Whole-Genome and Epigenomic Landscapes of Etiologically Distinct Subtypes of Cholangiocarcinoma

Authors

Apinya Jusakul

Ioana Cutcutache

Chern Han Yong

Jing Quan Lim

Mi Ni Huang

Nisha Padmanabhan

Vishwa Nellore

Sarinya Kongpetch

Alvin Wei Tian Ng

Ley Moy Ng

Su Pin Choo

Swe Swe Myint

Rajasegaran Thanan

Sanjanaa Nagarajan

Weng Khong Lim

Cedric Chuan Young Ng

Arnoud Boot

Mo Liu

Choon Kiat Ong

Vikneswari Rajasegaran

Stefanus Lie

Alvin Soon Tiong Lim

Tse Hui Lim

Jing Tan

Jia Liang Loh

John R. McPherson

Narong Khuntikeo

Vajaraphongsa Bhudhisawasdi

Puangrat Yongvanit

Sopit Wongkham

Yasushi Totoki

Hiromi Nakamura

Yasuhito Arai

Shinichi Yamasaki

Pierce Kah-Hoe Chow

Alexander Yaw Fui Chung

London Lucien Peng Jin Ooi

Kiat Hon Lim

Simona Dima

Dan G. Duda

Irinel Popescu

Philippe Broet

Sun-Whe Kim

Dong Wook Choi

Sung-Hoon Moon

Jin-Young Jang

Yong Bok Yang

Pankaj K. Singh

Sandhya Annaiah

Vinay Tergaonkar

Patrick Tan

Steven G. Rozen

Bin Tean Teh

Tatsuhiro Shibata

Chawalit Pairojkul

Raluca Gordân

Doi

10.1158/2159-8290.CD-17-0368

PMID: 28667006 · DOI: 10.1158/2159-8290.CD-17-0368 · Journal: Cancer Discovery (2017)

TL;DR

Jusakul et al., on behalf of the International Cancer Genome Consortium, profiled 489 cholangiocarcinomas (CCAs) from 10 countries — combining whole-genome (71), exome (200), targeted (188), copy-number (175), DNA methylation (138), and gene-expression (118) platforms — to define four etiology-driven molecular subtypes. Fluke-Positive Clusters 1 and 2 are enriched in ERBB2 amplifications and TP53 mutations; Fluke-Negative Clusters 3 and 4 either exhibit high copy-number alterations with PD-1/PD-L2 immune-checkpoint expression (Cluster 3) or carry chromatin-modifier mutations (IDH1/IDH2, BAP1) with FGFR/PRKA-related rearrangements (Cluster 4). The study nominates four new CCA driver genes (RASA1, STK11, MAP2K4, SF3B1), reports the first FGFR3-TACC3 fusion in CCA, identifies FGFR2 3′UTR deletions as a new mechanism of FGFR2 upregulation, and reveals two epigenetically distinct hypermethylation subtypes (CpG-island vs CpG-shore) reflecting carcinogen-driven vs genetically-driven oncogenesis PMID:28667006.

Cohort & data

  • Total cohort: 489 cholangiocarcinomas (133 Fluke-Positive: 132 Opisthorchis viverrini, 1 Clonorchis sinensis; 356 Fluke-Negative; 39 HBV/HCV-positive; 5 PSC-positive) from 10 countries (Singapore, Romania, Thailand, Italy, France, South Korea, Brazil, Taiwan, China, Japan), all staged AJCC 7th edition, none pretreated PMID:28667006.
  • Anatomical breakdown: intrahepatic (IHCH), perihilar (PHCH), and distal extrahepatic (EHCH) tumors, with anatomical and survival data on 459 samples PMID:28667006.
  • Genomic platforms: whole-genome sequencing on 71 tumor/normal pairs at average 64.2× depth (Illumina HiSeq X10/2500/2000); whole-exome sequencing on 200 cases (previously published); targeted DNA sequencing of 404 genes via SureSelect XT2 capture on 188 cases (HiSeq 4000, 99.6% coding coverage); HumanOmniExpress SNP arrays on 175 cases (Affymetrix-style SNP CN profiling); Illumina 450K methylation BeadChip on 138 cases; HumanHT-12 Expression BeadChip (microarray gene expression, Illumina platform) on 118 cases PMID:28667006.
  • Integrative clustering: iCluster (iClusterPlus) on the 94 samples with all four data types (sSNVs/indels, sCNAs, mRNA, methylation); validated by randomized subsampling and an expanded 121-sample reanalysis with 90% cluster-prediction accuracy PMID:28667006.
  • Validation cohort: newly classified samples plus a published 38-sample US TCGA CCA series for survival reproducibility PMID:28667006.
  • Cell lines: H69 (immortalized cholangiocyte), HEK293T, EGI-1 (DSMZ ACC 385), HUCCT1 (JCRB0425), and M213 (JCRB1557, from the Liver Fluke and Cholangiocarcinoma Research Center) were used for luciferase reporter and shRNA functional assays PMID:28667006.
  • Dataset: chol_icgc_2017. Comparison study: chol_jhu_2013.

Key findings

  • Four molecular CCA clusters from integrative multi-omic clustering. Cluster 1 (mostly Fluke-Pos): CpG-island promoter hypermethylation, enriched in ARID1A (p < 0.01) and BRCA1/BRCA2 (p < 0.05) mutations, high non-synonymous mutation burden, H3K27me3-promoter mutation enrichment. Cluster 2 (mixed): upregulated CTNNB1, WNT5B, AKT1 expression. Clusters 1 and 2 are enriched in TP53 mutations (p < 0.001) and ERBB2 amplifications (p < 0.01), Fisher’s exact test. Cluster 3 (Fluke-Neg): highest sCNA burden, immune-checkpoint upregulation (PDCD1, PDCD1LG2, BTLA). Cluster 4 (Fluke-Neg): BAP1, IDH1/IDH2 mutations, FGFR alterations (all p < 0.01), CpG-shore hypermethylation PMID:28667006.
  • Cluster correlates with anatomy and prognosis. Clusters 1 and 2 are enriched in extrahepatic tumors; Clusters 3 and 4 are almost entirely intrahepatic (p < 0.001). Patients in Clusters 3 and 4 have significantly better overall survival (log-rank p < 0.001), independent of fluke status, anatomical location, and stage (Cox p < 0.05); reproduced in an independent validation cohort PMID:28667006.
  • Mutation burden: average 82 non-silent somatic mutations/tumor (median 47); 64 sSNVs (median 41) and 18 indels (median 6). Three CCAs were hypermutated (5.91 sSNVs/Mb and 24.17 indels/Mb) with MSI signatures and two carrying POLE mutations. Excluding hypermutators, Fluke-Pos CCAs carried significantly more somatic mutations than Fluke-Neg CCAs (median 4,700 vs 3,143/tumor, p < 0.05) PMID:28667006.
  • Four new CCA driver genes (32 SMGs total by MutSigCV and IntOGen, q < 0.1 by both): RASA1 (4.1%, mostly frame-shift/nonsense; shRNA knockdown enhances migration/invasion in CCA cell lines), STK11 (5%, mostly inactivating), MAP2K4 (homozygous deletions in 2 Fluke-Pos cases plus 2.2% mutations, half inactivating), and SF3B1 (4.6%, hotspots at codons 625 and 700, previously seen in uveal melanoma and breast cancer) PMID:28667006.
  • ERBB2 amplification is enriched in Fluke-Pos CCAs (10.4% vs 2.7% in Fluke-Neg, p < 0.01). ERBB2-amplified samples averaged 14 copies (by ASCAT on SNP array or Quandico on sequencing data) and were independently validated by FISH. Activating ERBB2 mutations (S310F/Y, G292R, T862A, D769H, L869R, V842I, G660D) were detected in 9 cases (2%) PMID:28667006.
  • Other recurrent CN events: MYC amplification (n=12), MDM2 (n=9), EGFR (n=11), CCND1 (n=7) amplifications; CDKN2A (n=17), UTY (n=17), KDM5D (n=16) deletions PMID:28667006.
  • Structural variants: CREST called ~93 somatic SVs/tumor (median 69, range 0–395), 91% PCR-validated, mostly intra-chromosomal (65%), associated with ARID1A, CDKN2A/B, TTC28, and fragile site 1q21.3. SV burden varied across clusters (Kruskal–Wallis p < 0.05); TP53, FBXW7, and SMAD4 mutation status was associated with increased SV burden (q < 0.1) PMID:28667006.
  • FGFR rearrangement landscape. Five in-frame fusions with intact tyrosine-kinase domains: FGFR2-STK26, FGFR2-TBC1D1, FGFR2-WAC, FGFR2-BICC1, and FGFR3-TACC3. The FGFR3-TACC3 fusion is the first reported in CCA. All FGFR2 rearrangements occurred exclusively in Cluster 4 (p < 0.001) PMID:28667006.
  • FGFR2 3′UTR loss as a new activation mechanism. Recurrent truncating SVs translocated FGFR2 without its 3′UTR to intergenic regions; FGFR2-truncated CCAs had significantly higher FGFR2 transcript levels (p < 0.01), and luciferase reporter assays in HEK293T and H69 cells confirmed that the intact FGFR2 3′UTR represses expression PMID:28667006.
  • PRKACB (PKA catalytic subunit B) rearrangements. ATP1B1-PRKACB and LINC00261-PRKACB fusions retain the PKA pseudokinase domain and may activate downstream MAPK signaling PMID:28667006.
  • L1 (LINE-1) retrotransposition is recurrent and Fluke-Pos-associated. 52 events in 20/71 (28.2%) of WGS tumors, 98% PCR-validated, predominantly originating from a TTC28 intron-1 L1 element. L1 retrotransposition was enriched in Fluke-Pos tumors (p < 0.01) and correlated with increased SV burden (p < 0.05) PMID:28667006.
  • TERT promoter mutations are rare in CCA. Only 2/71 WGS cases (2.8%) carried TERT promoter mutations (chr5:1295228); no other recurrent non-coding promoter point mutations were identified PMID:28667006.
  • FIREFLY (a new method) identifies non-coding regulatory dysregulation at the gene-set level. Applied to 70 WGS samples and 6,639 mutated promoters, FIREFLY — which integrates protein-binding microarray data for 486 TFs — identified four gene sets enriched for promoter mutations that alter TF binding and produce concordant transcriptional dysregulation. Two of the four sets (MIKKELSEN_MCV6_HCP_WITH_H3K27ME3 and MIKKELSEN_MEF_ICP_WITH_H3K27ME3) are PRC2 targets bearing the H3K27me3 mark, and 3 of 4 sets were preferentially mutated in Cluster 1, supporting a role for PRC2/H3K27me3 dysregulation in that subtype PMID:28667006.
  • Two distinct DNA-methylation subgroups. Unsupervised methylation clustering on 138 CCAs reproduced Cluster 1 (CpG-island hypermethylation, Fluke-Pos) and Cluster 4 (CpG-shore hypermethylation, Fluke-Neg) at 96.3% and 86.1% concordance with the integrative clusters. GSEA showed both target PRC2 pathways but at different genomic features. Promoter methylation was inversely correlated with transcript level in both clusters (q < 0.05) PMID:28667006.
  • Distinct mechanisms drive Cluster 1 vs Cluster 4 hypermethylation. Cluster 1 shows downregulation of the demethylase TET1 and upregulation of histone methyltransferase EZH2. Cluster 4 is enriched in IDH1/IDH2 mutations (31.6% vs 1.0% in other clusters, q < 0.001) and, among IDH-WT Cluster 4 tumors, enriched in BAP1 inactivating point mutations and regional deletions (q < 0.001 and q < 0.05 respectively) PMID:28667006.
  • Mutational signatures: ten established signatures detected — COSMIC Signatures 1, 5, 8, 16, 17; APOBEC (Signatures 2, 13); MMR-deficient (Signatures 6, 20); aristolochic acid (Signature 22). Fluke-Pos CCAs were enriched for APOBEC mutation burden (p < 0.001). Signature 1 (CpG>TpG) was elevated in Cluster 1 even after age adjustment (p < 0.001), and CpG>TpG mutations were preferentially located near hypermethylated regions in Cluster 1 (p < 0.001) but not Cluster 4 — consistent with deamination of methylated cytosines as a Cluster 1 mutational driver PMID:28667006.
  • Cluster 1 is subclonally heterogeneous, Cluster 4 is clonal. VAF-distribution analysis of point mutations (in copy-neutral regions, purity-adjusted) showed wide spread in Cluster 1 indicating heterogeneous subclones, vs tightly clonal structure in Cluster 4 PMID:28667006.
  • Driver-gene–anatomy associations: BAP1 and KRAS were more frequently mutated in intrahepatic CCAs (q < 0.1, Fisher’s exact test), persisting after fluke-status adjustment in multivariate regression PMID:28667006.
  • Immune signal in Cluster 3. ESTIMATE showed immune infiltration in both Clusters 2 and 3, but only Cluster 3 specifically upregulated immune-checkpoint genes (PDCD1, PDCD1LG2, BTLA) and antigen cross-presentation, CD28 co-stimulation, and T-cell signaling pathways PMID:28667006.

Genes & alterations

  • ERBB2 — Amplification in 3.9–8.5% of CCAs, enriched in Fluke-Pos (10.4% vs 2.7%, p < 0.01); average 14 copies; FISH-validated. Activating point mutations (S310F/Y, G292R, T862A, D769H, L869R, V842I, G660D) in 2% of cases. Defining co-feature of Clusters 1 and 2; nominated as anti-HER2 therapeutic target PMID:28667006.
  • TP53 — Significantly enriched in Clusters 1 and 2 (p < 0.001). TP53 mutation associated with increased SV burden (q < 0.1) PMID:28667006.
  • ARID1A — Enriched in Cluster 1 (p < 0.01); also a recurrent SV target PMID:28667006.
  • BRCA1/BRCA2 — Enriched in Cluster 1 (p < 0.05) PMID:28667006.
  • IDH1/IDH2 — Enriched in Cluster 4 (31.6% vs 1.0% in other clusters, q < 0.001); proposed driver of CpG-shore DNA hypermethylation via 2-hydroxyglutarate oncometabolite production. Suggested as candidates for IDH inhibitors (e.g. ivosidenib, then in trial NCT02073994) PMID:28667006.
  • BAP1 — Enriched in IDH-WT Cluster 4 (q < 0.001 for inactivating point mutations, q < 0.05 for regional deletions); BAP1-mutant CCAs show increased CpG hypermethylation. More frequent in intrahepatic CCAs (q < 0.1) PMID:28667006.
  • FGFR2 — In-frame fusions (FGFR2-STK26, -TBC1D1, -WAC, -BICC1) and a new class of recurrent truncating rearrangements removing the 3′UTR — both classes elevate FGFR2 expression. Indels (n=3), SNVs (n=10), and copy-gain (n=1) also observed. Rearrangements are exclusive to Cluster 4 (p < 0.001); aggregated FGFR2 alterations enriched in Cluster 4 (p < 0.01) PMID:28667006.
  • FGFR3–TACC3 — First reported in CCA; previously characterized as oncogenic in bladder cancer, glioblastoma, and lung cancer PMID:28667006.
  • PRKACB — ATP1B1-PRKACB and LINC00261-PRKACB fusions retain the pseudokinase domain; proposed to activate MAPK signaling PMID:28667006.
  • RASA1 — Newly nominated CCA driver; inactivating mutations in 4.1% (10 frame-shift, 4 nonsense) plus focal CN losses, both correlated with reduced expression. shRNA knockdown in CCA cell lines (M213, HUCCT1) enhanced migration and invasion in Transwell assays — supporting tumor-suppressor function PMID:28667006.
  • STK11 — Newly nominated CCA driver; mutated in 5%, mostly inactivating (7 nonsense, 9 frame-shift) PMID:28667006.
  • MAP2K4 — Newly nominated CCA driver; homozygous deletions in 2 Fluke-Pos cases plus 2.2% mutations (half inactivating) — consistent with tumor-suppressor role PMID:28667006.
  • SF3B1 — Newly nominated CCA driver; mutated in 4.6% at hotspots codon 625 (23%) and codon 700 (14%) — implicating splicing dysregulation in CCA, paralleling uveal melanoma and breast cancer hotspots PMID:28667006.
  • KRAS — Significantly more frequent in intrahepatic CCAs (q < 0.1) PMID:28667006.
  • TERT — Promoter mutations rare in CCA (2/71 WGS cases, 2.8%) at chr5:1295228 PMID:28667006.
  • FBXW7, SMAD4 — Mutations associated with elevated SV burden (q < 0.1) PMID:28667006.
  • CTNNB1, WNT5B, AKT1 — Upregulated expression in Cluster 2 (p < 0.05) PMID:28667006.
  • MYC, MDM2, EGFR, CCND1 — Recurrent oncogene amplifications (n=12, 9, 11, 7 respectively) PMID:28667006.
  • CDKN2A, UTY, KDM5D — Recurrent deletions (n=17, 17, 16 respectively) PMID:28667006.
  • TTC28 — Source of recurrent somatic L1 retrotransposition in 28.2% of WGS tumors PMID:28667006.
  • POLE — Mutated in 2 of the 3 hypermutator CCAs (alongside MSI mutational signatures) PMID:28667006.
  • PDCD1 (PD-1), PDCD1LG2 (PD-L2), BTLA — Specifically upregulated in Cluster 3, motivating immune-checkpoint blockade as a candidate therapeutic strategy in that subtype PMID:28667006.
  • TET1, EZH2 — Cluster 1 shows downregulated TET1 and upregulated EZH2 expression, suggesting loss-of-demethylation plus gain-of-repressive-methylation as the epigenetic mechanism for Cluster 1 hypermethylation PMID:28667006.

Clinical implications

  • Molecular subtype provides prognostic information beyond anatomy. Clusters 3 and 4 carry significantly better overall survival than Clusters 1 and 2 (log-rank p < 0.001), independent of fluke status, anatomical site, and stage on multivariate Cox regression (p < 0.05); reproduced in an independent validation cohort PMID:28667006.
  • Cluster-specific therapeutic hypotheses (require clinical validation):
    • Clusters 1 and 2 — anti-HER2. ERBB2-amplified CCAs (predominantly Fluke-Pos) are candidates for HER2-targeted agents; cell-line data referenced by the authors suggest high-ERBB2 CCAs are more sensitive to ERBB2 inhibition than low-ERBB2 cases PMID:28667006.
    • Cluster 3 — immune checkpoint blockade. Upregulation of PDCD1 (PD-1), PDCD1LG2 (PD-L2), and BTLA, combined with antigen-presentation and T-cell signaling pathway upregulation, motivates immunotherapy trials in this subtype — though sample size is small PMID:28667006.
    • Cluster 4 — IDH inhibitors and FGFR-targeted agents. IDH1/IDH2 mutations and FGFR2/FGFR3 rearrangements suggest opportunities for IDH inhibitors (e.g. ivosidenib, then in NCT02073994) and FGFR inhibitors. FGFR2 3′UTR-loss truncating rearrangements expand the FGFR-targetable population beyond canonical in-frame fusions and activating mutations PMID:28667006.
  • Anatomical classification alone is insufficient for treatment decisions. Tumors in different anatomical sites can be molecularly similar, and tumors in the same anatomical site span all four molecular clusters; current oncology guidelines do not discriminate CCA treatment by anatomical site PMID:28667006.
  • Liver fluke status is not equivalent to molecular subtype. Cluster-associated survival differences persist after fluke-status adjustment, and Cluster 2 contains both Fluke-Pos and Fluke-Neg tumors — cautioning against using fluke status as a proxy for molecular classification PMID:28667006.

Limitations & open questions

  • Platform heterogeneity. Sample-resource constraints (DNA, RNA, FFPE) prevented all platforms from being applied to all samples; the integrative clustering rests on 94–121 samples with multi-platform data while the full driver/SV/CN analyses use larger but partially-overlapping subsets. Authors used statistical models and overlap concordance to argue platform bias is minimal, but cross-platform merging restricts analysis to common genomic regions PMID:28667006.
  • Cluster 3 is small. Immunotherapy nominations for Cluster 3 rest on a small sample size; the mechanistic basis for immune-checkpoint upregulation in this subtype is unresolved (the authors note that aneuploidy-associated immunogenicity is one possible explanation) PMID:28667006.
  • Center-specific pre-processing differences (collection, biopsy site, processing protocols) may introduce sequencing biases; partially mitigated by standardized AJCC 7th-edition histology review and tumor-cell-content estimation PMID:28667006.
  • Functional validation is restricted. Only RASA1 (shRNA knockdown migration/invasion) and FGFR2 3′UTR (luciferase reporter) and 2/3 selected promoter mutations (FIREFLY-prediction validation) were experimentally tested; the other newly nominated drivers (STK11, MAP2K4, SF3B1) and the proposed therapeutic vulnerabilities require further functional and clinical validation PMID:28667006.
  • FIREFLY does not predict directionality of expression change for individual TF binding-change mutations, since TFs can be activators or repressors; gene-set-level effects are inferred without per-mutation directionality PMID:28667006.
  • Cell-of-origin remains a competing explanation for the Cluster 1 vs Cluster 4 hypermethylation differences. The biliary system contains multipotent stem/progenitor cells; liver fluke infection primarily affects large bile ducts while parenchymal liver diseases affect canals of Hering and bile ductules — distinct vulnerable populations could partly explain the molecular differences alongside the proposed extrinsic-vs-intrinsic carcinogenesis model PMID:28667006.
  • Comparison with the contemporary US TCGA CCA series (38 samples, exclusively fluke-negative, mostly intrahepatic, North American — overlaps with chol_jhu_2013 lineage of CCA studies). The TCGA “IDH” and “METH3” groups map approximately to Cluster 4; “ECC” to Cluster 2; the TCGA “METH2” group (CCND1 amplifications) and Cluster 3 are not obviously matched PMID:28667006.
  • Generalizability of FIREFLY. Whether the gene-set-level promoter-mutation framework recovers analogous H3K27me3/PRC2 signals in other cancer types remains an open empirical question raised by the authors PMID:28667006.

Citations from this paper used in the wiki

  • “We analysed 489 CCAs from 10 countries, combining whole-genome (71 cases), targeted/exome, copy-number, gene expression, and DNA methylation information” (Abstract).
  • “Integrative clustering defined four CCA clusters – Fluke-Positive CCAs (Clusters 1/2) are enriched in ERBB2 amplifications and TP53 mutations, conversely Fluke-Negative CCAs (Clusters 3/4) exhibit high copy-number alterations and PD-1/PD-L2 expression, or epigenetic mutations (IDH1/2, BAP1) and FGFR/PRKA-related gene rearrangements” (Abstract).
  • “Whole-genome analysis highlighted FGFR2 3′UTR deletion as a mechanism of FGFR2 upregulation” (Abstract).
  • “Our analysis revealed four potentially new CCA driver genes not highlighted in previous CCA publications – RASA1, STK11, MAP2K4, and SF3B1” (Results, p. 5).
  • “RASA1 … was predicted to be inactivated in 4.1% of cases (10 frame shift, 4 nonsense) … shRNA-mediated knockdown of RASA1 resulted in significantly enhanced migration and invasion” (Results, p. 5).
  • “ERBB2 amplifications were more frequent in Fluke-Pos cases (10.4% in Fluke-Pos vs 2.7% in Fluke-Neg CCA, p < 0.01, Fisher’s exact test)” (Results, p. 5).
  • “we identified 5 in-frame gene fusions with intact tyrosine kinase domains – four involving FGFR2 (FGFR2-STK26, FGFR2-TBC1D1, FGFR2-WAC, and FGFR2-BICC1) and one involving FGFR3 (FGFR3-TACC3) … this is the first report of FGFR3 fusions in CCA” (Results, p. 6).
  • “FGFR2 3′UTR loss may thus represent a new and additional mechanism for enhancing FGFR2 expression in CCA” (Results, p. 6).
  • “FGFR2 rearrangements were observed exclusively in Cluster 4 (p < 0.001, Fisher’s exact test)” (Results, p. 6).
  • “we observed frequent somatic L1 retrotranspositions, particularly originating from an L1 element in intron 1 of the TTC28 gene (52 events in 20/71 tumors, 28.2%)” (Results, p. 7).
  • “only two CCAs (2.8%) harboured TERT-promoter mutations (chr5:1295228)” (Results, p. 7).
  • “FIREFLY (FInding Regulatory mutations in gEne sets with FunctionaL dYsregulation) … uses experimentally determined high-throughput TF-DNA binding data for 486 TFs” (Results, p. 7).
  • “two of these (MIKKELSEN_MCV6_HCP_WITH_H3K27ME3 and MIKKELSEN_MEF_ICP_WITH_H3K27ME3) are subsets of PRC2 target genes” (Results, p. 8).
  • “Cluster 1 … was dominated by hypermethylation in promoter CpG islands, while Cluster 4, enriched in Fluke-Neg CCAs, was dominated by hypermethylation in promoter CpG island shores” (Results, p. 9).
  • “Cluster 4 CCAs were significantly enriched in IDH1/2 mutations, which are known to be associated with CCA hypermethylation (31.6% in Cluster 4 versus 1.0% in other clusters, q < 0.001)” (Results, p. 9).
  • “patients in Clusters 3 and 4 had significantly better overall survival relative to the other 2 clusters (p < 0.001, log-rank test)” (Results, p. 4).
  • “Cluster 4 CCAs, which are associated with IDH1/2 mutations and FGFR2 and PRKA-related gene rearrangements, might also be tested with recently described IDH inhibitors (ClinicalTrials.gov identifier: NCT02073994) or FGFR-targeting agents” (Discussion, p. 11).

This page was processed by crosslinker on 2026-05-15.