Integrated DNA and RNA profiling refines prognostic stratification independent of therapeutic actionability in cholangiocarcinoma

Abstract

Background

Cholangiocarcinoma (CCA) exhibits marked biological heterogeneity. While genomic profiling identifies targetable alterations such as FGFR2 fusions and IDH1 mutations, transcriptomic analyses reveal distinct immune, proliferative and mesenchymal profiles. How these genomic and transcriptomic features jointly influence clinical outcomes remains unclear.

Methods

We performed integrated DNA and RNA profiling in 62 patients with intrahepatic and extrahepatic CCA treated at a single tertiary center. Targeted DNA sequencing assessed single-nucleotide variants, copy-number alterations, and microsatellite instability. Extended RNA sequencing evaluated gene fusions and pathway-level transcriptomic subtypes. Associations with progression-free survival (PFS) and overall survival (OS) were examined.

Results

Among 58 DNA-profiled tumors, the most frequent alterations were TP53 (43.1%) and KRAS (29.3%), followed by IDH1 (10.3%, restricted to intrahepatic CCA). Actionable alterations (IDH1, FGFR2 fusions, ERBB2 amplification, microsatellite instability-high) were detected in 25% of cases and associated with longer overall survival (67.5 vs. 20.8 months; p = 0.0004), consistent with benefit from matched therapies rather than intrinsic tumor biology. RNA profiling identified five transcriptomic subtypes: Immune (21%), Proliferative (18%), Mesenchymal (42%), Immune-Proliferative (8%), and Unclassified (11%). KRAS-TP53 co-mutation was the strongest adverse prognostic factor (38.1 vs. 13.2 months; p = 0.0004). The Mesenchymal subtype was associated with shorter PFS (5.9 vs. 17.9 months; p = 0.045) in patients treated with chemotherapy ± immunotherapy but did not significantly affect OS.

Conclusions

Integrated genomic and transcriptomic profiling refines prognostic stratification in cholangiocarcinoma independent of therapeutic actionability. KRAS-TP53 co-mutation and the Mesenchymal transcriptomic subtype represent independent high-risk markers detectable on routine FFPE tissue. These features complement actionable alterations and may inform patient selection and clinical trial design.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12885-026-16064-5.

Introduction

Cholangiocarcinoma is an aggressive malignancy of the biliary tract with rising incidence and consistently poor outcomes. For more than a decade, gemcitabine–cisplatin represented the standard first-line therapy, yielding a median overall survival of approximately one year [1]. More recently, the addition of immune checkpoint inhibitors has modestly improved outcomes in advanced disease (TOPAZ-1, KEYNOTE-966), establishing chemo-immunotherapy as the new standard of care [2, 3]. Current ESMO guidelines therefore recommend systematic molecular profiling to identify actionable alterations such as FGFR2 fusions, IDH1 mutations, and MSI-H status, which may guide therapeutic decisions [4].

Large-scale genomic studies have revealed the marked heterogeneity of CCA, including recurrent alterations in KRAS, TP53, SMAD4, and BAP1, defining distinct molecular subgroups [5–8]. In parallel, transcriptomic analyses have identified reproducible immune, proliferative, and mesenchymal programs that reflect tumor–microenvironment interactions and correlate with clinical behavior [9–11]. Despite these advances, actionable genomic alterations alone do not account for the wide variability in early progression observed in patients treated with standard chemotherapy, suggesting that additional biological determinants of outcome remain to be identified. We therefore hypothesized that integrating genomic co-mutation patterns with pathway-level transcriptomic programs would identify biologically distinct CCA subgroups associated with divergent survival trajectories, independently of therapeutic actionability. To test this hypothesis, we applied combined DNA and RNA profiling to a hospital-based CCA cohort and examined whether the resulting molecular features could refine prognostic stratification beyond what is captured by actionable alterations alone, with potential implications for patient risk stratification and clinical trial design.

Materials and methods

Patient cohort and sample collection

Patients with histologically confirmed cholangiocarcinoma (CCA) treated at Montpellier University Hospital between January 2016 and June 2023 were retrospectively included. Clinical data, including tumor location, stage, treatments received, and survival outcomes, were obtained from institutional databases. All patients provided written informed consent for molecular testing. The study was approved by the institutional ethics committee (IRB/CPP No. 2023-CCA-01) and conducted in accordance with the Declaration of Helsinki.

Formalin-fixed paraffin-embedded (FFPE) tumor blocks underwent pathological review to confirm diagnosis and assess tumor cellularity. Macrodissection was performed when required to achieve ≥ 20% tumor content. Genomic DNA (gDNA) and total RNA were extracted using Maxwell^® RSC FFPE DNA and RNA kits (Promega), following the manufacturer’s instructions. Nucleic acid concentrations were measured using Qubit dsDNA HS and RNA HS assays (Thermo Fisher Scientific).

DNA sequencing

Genomic DNA was analyzed with the Advanta™ Solid Tumor NGS Panel (49 oncology-relevant genes, Fluidigm, supplementary Table S1) as previously described [12]. Libraries were sequenced on a NextSeq 500 instrument (2 × 150 cycles, Illumina). Single nucleotide variants (SNVs), small insertions/deletions (indels), copy number variations (CNV) and microsatellite instability (MSI) were detected using in-house bioinformatics workflows [13–16]. Variants with variant allele frequency (VAF) ≥ 5% were annotated using OncoKB, ClinVar, and COSMIC. Four DNA libraries failed quality control and were excluded from SNV/indel analyses.

RNA sequencing

The Archer^® FusionPlex^® Lung kit (IDT) based on an anchored multiplex PCR method was first used to detect recurrent fusions, including FGFR2, as previously described [17]. For this, 200 ng of total RNA was used as input for library generation according to the manufacturer’s instructions. Libraries were sequenced on a MiniSeq instrument (2 × 150 cycles, Illumina). Results were analyzed using the Archer Analysis v6.2.7 software. A sample was considered as positive when the fusion breakpoint was supported by at least five unique reads.

An extended targeted RNA sequencing was additionally performed in a translational research setting using the RNA Prep with Enrichment (L) Tagmentation in combination with the RNA Pan-cancer panel (1,362 genes, Illumina). Briefly, 50 ng of DNAse-treated total RNA was reverse-transcribed into double-stranded cDNA, tagmented and amplified with unique dual indexes to generate indexed libraries. After quantification and normalization to 200 ng, three libraries were pooled and subjected to hybridization-based capture using biotinylated probes, followed by magnetic bead capture and stringent heated washes. The enriched libraries were finally re-amplified, purified, and quality-checked prior to sequencing on a NextSeq 500 instrument (2 × 150 cycles, Illumina). Raw counts matrix obtained using the DRAGEN RNA pipeline (v4.2.7, Illumina) were further analyzed to stratify samples based on their signaling activity.

Ligand-receptor based pathway activity inference from transcriptomic analysis

We used BulkSignalR because established CCA transcriptomic classifiers [9, 10] require genome-wide expression data incompatible with our targeted panel. BulkSignalR enables pathway activity inference from targeted bulk data by integrating ligand-receptor correlations with downstream pathway information. Pathway-level scores were inferred from bulk transcriptomic data using the BulkSignalR R package (v3.22) [18]. Briefly, all potential ligand-receptor (L-R) interactions and their associated downstream pathways were inferred. To minimize redundancy from reference databases, this exhaustive list was systematically filtered following BulkSignalR’s documentation. L-R interactions were first aggregated by their common downstream pathway (reduceToPathway function), retaining the most significant p-value for each pathway. This set was further refined by assigning each L-R interaction to only its single best-scoring pathway (reduceToBestPathway function). Gene signatures were then generated for the resulting pathways using a significance threshold (q-value < 0.01). Finally, these signatures were used to calculate a pathway activity score for each sample, forming the basis for transcriptomic subtype assignment.

Molecular stratification

Patients were categorized by: (i) actionable alterations according to the ESMO Scale for Clinical Actionability of molecular Targets (ESCAT I-II) [19]: IDH1, FGFR2 fusions, ERBB2 amplification, MSI-H; (ii) KRAS mutation; (iii) TP53 mutation; (iv) KRAS-TP53 co-mutation; and (v) transcriptomic subtype.

Statistical analysis

Overall survival (OS) and progression-free survival (PFS) were estimated using Kaplan–Meier method and the significance of difference between survival rates was ascertained with the log-rank test using SPSS^® Software (SPSS Inc). Multivariable analysis used Cox models adjusted for age, sex, location and treatment. p < 0.05 was considered significant.

Results

Patient cohort characteristics

Sixty-two patients with histologically confirmed CCA were included, with a median age of 66 years (range 33–80) and equal sex distribution (31 males, 31 females; 50% each). Most tumors were intrahepatic (iCCA: 47/62, 75.8%), with the reminder classified as extrahepatic (eCCA: 15/62, 24.2%). Most patients were diagnosed at advanced stages III–IV (53/62, 85.5%; Table 1). DNA sequencing was successful in 58 of 62 samples (93.5%); four were excluded for insufficient DNA quality. RNA profiling was completed in all 62 cases (100%).

Table 1.

Patient and sample characteristics

Characteristic		n
Age, years (median, range)		66 (Range 33–80)
Sex
	Male	31 (50%)
	Female	31 (50%)
Tumor location
	iCCA	47 (75.8%)
	eCCA	15 (24.2%)
Disease stage (AJCC)
	Stage I-II	8 (12.9%)
	Stage III-IV	53 (85.5%)
	Unknown	1 (1.6%)
Sample type
	Biopsy	36 (58.1%)
	Tumor resection	26 (41.9%)
Treatment
	Chemotherapy alone	22
	Chemotherapy + Immunotherapy	16
	Chemotherapy + Targeted therapy	7
	Chemotherapy + Immunotherapy + Targeted therapy	6
	Immunotherapy alone	1
	No treatment	9

Molecular landscape

Among the 58 DNA-profiles tumors, TP53 (25/58, 43.1%) and KRAS (17/58, 29.3%) were the most frequently altered genes, followed by IDH1 (6/58, 10.3%, exclusively in iCCA). The global distribution of genomic alterations across the cohort is shown in Fig. 1A. Additional alterations included PIK3CA (6/58, 10.3%), SMAD4 (4/58, 6.9%), and STK11 (4/58, 6.9%) (Supplementary Table S2). MSI-H status was detected in three tumors (4.8%). Mutation frequencies stratified by tumor location (iCCA vs. eCCA) are presented in Fig. 1B. Actionable alterations according to ESCAT I–II criteria [19], comprising IDH1 mutations (n = 6), FGFR2 fusions (n = 5), MSI-H (n = 3), ERBB2 amplification (n = 1) and RET fusion (n = 1) were found in 16 of 62 tumors (25.8%, Table 2). As expected, FGFR2 fusions were mutually exclusive with KRAS and TP53 mutations. CNV analysis revealed recurrent losses at CDKN2A and NOTCH1 and gains at FGFR3, TERT and ERBB2 (Fig. 1C and Supplementary Table S3). Both RNA sequencing analyses detected FGFR2 fusions in five tumors (8%), all arising in iCCA, involving partners BICC1 (n = 3), CTNNA3 (n = 1), and TNIP3 (n = 1), as well as a RET-NCOA4 fusion. No NTRK or ALK fusions were observed.

Fig. 1.

Integrated genomic landscape of cholangiocarcinoma. A Genomic alterations across the 62 cholangiocarcinoma samples. Each column represents one sample; each row represents one gene. Alterations are color-coded as follows: blue, wild-type (WT); black dot, presence of a variant; light yellow, copy loss; red, copy gain; light grey, not available (NA). Gene fusions are represented as colored annotation tracks above the oncoprint: magenta, FGFR2 fusion; green, RET fusion. Microsatellite stability status is shown as an additional top track: violet, MSI. B Vairant frequency by gene stratified by tumor location (blue, iCCA, n = 47; orange, eCCA, n = 15). C Copy number variation frequency by gene (blue, iCCA; orange, eCCA). iCCA, intrahepatic cholangiocarcinoma; eCCA, extrahepatic cholangiocarcinoma

Table 2.

Molecular alterations

Alteration		Total	iCCA (n=47)	eCCA (n=15)	Actionable (ESCAT I-II)	Notes
DNA alterations*
	TP53 mutation	25/58 (43.1%)	17	8	No	Poor prognosis marker
	KRAS mutation	17/58 (29.3%)	9	8	No	Poor prognosis marker
	IDH1 mutation	6/58 (10.3%)	6	0	Yes	Ivosidenib approved; iCCA-restricted
	ERBB2 amplification	1/62 (1.6%)	1	0	Yes	HER2-directed therapies
Microsatellite instability		3/62 (4.8%)	2	1	Yes	Immune checkpoint inhibitor therapies
RNA fusions
	FGFR2 fusions	5/62 (8.1%)	5	0	Yes	Pemigatinib approved; Partners: BICC1 (n=3), CTNNA3 (n=1), TNIP3 (n=1)
	RET fusions	1/62 (1.6%)	1	0	Yes	Traitement

*Due to low DNA quality, SNV and indels could be analysable for 58 samples only

iCCA Intrahepatic cholangiocarcinoma, eCCA Extrahepatic cholangiocarcinoma, ESCAT ESMO Scale for Clinical Actionability of molecular Targets

Transcriptomic subtype classification

Analysis of ligand–receptor (L–R) expression profiles identified a set of biologically coherent activated pathways, which clustered into three major functional categories (Supplementary Table S4): immune-related pathways; mesenchymal programs involving extracellular matrix organization and developmental signaling (including TGF-β, WNT, and NOTCH); and proliferative pathways related to growth factor signaling and cell-cycle progression. Based on these pathway activity patterns, most tumors could be assigned to a transcriptomic subtype: Mesenchymal (26/62, 42%), Immune (13/62, 21%), Proliferative (11/62, 18%), or Immune–Proliferative (5/62, 8%), while 7 samples (11%) remained unclassified (Fig. 2).

Fig. 2.

Transcriptomic subtypes of cholangiocarcinoma. Heatmap of BulkSignalR-inferred pathway activation scores across the 62 samples stratified by transcriptomic subtype. Each column represents one sample; each row represents a biological pathway. Rows are grouped into three functional categories: immune-related pathways (top); mesenchymal-related pathways (middle); and proliferative-related pathways (bottom). Color scale represents pathway activation score: blue, low activity; white, intermediate activity; red, high activity. Samples are grouped from left to right into five transcriptomic subtypes: Unclassified (n = 7, 11%), Mesenchymal (n = 26, 42%), Immune-Proliferative (n = 5, 8%), Proliferative (n = 11, 18%), and Immune (n = 13, 21%)

Prognosis significance of molecular subtypes

Survival analyses revealed distinct prognostic associations. Median OS was 26.8 months. The presence of actionable alterations was associated with significantly longer OS (67.5 vs. 20.8 months; HR 3.65, 95% CI 1.78–7.44, p = 0.0004; Table 3; Fig. 3A), likely reflecting benefit from matched targeted therapies. Conversely, both KRAS and TP53 mutations were individually associated with shorter OS (18.3 vs. 38.1 months, p = 0.005; and 15.1 vs. 39.1 months, p = 0.005, respectively) (Fig. 3B–C). KRAS-TP53 co-mutation identified patients with the worst prognosis (13.2 vs. 38.1 months; HR 0.15, 95% CI 0.05–0.43, p = 0.0004; Fig. 3D). Transcriptomic subtypes were not significantly associated with OS (p = 0.71). Similarly, tumor location (iCCA vs. eCCA) showed no significant OS difference (HR 1.14, 95% CI 0.53–2.44, p = 0.73). In multivariable Cox models, both the presence of actionable alterations (p = 0.007) and KRAS-TP53 co-mutation (p = 0.016) remained independent predictors of OS (Table 3).

Table 3.

Univariate and multivariates analysis of clinical parameters with regards to overall survival

OS	Number of samples	Univariate analysis			Multivariate analysis
		HR	95% CI	Pa	HR	95% CI	P
Sex (female; male)	62	0.65	0.33-1.27	NS (0.21)	ND
Age at diagnosis (<60; ≥60-year-old)	62	0.97	0.49-1.93	NS (0.94)	ND
Tumor location (eCCA; iCCA)	62	1.14	0.53-2.44	NS (0.73)	ND
Targetable alteration (no; yes)	61	3.65	1.78-7.44	0.0004	4.79	1.58-14.56	0.007
Co-mutation KRAS-TP53 (no; yes)	58	0.15	0.05-0.43	0.0004	0.39	0.18-0.85	0.016
Transcriptomic subtype (others; mesenchymal)	62	1.13	0.59-2.19	NS (0.71)	ND

HR Hazard ratio, 95% CI 95% confidence interval, iCCA Intrahepatic cholangiocarcinoma, eCCA Extrahepathic cholangiocarcinoma, NS Not significant, ND Not done

^aP was considered significant when P < 0.05

Fig. 3.

Prognostic impact of genomic alterations on overall survival. Kaplan-Meier curves showing overall survival (OS) stratified by genomic features and compared using the log-rank test. A OS according to the presence of an actionable alteration (IDH1 mutation, FGFR2 fusion, MSI-H, ERBB2 amplification or RET fusion; n = 16) vs. absence (n = 42); P = 0.0004. B OS according to the presence of a KRAS mutation (KRAS^mut, n = 17) vs. the absence of a KRAS mutation (KRAS^wt, n = 41). C OS according to the presence of a TP53 mutation (TP53^mut, n = 25) vs. absence of a TP53 mutation (TP53^wt, - n = 33). D OS according to presence of a KRAS–TP53 comutation (comutation KRAS-TP53, n = 12) vs. others (n = 46)

For patients treated with chemotherapy ± immunotherapy (n = 52), median PFS was 8.2 months. The Mesenchymal subtype was associated with significantly shorter PFS (5.9 vs. 17.9 months; HR 0.54, 95% CI 0.30–0.99, p = 0.045; Fig. 4A). KRAS-TP53 co-mutation similarly predicted inferior PFS (8.8 vs. 17.8 months; HR 0.42, 95% CI 0.18–0.96, p = 0.039; Fig. 4B). Both variables remained independently associated with shorter PFS in multivariable analysis (p = 0.039 and p = 0.015, respectively; Table 4). Notably, the Mesenchymal subtype did not influence OS, suggesting a predominant impact on early progression dynamics rather than long-term survival.

Fig. 4.

Prognostic impact of transcriptomic and genomic features on progression-free survival. Kaplan–Meier curves showing progression-free survival (PFS) in patients treated with chemotherapy ± immunotherapy (n = 52). A PFS in patients with a tumor sample classified in the mesenchymal subtype (nesenchymal, n = 24) vs. others (n = 28). B PFS in patients with a KRAS–TP53 co-mutation (comutation KRAS-TP53, n = 12) vs. others (n = 37)

Table 4.

Univariate and multivariates analysis of clinical parameters with regards to progression-free survival

	Number of samples	Univariate analysis			Multivariate analysis
		HR	95% CI	Pa	HR	95% CI	P
Sex (female; male)	52	1.15	0.64-2.06	NS (0.64)	ND
Age at diagnosis (<60; ≥60-year-old)	52	0.66	0.37-1.19	NS (0.17)	ND
Tumor location (eCCA; iCCA)	52	0.78	0.42-1.46	NS (0.44)	ND
Targetable alteration (no; yes)	51	1.79	0.97-3.27	NS (0.06)	ND
Co-mutation KRAS-TP53 (no; yes)	49	0.42	0.18-0.96	0.039	0.41	0.20-0.84	0.015
Transcriptomic subtype (others; mesenchymal)	52	0.54	0.30-0.99	0.045	0.52	0.38-0.97	0.039

HR Hazard ratio, 95% CI 95% confidence interval, iCCA Intrahepatic cholangiocarcinoma, eCCA Extrahepathic cholangiocarcinoma, NS Not significant, ND Not done

^aP was considered significant when P < .05

Discussion

This study demonstrates the feasibility and clinical value of integrated DNA and RNA profiling in a routine hospital setting. Patients harboring actionable alterations, for which targeted therapies such as futibatinib for FGFR2 fusions [20] and ivosidenib for IDH1 mutations [21–23] have recently received European regulatory approval, showed improved survival. Conversely, KRAS-TP53 co-mutation and the Mesenchymal transcriptomic subtype emerged as independent adverse prognostic markers, identifying patients at high risk of early progression under standard therapy. Our results confirm that transcriptomic heterogeneity described in large-scale studies [9, 10] remains detectable using a targeted RNA panel applied to routine FFPE samples. While Mody et al. [11] demonstrated the value of combining genomic and transcriptomic profiling in biliary tract cancers, our study extends these observations by showing that prognostic stratification can be achieved without genome-wide expression profiling, using markers readily assessable from routine diagnostic workflows.

At the molecular level, KRAS-TP53 co-mutation combines two complementary oncogenic mechanisms. KRAS activation sustains proliferative signaling through the MAPK and PI3K pathways, while TP53 loss abrogates cell-cycle checkpoints and DNA damage repair, fostering genomic instability and clonal evolution. Their convergence also promotes epithelial-mesenchymal transition, enhancing invasive potential and metastatic dissemination [24]. Consistent with preclinical models, several of the genetic alterations identified in our cohort, including KRAS and TP53 mutations, have been associated with oncogenic signaling and aggressive phenotypes in established CCA cell lines. However, such models do not fully recapitulate the complexity of primary tumors, particularly microenvironment-driven transcriptomic features such as those defining the Mesenchymal subtype. These features likely explain the shorter response under chemotherapy observed in this subgroup and suggest that conventional cytotoxic regimens may be insufficient to control disease progression. The Mesenchymal subtype, characterized by upregulation of extracellular matrix remodeling, TGF-β, WNT and NOTCH signaling, likely reflects a distinct tumor microenvironment dominated by stromal desmoplasia and immune exclusion. Dense fibrotic stroma may impair drug delivery and create hypoxic niches that favor tumor cell survival. Additionally, TGF-β-driven immunosuppression may limit the efficacy of checkpoint inhibitors, potentially explaining why this subtype showed reduced progression-free survival despite the addition of immunotherapy to chemotherapy backbone [25].

Recognizing these high-risk signatures at diagnosis opens opportunities for therapeutic innovation. In patients with tumors exhibiting mesenchymal subtype, strategies aimed at disrupting the stromal barrier, such as anti-fibrotic agents or TGF-β inhibitors, could restore drug penetration and immune infiltration [26]. For KRAS-mutant patients, the therapeutic landscape is rapidly evolving. Although G12C prevalence remains low in cholangiocarcinoma, G12D inhibitors and pan-KRAS agents such as RMC-6236, which block the nucleotide-exchange pocket of multiple hotspot variants including G12D and G12V, are currently being evaluated in basket studies enrolling biliary tract cancer patients [27]. Synthetic lethal approaches targeting downstream effectors or metabolic vulnerabilities of KRAS-TP53 co-mutated tumors also warrant exploration.

Based on these findings, we propose a stepwise molecular approach for clinical practice: first, targeted DNA sequencing to identify actionable alterations guiding treatment selection; second, systematic reporting of KRAS-TP53 co-mutation status to flag high-risk patients warranting closer monitoring or early clinical trial enrollment; and third, extended RNA profiling when feasible to refine progression-free survival prediction, particularly in patients receiving chemo-immunotherapy combinations.

Extended RNA profiling required substantially more laboratory time, bioinformatics processing and overall resources compared to DNA sequencing. These constraints may contribute to the fact that RNA-based classifiers are not yet routinely implemented in diagnostic workflows. Established CCA classifiers [9, 10] and compact signatures such as CORE-37 [28] or Wada-8 [29] could not be directly applied, as only very few of their constituent genes were covered by our targeted RNA panel, preventing meaningful benchmarking against these models. This limited overlap justified the use of a pathway-level approach with BulkSignalR, which provided biologically interpretable stratification despite restricted transcriptomic coverage. However, further optimization and harmonization of targeted RNA panels will be important to enable broader comparability across studies. Until their clinical utility is demonstrated in treatment selection or risk-adapted strategies, comprehensive RNA profiling will likely remain a translational tool rather than a diagnostic standard.

Our study has several limitations. The modest sample size (n = 62) reduces statistical power for detecting moderate effects and may increase overfitting risk in transcriptomic subtype classification. Immunohistochemical analysis of key proteins characterizing transcriptomic subtypes would be of great interest; however, it was not feasible in the present cohort due to limited residual FFPE material following nucleic acid extraction. Future prospective studies should consider such analyses to establish correlations between protein expression and the identified transcriptomic phenotypes. Treatment heterogeneity, particularly variable access to targeted therapies and immunotherapy combinations over the study period, may have contributed to some of the observed survival differences. Finally, retrospective design and single-center recruitment limit generalizability. Prospective validation in larger, multi-center cohorts with standardized treatment protocols is warranted before these markers can be implemented in routine clinical decision-making.

Conclusions

Integrated genomic and transcriptomic profiling refines prognostic stratification in cholangiocarcinoma beyond therapeutic actionability. KRAS-TP53 co-mutation and Mesenchymal transcriptomic subtype represent independent high-risk markers that complement actionable alterations. Both features are assessable from routine FFPE tissue using targeted approaches and provide clinically relevant prognostic information, particularly for progression-free survival, and may help inform the design of future biomarker-driven trials. Prospective validation and development of compact classifiers will be key for implementation in routine diagnostics.

Abbreviations

CCA

Cholangiocarcinoma

iCCA

Intrahepatic cholangiocarcinoma

eCCA

Extrahepatic cholangiocarcinoma

OS

Overall survival

PFS

Progression-free survival

SNV

Single-nucleotide variant

CNV

Copy-number variation

MSI

Microsatellite instability

MSI-H

Microsatellite instability-high

FFPE

Formalin-fixed paraffin-embedded

ESCAT

ESMO Scale for Clinical Actionability of molecular Targets

NGS

Next-generation sequencing

HR

Hazard ratio

CI

Confidence interval

DNA

Deoxyribonucleic acid

RNA

Ribonucleic acid

Authors’ contributions

JAV and JS designed the study. JAV, SCA, and ID performed data collection and analysis. SCA developed bioinformatics pipelines. EA provided clinical data and patient samples. JC performed transcriptomic analysis interpretation. JAV and JS wrote the manuscript. All authors reviewed and approved the final manuscript.

Funding

No external funding was received for this study.

Data availability

The datasets generated and/or analysed during the current study are available in the NCBI Gene Expression Omnibus (GEO) repository, under accession number GSE317889 (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE317889).

Declarations

Ethics approval and consent to participate

This study was approved by the Montpellier University Hospital Ethics Committee (IRB/CPP No. 2023-CCA-01). All patients provided informed consent for molecular testing in accordance with the Declaration of Helsinki.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.