mice were randomly
allocated to 3 groups (8 mice per group). After acclimatization for 1
week, 5 × 10^6 CCLP1 cells mixed with 2 × 10^6 ipNF95.6 cells were
injected subcutaneously on the right flank of the mice. ipNF95.6 cells
were pretreated with PBS, lactate, or lactate combined with
glycyrrhizin for over three generations. Following inoculation, tumor
volume was measured every three days and calculated using the formula
V = ab^2/2, where a is the long diameter and b is the short diameter.
Statistical analysis
Data were analyzed using GraphPad Prism (version 8.0.1.244, GraphPad
Software Inc., San Diego, USA). The Shapiro-Wilk test was used to
assess the normality of each distribution. Student’s t-test or the
Wilcoxon rank-sum test was employed to compare two groups, while ANOVA
followed by Tukey’s multiple comparisons test was used for multiple
group comparisons. Chi-square test was used for categorical variables
distribution test. Kaplan-Meier curves were constructed and analyzed
using the log-rank test. All data were presented as means ± standard
deviation. All in vitro experiments were conducted in at least three
independent experiments. *P < 0.05, **P < 0.01, ***P < 0.001, ns: not
significant.
Supplementary information
[242]Supplemental Figures^ (4.5MB, docx)
[243]Supplemental Tables^ (35.9KB, xlsx)
[244]Raw WB data^ (8.3MB, pdf)
Acknowledgements