Comparative analysis of genomic profiles and clinical outcomes in cholangiocarcinoma and gallbladder cancer

Abstract

Cholangiocarcinoma (CCA) and gallbladder cancer (GBC) exhibit distinct biological behaviors and treatment responses, potentially driven by differences in their genomic landscapes. However, genomic investigations of biliary tract carcinoma (BTC) remain limited. A retrospective analysis was performed on next-generation sequencing data from 258 patients with BTC (188 with CCA, 70 with GBC), with an external cohort from Memorial Sloan-Kettering Cancer Center (MSKCC cohort) included for comparison. The most frequently altered genes in BTC included TP53 (56.6%), KRAS (27.9%), CDKN2A (21.3%), TERT (17.8%), and MCL1 (14.3%). CCA was predominantly characterized by mutations in ARID1A (15.4% vs. 5.7%, P = 0.04) and PBRM1 (10.1% vs. 0%, P = 0.003), as well as MCL1 amplifications (17.0% vs. 5.7%, P = 0.03), while GBC showed higher frequencies of TP53 alterations (78.6% vs. 47.9%, P < 0.001), ARID2 (15.7% vs. 3.7%, P = 0.002), TERT (24.3% vs. 11.2%, P = 0.02), CCNE1 amplifications (21.4% vs. 2.7%, P < 0.001), and SOX2 amplifications (4.3% vs. 0%, P = 0.02). GBC also exhibited an enrichment of alterations in the DNA damage response (DDR) (84.3% vs. 71.3%, P = 0.04) and p53 signaling (88.6% vs. 59.6%, P < 0.001) pathways. Additionally, 39.9% of patients with BTC harbored at least one actionable genomic alteration. Germline variants were detected in 7.0% (18/258) of patients, with the majority occurring in DDR (61.1%) and p53 (22.2%) pathways. Comparisons with the MSKCC cohort revealed similar genomic features and alterations. Patients with CCA exhibited significantly longer overall survival (OS) than patients with GBC (median OS, 41.2 vs. 24.3 months, P = 0.003). In the MSKCC cohort, high tumor mutational burden (TMB-H_med) correlated with poorer OS (HR = 1.43, P = 0.01), while PBRM1 mutations were associated with improved survival (HR = 0.50, P = 0.02). This study underscores the distinct genomic profiles of GBC and CCA, offering valuable insights into the molecular underpinnings of these aggressive cancers and supporting the development of precision medicine strategies.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-026-56230-0.

Introduction

Biliary tract carcinoma (BTC) is an aggressive malignancy originating from the epithelial cells lining the bile ducts, encompassing cholangiocarcinoma (CCA) and gallbladder cancer (GBC). In recent years, both the incidence and mortality rates of BTC have risen significantly^1,2. Due to the lack of early symptoms, most patients are diagnosed at advanced stages³, resulting in a generally poor prognosis, with a 5-year survival rate of less than 10% for those with unresectable BTC^4,5. Although non-radical surgeries and traditional systemic therapies can offer life-prolonging benefits, the overall prognosis for patients with advanced BTC treated with conventional regimens remains dismal, with no substantial survival improvements observed in recent years⁶. Consequently, there is an urgent need for novel therapeutic strategies. Previous molecular profiling studies utilizing whole-exome sequencing (WES) and targeted panel sequencing have identified recurrent genomic alterations and distinct genomic landscapes across BTC subtypes^7,8. Specifically, mutations in KRAS, ARID1A, SMAD4, and PBRM1 are prevalent in CCA⁹, while alterations in TP53, SMAD4, NOTCH1, ERBB2, and PIK3CA are more common in GBC¹⁰. Additionally, multiple actionable genetic alterations have been identified, including IDH1 mutations, FGFR2 rearrangements, ERBB2 amplification, PIK3CA mutations, NRAS mutations, as well as low-frequency BRCA1/2 mutations, BRAF V600E point mutation, and MET amplification¹¹.

Emerging molecular-targeted therapies offer promising strategies for treating refractory BTC¹². For instance, the phase 2 FIGHT-202 study, a multicenter, single-arm trial, evaluated the FGFR inhibitor pemigatinib in patients with advanced CCA harboring FGFR2 rearrangements who had previously undergone treatment. This study reported an objective response rate (ORR) of 35.5%, with a median progression-free survival (PFS) of 6.9 months (95% CI: 6.2–9.6) and a median overall survival (OS) of 21.1 months (95% CI: 14.8–NR)¹³. Additionally, the ClarIDHy study, a phase 3, multicenter, randomized controlled trial involving 185 patients with intrahepatic CCA who were resistant to prior therapies, demonstrated the efficacy of ivosidenib, an isocitrate dehydrogenase 1 (IDH1) inhibitor. Treatment with ivosidenib of cases harboring IDH1 mutations resulted in improved PFS compared to placebo, with a median PFS of 2.7 months versus 1.4 months, respectively¹⁴. Overall, targeted therapies showed response or clinical benefit in 64% of treated patients¹¹. Consequently, further exploration of actionable genetic alterations, as well as the identification of prognostic and predictive biomarkers, is critical to enhancing treatment efficacy.

This study utilized high-throughput sequencing and a clinico-genomic database comprising 258 patients with BTC to characterize the genomic profiles of BTC subtypes. Additionally, an external cohort of 410 patients was leveraged to validate our findings and explore potential molecular correlates of survival outcomes.

Materials and methods

Patients and samples

This study included a total of 258 patients with BTC from the clinico-genomic database of Nanjing Geneseeq Technology Inc. Among these, 188 patients were diagnosed with CCA and 70 with GBC. All patients underwent targeted next-generation sequencing using a 425-gene panel (GENESEEQPRIME^®→, Nanjing Geneseeq Technology Inc.) on baseline tumor tissues between June 2016 and June 2021 (Fig. 1). In this retrospective study, we prioritized data homogeneity by strictly selecting primary tumor sequencing results for all 258 patients. From our initial database, patients with only metastatic site sequencing data were excluded. For patients with Stage IV disease, only those with available primary tumor genomic profiles were included in this analysis to ensure a consistent comparison across BTC subtypes without confounding from varied metastatic microenvironments. The details of tumor DNA profiling were provided in the Supplementary Methods. Clinical data, including age at diagnosis, family and personal cancer history, and disease stages, were extracted from the medical records. The study adhered to the standards outlined in the Declaration of Helsinki and was approved by the Medical Ethics Committee of Nanjing Geneseeq Medical Laboratory (NSJB-MEC-2022-02).

Fig. 1.

Study flowchart.

Memorial Sloan Kettering cohort for validation and survival analysis

To validate the results of this study and perform survival analysis, we obtained two publicly available Memorial Sloan-Kettering Cancer Center (MSKCC) datasets^11,15 from the cBioPortal for Cancer Genomics (https://www.cbioportal.org). The initial combined dataset included 437 tumor samples from 424 patients diagnosed with BTC. The following screening and selection criteria were applied: first, we excluded 14 samples with no detectable somatic mutations; second, for patients with multiple sequenced samples, we retained only the primary tumor sample (sample type = “primary”) to ensure consistency, resulting in the exclusion of 13 non-primary tumor samples. After applying these filters, a final cohort of 410 patients (182 CCA and 228 GBC, Table S1) was defined for comparative genomic analysis. For survival analysis, we selected a subset of 280 patients (182 CCA and 98 GBC, Table S2) from the final cohort for whom OS data were available. For each patient, we extracted comprehensive genomic and clinical data from cBioPortal. Genomic data included somatic mutations, copy number variations (CNVs), and gene fusions, while clinical data included age, sex, and OS information. All data were de-identified and analyzed in accordance with cBioPortal’s data usage policies.

TMB calculation

Tumor mutational burden (TMB) was evaluated using the GENESEEQPRIME^® 425-gene panel, which encompasses a targeted region coverage of 1.5 Mb. TMB was calculated by summing all single-nucleotide variants (SNVs) and insertions/deletions (indels) within the coding region of the targeted genes, including synonymous alterations and excluding known driver mutations, as previously described¹⁶. The panel-derived TMB calculation demonstrated high analytical concordance with whole-exome sequencing (Pearson r = 0.93). To accommodate different analytical objectives, patients were stratified using two distinct classification schemes. For the analysis of potential therapeutic targets, a well-established clinical cut-off reflecting immunotherapy efficacy was applied¹⁷, categorizing patients into clinical TMB-high (TMB-H_clin, ≥ 10 mut/Mb) and clinical TMB-low (TMB-L_clin, < 10 mut/Mb) groups. This specific 10 mut/Mb threshold has been rigorously established and prospectively validated using WES-confirmed clinical samples. Separately, for prognostic survival analysis and the profiling of genomic characteristics, patients were stratified based on their respective cohort median TMB values, defining median TMB-high (TMB-H_med, ≥ median) and median TMB-low (TMB-L_med, < median) cohorts.

Microsatellite instability analysis

The GENESEEQPRIME^® panel covered 52 microsatellite sites, including classic microsatellite instability (MSI) sites such as BAT-25, BAT-26, NR-21, NR-24, and MONO-27. A locus was considered valid for analysis strictly if it achieved a minimum sequencing coverage depth of 101x. The stability of these sites was assessed using a validated bioinformatic algorithm under CLIA/CAP-accredited laboratory protocols, employing a tumor-normal comparison approach. The stability of these MSI sites was assessed and compiled into an overall MSI score. A sample was classified as MSI if > 40% of the qualified sites exhibited instability; otherwise, it was categorized as microsatellite stable (MSS). This > 40% classification threshold and the bioinformatic pipeline have been analytically validated against orthogonal immunohistochemistry (IHC) testing, demonstrating 97.5% positive and 90.4% negative agreement¹⁸, and have been well-established in previous cohorts^19,20.

Mutational Signature

Text files created in the Mutation Annotation Format (MAF), containing both synonymous and non-synonymous somatic mutations in the coding region, were further processed using the “maftools” and “sigminer” packages in R (version 4.1.3)²¹. The 30 mutational signatures available on the COSMIC website (https://cancer.sanger.ac.uk/signatures/signatures_v2/) were classified into ten groups based on their annotations, including age (COSMIC1), APOBEC (COSMIC2 and COSMIC13), BRCA (COSMIC3), smoking (COSMIC4), DNA mismatch repair deficiency (COSMIC6, COSMIC15, COSMIC20, and COSMIC26), ultraviolet (COSMIC7), POLE (COSMIC10), temozolomide (COSMIC11), immunoglobulin (COSMIC9), and others (the remaining signatures). For each patient, the contribution of each mutational signature was calculated as the proportion of the selected signature relative to the total signatures, as previously described^22,23.

Statistical analysis

For group comparisons, Fisher’s exact tests were used for categorical data, while the Mann-Whitney U test was applied to non-normally distributed continuous variables. A two-sided P-value below 0.05 was considered statistically significant. P values were adjusted for multiple comparisons using Bonferroni’s post hoc test or the false discovery rate (FDR) adjustment method, as appropriate²⁴. FDR is a statistical method that controls the expected proportion of false positives among significant results, which is especially crucial in genomic analyses involving large-scale comparisons^25,26. This method ensures a balanced trade-off between discovery and reliability in high-throughput data contexts. Adjusted P-values < 0.05 were considered significant. To evaluate the potential impact of missing clinical information, sensitivity analyses were performed by comparing the clinicopathological and genomic characteristics of patients with and without available TNM staging data, and by repeating the primary comparative analyses within the staged sub-cohort. For survival analyses, Kaplan–Meier curves were compared using the log-rank test, and hazard ratios (HRs) were estimated by Cox proportional hazards models. When survival curves crossed and the log-rank test yielded a P value < 0.05, a two-stage test was additionally performed to assess the survival differences. All statistical analyses were performed using R software (version 4.1.2).

Results

Patient characteristics

The baseline characteristics of the study cohort are summarized in Table 1. The median age at diagnosis for the overall BTC cohort was 61 years (range 23–87), and 49.6% of the patients were male. Among the 137 patients with sufficient tumor, node, metastasis (TNM) stage data, 65.7% (n = 90) were diagnosed at stage IV. Additionally, 40.5% (n = 85) of the 210 patients assessed for a family history of cancer reported a family history, with 5.7% (n = 12) indicating a family history of BTC or hepatocellular carcinoma (HCC). The cohort included 70 patients with GBC and 188 with CCA. The GBC subgroup had a significantly higher proportion of female patients compared to the CCA subgroup (67.1% vs. 44.1%, P = 0.001). Within the GBC subgroup, females outnumbered males across all age groups (Fig. S1). In contrast, in the CCA subgroup, males outnumbered females in all age groups, except for the 30–39 age group (Fig. S1). The GBC subgroup also had a higher proportion of patients aged over 60 compared to the CCA subgroup (62.9% vs. 48.9%, P = 0.05) (Table 1). A trend toward a higher proportion of family history of BTC/HCC was observed in the GBC subgroup compared to the CCA subgroup (10.9% vs. 3.9%, P = 0.08) (Table 1).

Table 1.

Clinical characteristics of patients MSI, microsatellite instability; MSS, microsatellite stability; BTC, biliary tract carcinoma; HCC, hepatocellular carcinoma.

Characteristics	Biliary tract carcinoma	Cholangiocarcinoma	Gallbladder cancer	P value
	(n = 258)	(n = 188)	(n = 70)
Age at diagnosis, years				0.05
> 60	136(52.7%)	92(48.9%)	44(62.9%)
≤ 60	122(47.3%)	96(51.1%)	26(37.1%)
Median (Range)	61(23–87)	59(23–81)	62(37–87)
Sex, No.(%)				0.001
Male	128(49.6%)	105(55.9%)	23(32.9%)
Female	130(50.4%)	83(44.1%)	47(67.1%)
TNM stage, No.(%)				0.68
I–III	47(34.3%)	36(35.6%)	11(30.6%)
IV	90(65.7%)	65(64.4%)	25(69.4%)
Unknown	121	87	34
Microsatellite instability status, No.(%)				0.30
MSS	247(95.7%)	178(94.7%)	69(98.6%)
MSI	11(4.3%)	10(5.3%)	1(1.4%)
Family history of cancer, No.(%)				0.75
No	125(59.5%)	91(58.7%)	34(61.8%)
Yes	85(40.5%)	64(41.3%)	21(38.2%)
Unknown	48	33	15
Family history of BTC/HCC, No.(%)				0.08
No	198(94.3%)	149(96.1%)	49(89.1%)
Yes	12(5.7%)	6(3.9%)	6(10.9%)
Unknown	48	33	15

MSI, microsatellite instability; MSS, microsatellite stability; BTC, biliary tract carcinoma; HCC, hepatocellular carcinoma.

Somatic genomic landscape of gallbladder cancer and cholangiocarcinoma

The genomic profiles of 258 patients with BTC revealed that the most frequently mutated genes included TP53 (56.6%), KRAS (27.9%), CDKN2A (21.3%), TERT (17.8%), and MCL1 (14.3%) (Fig. 2A). GBC showed significant enrichment in TP53 (78.6% vs. 47.9%, P < 0.001), ARID2 (15.7% vs. 3.7%, P = 0.002), TERT (24.3% vs. 11.2%, P = 0.02), BLM (5.7% vs. 0.5%, P = 0.02), and STK11 (12.9% vs. 4.3%, P = 0.02) mutations, along with amplifications of CCNE1 (21.4% vs. 2.7%, P < 0.001) and SOX2 (4.3% vs. 0%, P = 0.02) compared to CCA (Fig. 2B). In contrast, ARID1A mutations (15.4% vs. 5.7%, P = 0.04), PBRM1 mutations (10.1% vs. 0%, P = 0.003), and MCL1 amplifications (17.0% vs. 5.7%, P = 0.03) were more frequently observed in CCA (Fig. 2B). The distinct subtype enrichments, including alterations in ARID1A, PBRM1, CCNE1, as well as the DDR and p53 signaling pathways, all remained statistically significant (FDR-adjusted P < 0.05), confirming robust genomic divergence (Table S3).

Fig. 2.

Comparative landscape of somatic alterations in CCA and GBC. (A) Oncoprint illustrating the most frequently mutated genes and clinical features across 258 patients with BTC. Each vertical column denotes an individual case. The upper bar chart quantifies the number of alterations per patient. The top heatmap presents sex, cancer subtype, and age distributions. The lower panel depicts mutation frequencies of key driver genes. (B) Forest plot depicting associations between clinical characteristics, somatic alterations, and cancer subtypes within the Chinese cohort. (C) Distribution of tumor mutation burden in CCA compared to GBC.

The median TMB for the entire cohort was 4.47 muts/Mb (range 0-191.7 muts/Mb), with 15.5% (40/258) of samples classified as TMB-H. A trend towards higher TMB in patients with GBC compared to patients with CCA was observed (median, 5.24 vs. 4.33 muts/Mb; P = 0.06) (Fig. 2C), with comparable proportions of TMB-H patients between the two groups (18.6% vs. 14.4%, P = 0.44) (Table 2). To determine if TMB was influenced by clinical stage, we compared TMB levels between stage I-III and stage IV patients in the subset with available staging data (n = 137). No significant differences were observed in the overall cohort (P = 0.1224), nor within the CCA (P = 0.1775) and GBC (P = 0.5138) subgroups (Fig. S2). MSI was present in 4.3% (11/258) of patients (Table 1), with no significant difference in the proportion of MSI between GBC and CCA (5.3% vs. 1.4%, P = 0.30). Notably, we observed an extreme TMB outlier (191.7 mut/Mb) in a 53-year-old female patient with stage IV CCA. Although her tumor was MSS, genomic profiling revealed the presence of multiple mutations within the DDR pathway, including a nonsense mutation in ATR (p.W2094*) and missense mutations in ATM (p.P2648L), ATR (p.G807A), ATRX (p.D514N), PALB2 (p.G1068R), PMS1 (p.S251N). In light of previous studies suggesting that alterations in DDR genes may be associated with elevated mutational burdens across various cancers²⁷, it is possible that this extensive DDR pathway involvement contributed to the hypermutated phenotype observed in this case.

Table 2.

Actionable genetic alterations in Chinese biliary cancer patients.

Actionable genetic alteration	BTC(n = 258)		CCA(n = 188)		GBC(n = 70)		P value	Validated in MSK Cohort (BTC)
	No.	Percent	No.	Percent	No.	Percent		No.	Percent
IDH1/2 mutation	14	5.40%	14	7.40%	0	0.00%	0.01	51	12.44%
FGFR1 ~ 4 fusion/mutation	12	4.70%	11	5.90%	1	1.40%	0.19	21	5.12%
ERBB2 amplification	16	6.20%	6	3.20%	10	14.30%	< 0.01	31	7.56%
KRAS G12C	6	2.30%	5	2.70%	1	1.40%	> 0.99	3	0.73%
PIK3CA mutation	14	5.40%	4	2.10%	10	14.30%	< 0.001	26	6.34%
MET amplification/skipping	9	3.50%	8	4.30%	1	1.40%	0.45	6	1.46%
BRCA1/2 mutation	13	5.00%	7	3.70%	6	8.60%	0.12	16	3.90%
TMB-H	40	15.50%	27	14.40%	13	18.60%	0.44	36	8.78%
MSI-H	11	4.30%	10	5.30%	1	1.40%	0.3	9	2.20%
Total	103	39.90%	71	37.80%	32	45.70%	0.26	163	39.76%

BTC, biliary tract carcinoma; CCA, cholangiocarcinoma; GBC, gallbladder cancer; TMB-H, high tumor mutational burden; MSI, microsatellite instability.

Co-occurrence and mutual exclusivity patterns of the most frequently altered genes and clinical characteristics were subsequently analyzed. Among the entire BTC cohort, female patients showed a higher likelihood of carrying alterations in PIK3CA. TP53 alterations frequently co-occurred with mutations in TERT and CCNE1, while exhibiting mutual exclusivity with NRAS mutations. As expected, patients with TP53 and TERT alterations were more likely to present with TMB-H. Additional co-occurrences included SMAD2 mutations with MEN1 alterations and IDH1/2 mutations with PBRM1 alterations (Fig. S3A). In the GBC subtype, alterations in TP53, ARID2, CCNE1, SMAD2, and ERBB2 were more common, and these were all significantly associated with female sex (all FDR adjusted P < 0.05; Fig. S3A). Mutual exclusivity with PBRM1 was also observed (FDR adjusted P < 0.05). TERT alterations co-occurred with ARID2 mutations (Fig. S3A and B). In the CCA group, only the association between TP53 alterations and TMB-H was observed (Fig. S3C). Pathway-level analysis revealed enrichment of genomic abnormalities in the DDR (84.3% vs. 71.3%, P = 0.04) and p53 signaling (88.6% vs. 59.6%, P < 0.001) pathways in patients with GBC compared to those with CCA (Fig. S3D).

Analysis of the single base substitution (SBS) spectrum revealed enrichment of T[C > T]A and T[C > T]G substitutions in the GBC group, while C[C > T]G and G[C > T]G substitutions were more common in the CCA group (Fig. S4A). A further comparative analysis of the contribution of ten mutational signature groups between the CCA and GBC groups revealed a notably higher contribution of the APOBEC mutational signature in the GBC group compared to the CCA group (P = 0.03, Fig. S4B).

To ensure that these observed genomic differences were not biased by incomplete clinical information, we compared the clinicopathological and genomic features of staged and unstaged patients. No features showed significant differences after adjusting for multiple comparisons (all FDR-adjusted P > 0.05, Table S4). Furthermore, a restricted analysis within the staged subset yielded genomic landscapes and pathway enrichments highly consistent with those of the overall cohort (Fig. S5), confirming the robustness of our findings across the entire study population.

Characterization of actionable alterations in BTC

Actionable alterations were defined in this study according to OncoKB Levels of Evidence V2 (12/20/2019), referring to genomic alterations associated with FDA-approved targeted therapies. In the entire cohort, 39.9% of patients harbored at least one actionable alteration, including driver alterations in IDH1/2 (14, 5.4%), PIK3CA (14, 5.4%), and BRCA1/2 (13, 5.0%), ERBB2 amplifications (16, 6.2%), fusions or mutations in FGFR1-4 (12, 4.7%), MET exon 14 skipping or amplifications (9, 3.5%), KRAS G12C mutations (6, 2.3%), TMB-H (40, 15.5%), and MSI (11, 4.3%) (Table 2). Specifically, FGFR alterations were predominantly found in patients with CCA (91.7%, 11/12), with only one case identified in GBC (Table 2). FGFR3-TACC3 fusions were detected in four patients, and FGFR2 fusions in seven patients with diverse partner genes, including WAC, DAAM1, CTNNA3, BICC1, HOOK1, RBM20, and AHCYL1 (Fig. 3A). Breakpoints in FGFR2 were exclusively found in intron 17, whereas FGFR3 breakpoints were distributed across intron 2, intron 17, and exon 18 (FIg. S6A). Mutations in FGFR1 included p.P150S, p.N330H, and p.N65S (Fig. S6B), while FGFR2 mutations included p.S110A, p.S224C, p.D225Gfs*4, c.1439+936_1555delinsC, and p.A567V (Fig. S6C). FGFR3 mutations were all p.V329I (Fig. S6D), and FGFR4 mutations consisted of p.E212K, p.R401P, and p.A614T (Fig. S6E). IDH1/2 mutations were detected exclusively in the CCA group (7.4% vs. 0%, P = 0.01) (Table 2), with the prevalent mutation hotspots being p.R132X for IDH1 and p.R172X for IDH2 (Fig. 3B, C). ERBB2 amplification (14.3% vs. 3.2%, P = 0.002) and PIK3CA mutations (14.3% vs. 2.1%, P < 0.001) were enriched in GBC. The most common PIK3CA mutation sites were p.E545X and p.E542X, with frequencies of 25% and 21%, respectively (Fig. 3D). KRAS mutations were detected in 26.0% (67/258) of the patients, with the G12X mutation accounting for 76% (51/67) of KRAS mutations (Fig. 3E). In the entire cohort, 2.3% (6/258) of patients carried the actionable KRAS p.G12C alteration (Table 2). Among patients with CCA, 27 (14.4%) were classified as TMB-H, and 10 (5.3%) exhibited MSI (Table 2; Fig. 3F). In the GBC cohort, 13 (18.6%) patients were identified as TMB-H, and only one (1.4%) patient displayed MSI (Fig. 3F).

Fig. 3.

Landscape of actionable genomic alterations in Chinese patients with BTC. (A) Fusion partner genes identified in FGFR2 and FGFR3 fusions. (B) Lollipop plot for IDH1 mutations. (C) Lollipop plot for IDH2 mutations. (D) Distribution of PIK3CA mutations across BTC, CCA, and GBC. (E) Distribution of KRAS mutations in BTC, CCA, and GBC. (F) Prevalence of TMB-H_clin and MSI in CCA versus GBC. Abbreviations: BTC, biliary tract carcinoma; CCA, cholangiocarcinoma; GBC, gallbladder cancer; TMB-H, high tumor mutation burden; MSI, microsatellite instability.

Characterization of germline variants in BTC

A total of 20 pathogenic/likely pathogenic germline variants were identified in 18 (7.0%) patients with BTC, including 11 CCA (5.9%) and 7 GBC (10.0%) patients. The median age at diagnosis was 60 years (range 34–83). Among these 18 patients, nine reported a family history of various cancers, including breast, liver, lung, esophageal, rectal, pancreatic, brain, and throat malignancies (Table 3). The majority (61.1%, 11/18) carried variants in genes associated with the DDR signaling pathway, while 22.2% (4/18) had germline variants in p53 pathway genes. Additionally, two patients were found to have germline variants in the PI3K and RTK/RAS signaling pathways, respectively (Table 3). Patient P121, a male with CCA, carried two germline variants in BRCA2 (c.8954-5_8954-2delAACA and p.W2626*). His family history included lung cancer in his mother, rectal cancer in his father, and pancreatic cancer in his brother. Patient P222, a male with GBC, carried germline variants in both ATM (p.Q3037_Q3038delinsH*) and ERCC5 (p.Q233*), with a paternal history of esophageal cancer. The remaining 16 patients carried a single germline variant each (Table 3).

Table 3.

Eighteen samples with pathogenic/likely pathogenic germline variants.

ID	Cancer type	Age	Family history	Gene	AAChange		Variant_Classification
P007	CCA	34	Yes(Mother, breast cancer)	BLM	c.320dupT	p.L107Ffs*36	Frame_Shift_Ins
P012	CCA	39	Yes(Grandmother, liver cancer)	RAD50	c.2165dupA	p.E723Gfs*5	Frame_Shift_Del
P023	CCA	44	No	XPC	c.2722 C > T	p.R908*	Nonsense_Mutation
P046	CCA	52	No	MSH2	c.1 A > G	p.M1?	Translation_Start_Site
P078	CCA	57	Yes(Father, lung cancer; brother, esophageal cancer)	PALB2	c.448 C > T	p.Q150*	Nonsense_Mutation
P085	CCA	58	No	RAD50	c.2420delG	p.R807Kfs*13	Frame_Shift_Del
P121	CCA	63	Yes(Mother, lung cancer; Father, rectal cancer; brother, pancreatic cancer)	BRCA2	c.8954-5_8954-2delAACA		Splice_Region
				BRCA2	c.7878G > A	p.W2626*	Nonsense_Mutation
P127	CCA	64	No	RECQL4	c.1929_2058 + 47del		Splice_Site
P151	CCA	67	No	ATM	c.7921delC	p.Q2641Rfs*5	Frame_Shift_Del
P177	CCA	72	NA	SDHA	c.1054 C > T	p.R352*	Nonsense_Mutation
P180	CCA	73	Yes(Sister, brain cancer)	ERCC5	c.468-8_469delACTTTCAGTT		Splice_Site
P192	GBC	39	Yes(aunt, lung cancer)	TSC2	c.5061_5069-47dup		Nonsense_Mutation
P203	GBC	53	NA	XPC	c.2280 C > A	p.Y760*	Nonsense_Mutation
P204	GBC	53	Yes(Mother, lung cancer)	ALK	c.4765 C > T	p.Q1589*	Nonsense_Mutation
P222	GBC	62	Yes(Father, esophageal cancer)	ERCC5	c.697 C > T	p.Q233*	Nonsense_Mutation
				ATM	c.9111_9112delGCinsTT	p.Q3037_Q3038delinsH*	Nonsense_Mutation
P227	GBC	63	Yes(Mother, breast cancer; Father, liver cancer; brother, throat cancer)	BRCA2	c.3860dupA	p.N1287Kfs*2	Frame_Shift_Ins
P255	GBC	82	No	TP53	c.818G > A	p.R273H	Missense_Mutation
P256	GBC	83	No	ATM	c.1402_1403delAA	p.K468Efs*18	Frame_Shift_Del

CCA, cholangiocarcinoma; GBC, gallbladder ca.

Analysis of clinico-genomic features using an external cohort

To validate our findings, a cohort of 410 patients with BTC (182 CCA and 228 GBC) from the MSK-IMPACT dataset was analyzed. The clinical characteristics of the MSKCC cohort are summarized in Table S1, and no significant differences were observed between our cohort and the MSKCC cohort. At the genetic level, a significant difference was observed in the distribution of CNVs in CCNE1 (1.1% vs. 9.6%, P < 0.001) (Fig. S7) between CCA and GBC, which aligned with the findings in our study cohort. The differential distribution of mutations in TP53, ARID2, PBRM1, TERT, BLM, STK11, PIK3CA, IDH1, IDH2, ERBB2, NRAS, FGFR1-4, and POLE, as observed in our cohort, was also recapitulated in the MSK cohort (Figs. 2B and S7). Our analysis revealed distinct mutational patterns between GBC and CCA in our cohort compared to the MSKCC cohort. A higher prevalence of SOX2 CNVs was observed in patients with GBC compared to patients with CCA in our cohort (4.3% vs. 0%, P = 0.02), though this difference was not observed in the MSKCC cohort (0% vs. 0.5%, P = 0.44). Similarly, a higher frequency of MCL1 CNVs was observed in CCA compared to patients with GBC (17.0% vs. 5.7%, P = 0.03) in our cohort, whereas no significant difference was observed in the MSKCC cohort (3.8% vs. 1.3%, P = 0.12). Mutations in SMAD2, MEN1, IL7R, and KRAS were more common in GBC than in CCA in our cohort, whereas no difference was observed in the MSKCC cohort. Additionally, higher frequencies of ARID1A and KRAS mutations were observed in CCA compared to GBC in our cohort, but no such difference was observed in the MSKCC cohort (Figs. 2C and S7). At the pathway level, the differential involvement of genetic alterations in the DDR and p53 signaling pathways between CCA and GBC, as observed in our cohort, was corroborated by similar findings in the MSKCC cohort (Figs. S3D, S8A). Furthermore, the increased prevalence of T[C > T]A substitutions in GBC, along with a higher contribution of the APOBEC mutational signature, were confirmed in the MSKCC cohort (Fig. S4A, B; Fig. S8B, C).

Consistent with our findings, the MSKCC cohort also showed a correlation between GBC and female sex, as well as alterations in TP53, ERBB2, and CCNE1 (Fig. S9). Additionally, TP53 mutations frequently co-occurred with CCNE1 mutations and were associated with TMB-H. Co-occurrence of PBRM1 mutations with IDH1/2 mutations was also observed (Fig. S9). In contrast to our cohort, the MSKCC cohort exhibited an association of GBC with high TMB and mutual exclusivity between FGFR2 and IDH1/2 alterations (Fig. S9).

Association of prognosis with clinical characteristics and genomic alterations

In the MSKCC cohort, survival data were available for 280 patients with BTC (182 CCA, 98 GBC). Their clinical characteristics and genetic profiles were comparable to those of our cohort (Table S2). The cohort’s median OS was 35.1 months (95% CI: 26.7–45.0). Univariate analysis results, detailing associations between clinical features, genetic variations, and OS, are presented in Table S5. Patients with CCA had a significantly longer OS than patients with GBC (mOS: 41.2 vs. 24.3 months, HR [95% CI] = 0.63 [0.46–0.85], P = 0.003, Fig. 4A). Additionally, the presence of SMAD4 mutations (mOS: 17.7 vs. 41.2 months, HR [95% CI] = 2.02 [1.4–2.91], P < 0.001, Fig. 4B) or MYC amplification (mOS: 17.6 vs. 36.2 months, HR [95% CI] = 2.39 [1.3–4.42], P = 0.004, Fig. 4C) was linked to poorer survival outcomes. Conversely, PBRM1 mutations were associated with improved OS (mOS: 72.3 vs. 32.1 months, HR [95% CI] = 0.5 [0.27–0.92], P = 0.02, Fig. 4D). Moreover, patients with TMB-L showed better OS (mOS: 41.2 vs. 25.8 months, HR [95% CI] = 1.43 [1.07–1.91], P = 0.01, Fig. 4E). Median OS was numerically higher in IDH1/2-mutated patients compared to wild-type (46.8 vs. 32.3 months, HR [95% CI] = 0.8 [0.54–1.19], P = 0.27, Fig. S10). Multivariate analysis, adjusting for BTC subtypes and genetic alterations, identified SMAD4, and MYC as independent prognostic factors for the entire cohort (Table S6). In both CCA and GBC subgroups, SMAD4 and MYC alterations were consistently associated with poor outcomes (Fig. S11A, B). Low TMB correlated with prolonged OS in the CCA subgroup (mOS: 52.8 vs. 25.8 months, HR [95% CI] = 1.62 [1.14–2.31], P = 0.007), but not in the GBC subgroup (mOS: 35.1 vs. 22.5 months, HR [95% CI] = 0.83 [0.50–1.37], P = 0.47, Fig. S11C). Similarly, PBRM1 mutations shows a trend toward favorable outcomes in patients with CCA (mOS: 72.3 vs. 37.9 months, HR [95% CI] = 0.53 [0.27–1.05], P = 0.07, Fig. S11D), while no significant OS difference was observed in GBC individuals with or without PBRM1 mutations (mOS: 57.8 vs. 23.4 months, HR [95% CI] = 0.45 [0.11–1.87], P = 0.26, Fig. S11D). Patients with both TMB-H and PBRM1 wild-type were classified as “double hit,” with the remaining patients categorized as “others.” Among patients with CCA, the “double hit” group had significantly worse OS compared to the “others” group (mOS: 21.8 vs. 57.2 months, HR [95% CI] = 1.78 [1.25–2.54], P = 0.001, Fig. 4F). In contrast, no significant OS difference was observed in patients with GBC between the “double hit” and “others” groups (mOS: 25.3 vs. 23.4 months, HR [95% CI] = 0.96 [0.59–1.58], P = 0.88, Fig. 4F). Lastly, the effect of targetable alterations on OS was assessed. In the CCA subgroup, OS did not significantly differ between IDH1/2-mutant and wild-type patients (mOS: 46.8 vs. 40.0 months, HR [95% CI] = 0.94 [0.62–1.43], P = 0.78, Fig. S11E).

Fig. 4.

Prognostic impact of cancer subtype and genomic alterations on overall survival (OS). Kaplan-Meier survival curves from the MSKCC cohort: (A) stratified by cancer subtype (CCA vs. GBC); (B) by SMAD4 mutation status; (C) by MYC amplification; (D) by PBRM1 mutation; (E) by median TMB; (F) by subgroup classification into “double hit” (TMB-H_med and PBRM1 wild-type) versus others.

Discussion

This study provides an in-depth analysis of the genomic profiles of Chinese patients with BTC, distinguishing between CCA and GBC. A key strength of this study is the sample size (n = 258), supported by a large external validation cohort (n = 410). The consistency across these datasets enhances the reliability of the genomic signatures identified, providing a robust basis for clinical interpretation in rare and heterogeneous BTC subtypes.

Actionable mutations were identified in 39.9% of patients with BTC, with an additional 19.8% exhibiting high TMB or MSI. These findings are consistent with previous reports, which indicate a 30–50% prevalence of actionable alterations in patients with BTC^11,28–30. PBRM1 mutations were observed in 10% of CCAs, aligning with the MSKCC cohort and prior studies, which report frequencies between 6% and 10%^28,30–32. Although PBRM1 is also frequently mutated in clear cell renal carcinoma (ccRCC) (41% to 45%)^33,34, its prognostic impact appears to be cancer-type dependent. While our cohort showed a favorable trend in CCA, studies in ccRCC have generally linked PBRM1 loss to worse clinical outcomes³⁵. Furthermore, its predictive value remains controversial across different clinical trials³⁴. Therefore, the prognostic significance of PBRM1 in BTC requires further validation in larger independent cohorts.

TMB has been widely recognized as both a prognostic biomarker and a predictive marker for immunotherapy, including immune checkpoint inhibitors, across various cancer types^36–38. However, the clinical significance of TMB remains controversial. For example, one study reported no significant difference in OS between BTC individuals with high TMB and those with non-high TMB³⁹. Conversely, another study involving 1415 patients across multiple cancer types, including 13 BTC cases, identified TMB status as a critical prognostic marker, highlighting its relationship with OS⁴⁰. In our study, we observed that both PBRM1 mutations and low TMB were associated with improved OS. Notably, PBRM1 mutations were more frequent in CCA, which exhibited lower TMB values compared to GBC. This may help explain the superior OS observed in CCA compared to GBC. Nevertheless, the impact of other genetic alterations on OS cannot be disregarded. Therefore, the prognostic and predictive significance of TMB status and PBRM1 mutations in BTC warrants further investigation. Specifically, in our survival analysis of the MSKCC cohort, TMB-H was significantly associated with poorer overall survival in CCA, but not in GBC. While TMB-H is a well-established predictive biomarker for favorable responses to immunotherapy, in immunotherapy-naive patients, it generally reflects severe genomic instability and aggressive tumor biology, thereby serving as a prognostic biomarker for poor clinical outcomes (Valero et al. 2021; Ota et al. 2022). The differential prognostic impact of TMB-H between CCA and GBC further highlights their biological heterogeneity. GBC intrinsically possesses a highly altered genome, characterized by a higher baseline TMB and frequent TP53 and DDR pathway alterations. This inherent genomic instability drives aggressive tumor behavior regardless of the absolute TMB score, which may explain why TMB-H does not further stratify survival in GBC. In contrast, CCA typically features a more stable genome; therefore, the presence of TMB-H in a subset of CCA patients marks a highly divergent and aggressive phenotype, leading to a drastically reduced survival time.

Regarding other actionable alterations, the prognostic impact of IDH1/2 mutations appears inconsistent. In our analysis, the numerical survival trend in the overall BTC cohort was not sustained within the CCA subgroup (P = 0.78), likely reflecting the baseline survival advantage of CCA over GBC. This discrepancy underscores the importance of subtype-specific analysis to avoid confounding by the clinical heterogeneity of BTC. Several IDH inhibitors, including the FDA-approved ivosidenib and enasidenib⁴¹, as well as other inhibitors currently in clinical trials⁴², have shown anti-tumor properties. In our cohort, the frequency of alterations in key oncogenes such as TP53 and KRAS was higher compared to the MSKCC cohort, while alterations in IDH1/2 and FGFR were less prevalent. In both cohorts, an APOBEC-association mutational signature was enriched in GBC. APOBEC enzymes are established sources of cytosine deamination and mutagenesis in human cancers^43,44. However, our data do not identify the upstream drivers of this pattern, and factors such as inflammatory exposure or population context remain speculative^45,46. These findings therefore support the presence of subtype-specific mutagenic processes in GBC that warrant dedicated functional investigation.

Consistent with previous studies linking hereditary variants to BTC⁴⁷, our study identified pathogenic germline variants in a comparable proportion of patients. However, research on the prevalence of germline variants in BTC remains limited, likely due to the disease’s rarity. Variants in genes such as MSH2 and BRCA2 have been associated with an increased risk of BTC⁴⁸. In our cohort, loss-of-function mutations were detected in multiple genes involved in the DDR pathway, including those related to homologous recombination repair, such as BRCA2, MSH2, RAD50, PALB2, BLM, XPC, and ERCC5. This observation underscores the potential value of evaluating homologous recombination deficiency (HRD) in BTC, as tumors with HRD may respond to targeted therapies, such as platinum-based chemotherapies and poly-ADP ribose polymerase (PARP) inhibitors⁴⁹. Although clinical evidence supporting the use of PARP inhibitors in BTC remains limited, early studies suggest favorable responses to PARP inhibitors in BRCA-mutated BTC^50,51, further supporting the potential role of HRD testing in guiding therapeutic decisions. Incorporating germline testing into routine clinical practice could help identify patients with BTC who might benefit from targeted or intensified therapy.

Regarding the selection of biopsy sites, while metastatic biopsies are clinically standard for advanced BTC, we restricted our retrospective analysis to primary tumor samples to ensure a uniform tissue-of-origin across all stages. This controlled approach minimizes the genomic confounding associated with different metastatic sites. Although this might not capture acquired alterations specific to late-stage metastases, the high concordance of truncal drivers in BTC suggests that our primary tumor-focused cohort reliably reflects the core genomic divergence between GBC and CCA.

This study has several limitations. First, the retrospective design restricted our ability to gather complete anatomical data, preventing the differentiation between intrahepatic and extrahepatic subtypes. This nature also resulted in incomplete TNM staging and survival data for a portion of our cohort. While our sensitivity analysis showed that the core genomic findings remained consistent within the staged sub-cohort, the lack of complete clinical data may still limit the depth of survival analysis. Second, technical constraints exist due to the use of targeted panel sequencing. Although robustly calibrated, it inherently restricts the comprehensive detection of rare structural variants or complex fusions compared to whole-exome or whole-genome sequencing. Additionally, panel-derived mutational signature analysis is constrained by targeted genomic coverage and should therefore be interpreted cautiously. Furthermore, the comparative validation against the external MSK cohort introduces potential platform and bioinformatic heterogeneity. Third, our analysis may be influenced by clinical and statistical confounding factors. The cohort included both localized and advanced cases. Furthermore, we lacked detailed records on systemic treatments or prior surgical interventions, the latter of which serve as a major independent prognostic factor in BTC. Without stratification by these treatment modalities, it is difficult to definitively separate the intrinsic prognostic value of mutations from treatment-driven survival benefits. Finally, the statistical power to fully characterize rare genomic events, such as ARID1A and PBRM1 mutations, is limited by the current cohort size, necessitating further validation in larger, multi-center prospective studies. To address these limitations and the biological heterogeneity of BTC, future research should consider integrative multi-omics strategies, incorporating transcriptomics and proteomics to complement the DNA-level profiling presented in this study.

Conclusions

In conclusion, this comprehensive genomic analysis of a large cohort of Chinese patients with BTC revealed molecular differences between GBC and CCA. Potential predictive and prognostic factors that could inform clinical decision-making and facilitate the development of alternative treatment options were identified. These findings contribute to the growing body of evidence supporting personalized therapeutic strategies for this rare and challenging disease, highlighting the promise of precision medicine in improving outcomes for patients with BTC.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

Weiqun Lu: Conceptualization, Investigation, Formal analysis, and Funding acquisition. Rujiao Liu: Investigation. Cuicui Liu: Investigation, Writing – original draft, and Writing – review & editing. Jiaohui Pang: Conceptualization, Investigation, and Formal analysis. Jiani Yin: Investigation, Writing – review & editing. Zhe Zhang: Conceptualization, Methodology, Investigation, Supervision, Data curation, and Writing – review & editing.

Funding

W.L. received a research grant from the Natural Science Foundation General Program of Guangdong Province, China (No. 2021A1515010764).

Data availability

The raw sequencing data generated in the present study are deposited in the Genome Sequence Archive for Human (GSA-Human) repository (accession number HRA011746) and are accessible at https://ngdc.cncb.ac.cn/gsa-human/s/8K8sJU40.

Declarations

Competing interests

Cuicui Liu, Jiaohui Pang, and Jiani Yin are employees of Nanjing Geneseeq Technology Inc. The remaining authors declare no conflicts of interest.

Ethical approval and consent to participate

The study received approval from the Medical Ethics Committee of Geneseeq Medical Laboratory (NSJB-MEC-2022-02). Written informed consent was obtained from all participants prior to sample collection.

Consent for publication

Informed consent form was obtained from each patient.