Cholangiocarcinoma (CCA) includes a cluster of highly heterogeneous biliary malignant tumours that can arise at any point of the biliary tree. Their incidence is increasing globally,

currently accounting for ~15% of all primary liver cancers and ~3% of gastrointestinal malignancies. The silent presentation of these tumours combined with their highly aggressive nature and

refractoriness to chemotherapy contribute to their alarming mortality, representing ~2% of all cancer-related deaths worldwide yearly. The current diagnosis of CCA by non-invasive

approaches is not accurate enough, and histological confirmation is necessary. Furthermore, the high heterogeneity of CCAs at the genomic, epigenetic and molecular levels severely

summarize and critically discuss the latest advances in CCA, mostly focusing on classification, cells of origin, genetic and epigenetic abnormalities, molecular alterations, biomarker

discovery and treatments. Furthermore, the horizon of CCA for the next decade from 2020 onwards is highlighted.

On the basis of the anatomical site of origin, cholangiocarcinoma (CCA) is classified into intrahepatic CCA (iCCA), perihilar CCA (pCCA) and distal CCA (dCCA). iCCA is defined as a

malignancy located in the periphery of the second-order bile ducts, pCCA arises in the right and/or left hepatic duct and/or at their junction, and dCCA involves the common bile duct (that

is, the choledochus). Grossly, CCA can show three main patterns of growth: mass-forming, periductal-infiltrating, and intraductal-growing. Mass-forming CCA is a mass lesion in the hepatic

parenchyma. Periductal-infiltrating iCCA grows inside the duct wall and spreads longitudinally along the wall. Intraductal-growing CCA is a polypoid or papillary tumour growing towards the

duct lumen.

This international group of multidisciplinary experts in CCA (that is, oncologists, surgeons, hepatologists, geneticists, immunologists, basic scientists) has been intensively collaborating

within the ENS-CCA since 2015 with the main aims of improving our understanding of CCA and the management of patients. In this regard, this expert consensus is endorsed by the ENS-CCA. The

overall goal of this multidisciplinary statement is to provide a detailed critical overview of the current knowledge in this field, proposing some expert recommendations and highlighting

what is envisaged for the next decade.

J.M.B. and G.J.G. identified the areas of interest, stratified the consensus statement into the sections presented in the document and assigned them to selected ENS-CCA members or

non-European collaborators (L.R.R., S.G., S.I.I., J.H. and G.J.G.) who are expert in each field of knowledge and research. To write this document, a PubMed search was conducted by combining

the term ‘cholangiocarcinoma’ with the following terms: ‘epidemiology’, ‘risk factors’, ‘classification’, ‘cells of origin’, ‘diagnosis’, ‘staging’, ‘genetics’, ‘epigenetics’, ‘signalling

pathways’, ‘epithelial-to-mesenchymal transition’, ‘cancer stem cells’, ‘tumour microenvironment’, ‘immunobiology’, ‘in vitro and in vivo models’, ‘biomarkers’, ‘surgery’, ‘liver

transplantation’, ‘therapies’, ‘clinical trials’ and ‘chemoresistance’. No specific search dates were used. All the sections were merged into a first draft by P.M.R and J.M.B. and then

extensively revised to create the final document that was later circulated among all the authors for further correction, improvement, discussion and approval. The data presented in Fig. 2

were obtained by combining the values of mortality rates in men and women for both iCCA and eCCA reported in 2019 by Bertuccio et al.9. For the recommendations on CCA management and research

priorities, ideas were proposed, discussed and approved after final revision by all the authors to reach a consensus.

Global age-standardized annual mortality rates for cholangiocarcinoma (CCA) (deaths per 100,000 inhabitants, including intrahepatic CCA, perihilar CCA and distal CCA) obtained from Bertuccio

et al.9. Data refer to the periods 2000–2004 (2002), 2005–2009 (2007) and 2010–2014 (2012). Yellow indicates countries/regions with low mortality (4 deaths per 100,000 people). Mortality in

eastern countries/regions in which CCA is highly prevalent (that is, Thailand, China, Taiwan and South Korea) have not yet been reported and, therefore, CCA incidence is shown for these

countries1.

The global mortality for CCA increased worldwide during the periods 2000–2004, 2005–2009 and 2010–2014 (Fig. 2), according to the WHO and Pan American Health Organization databases for 32

selected locations in Europe, America, Asia and Oceania9. Furthermore, CCA mortality was higher in men than in women worldwide, and in countries/regions in Asia versus those in the West.

Accordingly, Asian individuals were reported to have the highest mortality (2.81 per 100,000 men in Japan). However, in the USA, the more noticeable increases in mortality between 2004 and

2014 were found for African American individuals (45%), followed by Asian (22%) and white (20%) individuals18. The age-standardized incidence of CCA shows considerable geographical

variation, with the highest incidence in Eastern countries/regions; incidence varies from 85 per 100,000 in northeastern Thailand (the highest reported value globally) to 0.4 per 100,000 in

Canada12. Variations in incidence probably reflect differences in local risk factors and potential genetic predispositions1,2,19.

The three subtypes of CCA can have different risk factors, pathobiology, clinical presentations, management and prognosis, as well as distinct epidemiological trends1,2. Over the past few

decades, the reported age-standardized incidence for iCCA has been steadily increasing in most locations worldwide, whereas the age-standardized incidence for eCCA has been decreasing2,19.

However, these trends need cautious interpretation given that all versions of the main International Classification of Diseases (ICD) have so far failed to include a separate code for the

largest group of CCA (pCCA) and previous versions of ICD–Oncology (ICD-O) have cross-referenced pCCA (technically extrahepatic) to iCCA. Importantly, for the first time, subsequent

iterations of both ICD and ICD-O (ICD-11 and ICD-O-4, respectively) — which are due to come into effect by 2021 — will have separate codes for recording iCCA, pCCA and dCCA20. Having clearly

defined codes for the three subtypes of CCA might facilitate more accurate and meaningful epidemiological data. In the meantime, reported epidemiological trends for CCA and/or biliary tract

cancer need to be interpreted carefully.

Several risk factors, both common and rare, have been linked to CCA (Table 1). Although some risk factors are shared by all forms of CCA, others seem to be more specific for one subtype and

seem to be more important in different regions. A common characteristic amongst many of these risk factors is that they are associated with chronic inflammation of the biliary epithelium and

bile stasis19. Several recognized risk factors have increased globally over recent decades (1990–2016) and could be contributing to increasing CCA rates. For instance, high alcohol

consumption, tobacco smoking and viral infections (hepatitis B virus (HBV) and hepatitis C virus (HCV)) have been reported to increase the risk of CCA development30. Moreover, it is also

important to highlight the global obesity pandemic, as well as the metabolic syndrome and/or presence of nonalcoholic fatty liver disease, as risk factors that deserve future central

attention30. However, in most locations, the majority of CCA cases remain sporadic, without any identifiable risk factor present. A number of studies are examining the potential influence of

commonly used drugs such as aspirin31,32,33 and lipid-lowering statins34,35 in the prevention of CCA. Notably, post-diagnosis aspirin usage has been found to be associated with a reduced

risk of death (HR 0.71) among patients with CCA36. Polymorphisms of host genes encoding enzymes involved in xenobiotic detoxification, DNA repair, multidrug resistance, immune response and

folate metabolism have been linked to CCA development19. There are currently no published genome-wide association studies (GWAS) in CCA, but an appropriately powered one is eagerly

anticipated.

iCCA can emerge at any point of the intrahepatic biliary tree, ranging from bile ductules to the second-order bile ducts (segmental bile ducts). In contrast to HCC, iCCA usually develops in

non-cirrhotic liver37. pCCA can arise in the right and/or left hepatic duct and/or at their junction (so-called perihilar bile ducts)38, and dCCA involves the common bile duct39. The current

term eCCA is now discouraged as it combines subtypes with distinct clinicopathological features, prognosis and therapeutic options, and also due to the difficulties in discriminating

between intrahepatic and extrahepatic origins of pCCA.

iCCA can show three main patterns of growth: mass-forming, periductal-infiltrating, and intraductal-growing1,38 (Fig. 1); pCCA and dCCA present as flat or poorly defined nodular sclerosing

tumours or, less frequently, as intraductal papillary tumours40. CCA can be preceded by pre-invasive lesions39. Histologically, although the vast majority of pCCA and dCCA are conventional

mucin-producing adenocarcinomas or papillary tumours40, iCCA shows several histological variants (that is, conventional, cholangiolocarcinoma and rare variants)41 (Fig. 3; Table 2).

Conventional iCCA can be further classified into two main histological subtypes according to the level or size of the affected duct42,43,44,45,46 (Fig. 3; Table 2). Small bile duct iCCA

presents as a small-sized tubular or acinar adenocarcinoma with nodular growth invading the liver parenchyma, and with no or minimal mucin production42,43,44,45,46. Large bile duct iCCA

arises in large intrahepatic bile ducts and comprises mucin-producing columnar tumour cells arranged in a large duct or papillary architecture38,46,47,48,49. Remarkably, the histological

subtyping parallels the high molecular heterogeneity of CCAs and can be ascribed to different cells of origin and pathogenesis41. Small bile duct iCCA can be characterized by isocitrate

dehydrogenase (IDH1, IDH2) mutations or fibroblast growth factor receptor 2 (FGFR2) fusions40,50,51,52,53,54,55. By contrast, large bile duct iCCA, similar to pCCA and dCCA, shows a high

frequency of mutations in KRAS and/or TP53 genes51,53,56,57. Interestingly, dCCA is also associated with ELF3 mutations58. Growing evidence demonstrates that distinct cells of origin within

an organ can give rise to different subtypes of cancer, typically tissue-specific stem and progenitor cells59,60,61,62. Evidence regarding the cells of origin of CCA in humans was obtained

by phenotyping the candidate tissues and/or cells of origin with respect to CCA subtypes through histological and gene expression analysis38,44,46,63,64,65,66,67,68, whereas indirect

evidence might be derived from risk factors67,68.

Based on the duct size, the intrahepatic biliary tree can be further subdivided into small and large intrahepatic bile ducts (iBDs). Small iBDs are lined by small cuboidal cholangiocytes

whereas columnar and mucous cholangiocytes line large iBDs. Typically, large iBDs contain peribiliary glands within their wall. The extrahepatic biliary tree shares anatomical features with

large iBDs. Histological cholangiocarcinoma (CCA) variants reflect the phenotype of the involved duct and the putative cell of origin. Conventional intrahepatic CCA (iCCA) has two main

variants: small duct-type iCCA arises in small iBDs with cuboidal cholangiocytes representing the putative cell of origin, and large duct-type iCCA involves large iBDs and is considered to

be derived from columnar cholangiocytes and peribiliary glands (seromucous glands; mucous acini are shown in light pink, serous acini are shown in green). Cholangiolocarcinoma (CLC) is a

frequent histological variant of iCCA and its phenotype suggests the origin from bile ductules or ductular reaction (DR) that occurs in chronic liver diseases. The vast majority of perihilar

CCA (pCCA) and distal CCA (dCCA) are considered to originate from the lining epithelium and peribiliary glands. This histological subtyping underlies distinct clinicopathological and

molecular features as summarized in Table 2. eBD, extrahepatic bile duct; HpSC, human pluripotent stem cell.

Small bile duct and cholangiolocarcinoma iCCA subtypes emerge at the level of smaller intrahepatic bile ducts, including bile ductules38,44,46,69. In these portions of the biliary tree,

hepatic stem or progenitor cells (HpSCs) and cuboidal cholangiocytes represent surface epithelium and are the putative cells of origin of these malignancies38,44,46,69 (Fig. 3; Table 2).

Interestingly, HpSCs have been implicated in CK19+ HCC70 and in combined HCC–CCA4,64,71. Notably, CCA-like HCC tumours display embryonic stem cell-like expression traits, further

substantiating the involvement of bipotent hepatic progenitor cells, and in humans display a worse prognosis than HCC72. In line with these findings, nestin, a maker of bipotent progenitor

oval cells, is greatly increased in HCC–CCA tumours and is associated with a worse prognosis, and has been proposed as a new possible diagnostic and prognostic biomarker3,26. Small bile duct

and cholangiolocarcinoma iCCA usually develop on a background of chronic liver disease (such as chronic viral hepatitis and cirrhosis)38,46,70, characterized by HpSC activation71,73,74.

Large bile duct iCCA, pCCA and dCCA derive from columnar mucous cholangiocytes or peribiliary glands38,44,46,47,49,69 (Fig. 3), which are also implicated in the origin of precursor lesions

(such as intraductal papillary neoplasm)46. These malignancies mainly develop in ducts affected by chronic inflammation as in primary sclerosing cholangitis (PSC) or liver fluke

infection46,49,69. In PSC, peribiliary gland cell proliferation, mucinous metaplasia, and dysplasia to cancer progression take place within bile ducts and along the biliary tree, mimicking

the cancerization field (‘field defect’)49,75.

Controversies exist regarding the cellular origins of iCCA based on lineage tracing studies in experimental carcinogenetic models76. Indeed, there is evidence in favour of HpSC,

cholangiocyte or hepatocyte origin of iCCA from these experimental settings76,77,78,79,80,81. Thus, a definitive determination of the origin of iCCA in humans cannot be reached based on

current evidence and requires further research. Moreover, it should be underlined that current experimental models of liver damage do not fully recapitulate the pathogenesis of human chronic

liver disease, including proliferative senescence in hepatocytes66,82,83, and that lineage tracing studies must be conducted and interpreted cautiously76,84,85,86,87.

CCAs are usually asymptomatic during early stages. The most frequent symptom of pCCA and dCCA is jaundice due to biliary tract obstruction88. In iCCA, jaundice is less frequent and mostly

associated with advanced disease. Other symptoms of advanced disease include asthenia, abdominal pain, malaise, nausea, anorexia and weight loss. iCCA is an incidental finding in around

20–25% of cases88. In patients with cirrhosis, ultrasonography surveillance for HCC enables iCCA diagnosis at an asymptomatic, early stage89. Unfortunately, the majority of iCCA cases occur

in the absence of known risk factors90, when the only chance for early diagnosis is by cross-sectional imaging performed for other reasons.

As no specific CCA radiology pattern exists, histopathological or cytological analysis is mandatory to confirm the diagnosis1,28. This diagnosis is based on the WHO classification of biliary

tract cancer showing an adenocarcinoma or mucinous carcinoma101, with tubular and/or papillary structures and a variable fibrous stroma102.

There is no widely used staging system for CCA, although it can be staged according to the American Joint Committee on Cancer (AJCC) TNM system103,104. Despite providing a clinically

meaningful classification correlated with prognosis105, the current TNM classification has some limitations. First, it has limited discriminatory ability between T2 and T3 tumours in

surgically resected iCCAs105,106. T2 tumours include multifocal disease or disease with intrahepatic vascular invasion that probably reflect disseminated disease and the OS in patients with

these tumours does not differ from the OS in patients with extrahepatic metastatic disease105. Similarly, there is also evidence supporting the negative effect of the presence of multifocal

iCCA (iCCA with liver metastases; T2) on prognosis (OS) when compared with other early stages, which might require consideration in future versions of the AJCC TNM classifications107.

Second, although size has been included for the first time as a prognostic factor for iCCA in the eighth edition of the AJCC Cancer Staging Manual, the only cut-off size considered is 5 cm

in T1 tumours. Several authors have shown that a 2 cm cut-off value might identify very early tumours with very low likelihood of dissemination and potential long-term survival with low

recurrence rates24,108. Finally, the TNM classification misses relevant prognostic factors such as the presence of cancer-related symptoms (such as abdominal pain or malaise) or the degree

of liver function impairment. As previously shown with HCC, future proposals from society guidelines should focus on stratifying non-surgical patients for clinical studies using clinical and

imaging data. Notably, Chaiteerakij et al. proposed a new staging system for pCCA based on tumour size and number, vascular encasement, lymph node and peritoneal metastasis, Eastern

Cooperative Oncology Group (ECOG) performance status (ECOG-PS), and CA19-9 level, which has shown a better performance in predicting survival than the TNM staging system109. Also, important

for stratification in clinical trials, radiographic staging parameters need to be developed in the absence of histological staging, and a radiographic staging system has been proposed for

pCCA109.

Initial efforts using integrative genomics approaches to stratify CCA based on prognosis have highlighted extensive deregulated transcriptomic landscapes showing augmented anti-apoptotic

signalling, angiogenesis, signal transduction and transcriptional control8,110. The main oncogenic networks comprised WNT-CTNNB1, MYC, ERBB, TNF and VEGF signalling, emphasizing cell

survival signalling pathways in patients with poor OS8. Regarding genomic alterations, CCA falls midway in the mutational spectrum of cancers111, with an approximately equal content of

genomic alterations in iCCA (median 39 non-synonymous mutations per tumour) and eCCA (median 35 non-synonymous mutations per tumour)56. Massive sequencing

studies56,112,113,114,115,116,117,118,119,120,121 have improved our understanding of the causal mechanisms in CCA, emphasizing the genomic complexity in prevalent oncogenic modules

affecting: cell cycle regulation; DNA damage and genomic instability (TP53, CDKN2A, CCND1, ATM, ROBO2, BRCA1 and BRAC2); MYC amplification; epigenetic regulation including NADPH metabolism

(IDH1 and IDH2), de-ubiquitination (BAP1), SWI–SNF complex (PBRM1, ARID1A, ARID1B, ARID2, SMARCA2, SMARCA4 and SMARCAD1) and histone (de-)methylation (MLL2, MML3, KMT2C, KDM4A, KDM5D, KDM6A

and KDM6B); kinase signalling (KRAS, ERBB1–3, BRAF, PIK3CA, PTEN, STK11, SMAD4 and FGFR1–3); immune dysregulation (JAK–STAT3 signalling); FGFR2 and PRKCA–PRKCB fusions; the WNT–CTNNB1

pathway (APC); Hippo signalling (NF2, SAV1 deletion); METLL13 amplifications; and deregulated Notch signalling. Interestingly, the predominant genomic alterations in CCA are associated with

epigenetic processes122. Indeed, the most clinically significant genomic breakthroughs in iCCA are the discovery of hotspot IDH mutations (IDH1R132 and IDH2R172) that cause an accumulation

of the oncometabolite 2-hydroxyglutarate (2-HG)57, as well as the constitutive active gene fusion event between FGFR2 and many different partners, including the most prevalent (BICC1

(refs50,112,113,114), PPHLN1 (ref.115), TACC3 (ref.112) and MGEA5 (ref.112)). These alterations are important as they are driving current marker-based phase III clinical trials testing

specific agents targeting these alterations in FGFR2 fusion-positive CCA (NCT03773302)123,124 and IDH-mutated CCA (NCT02989857).

To date, information on the inherited predisposing genetic risk factors causing CCA is very limited125. Data mostly stem from GWAS of patient cohorts diagnosed with PSC25,126, with increased

risk of CCA. However, the only detailed genomic association with aetiological risk factors investigated by genome sequencing has been the association with liver fluke infection

(Opisthorchis viverrini and Clonorchis sinensis), with fluke-positive tumours showing an overall higher mutational rate (median 4,700 versus 3,143 somatic mutations per tumour)116 with

prevalent mutations in SMAD4 and TP53 as well as ERBB2 amplifications116,117,118. Furthermore, although not in a high proportion, KRAS mutations have been recurrently found in all CCA

subtypes56,116,117. A statistically significant association has also been observed between TP53 mutation and HBV infection119,120. Few studies have investigated the molecular distinction

between iCCA, pCCA and dCCA8,56,116,121. Nakamura et al. emphasized the difference in anatomical location of the tumour, highlighting IDH, EPHA2 and BAP1 mutations and FGFR2 fusions in iCCA,

whereas extrahepatic tumours specifically show PRKACA and PRKACB fusions as well as mutations in ELF3 (similar to tumours in the ampulla of Vater)127 and ARID1B56. Based on these

fundamental causal alterations, tumours in distinct anatomical sites should probably be treated differently. Besides linking IDH mutations with the response to ivosidenib128, few studies

have related genomic alterations to high-throughput drug screening119,129,130. Among these, Nepal et al. used an approach of integrative genomics in a large cohort of iCCAs to elucidate

unique mutational signatures, structural variants and epigenomic alterations, emphasizing specific oncogenetic mechanisms in four distinct subsets of patients with potential drug responses

and categories: RNA synthesis inhibition, IDH mutant; microtubule modulator, KRAS mutant; topoisomerase inhibition, TP53 mutant; and mTOR inhibitors119.

Epigenetics was shown to play an important part in the initiation and progression of CCA, affecting tumour phenotype in the absence of changes in DNA sequences131. Deregulated patterns of

methylation, histone modifications and aberrant expression of non-coding RNAs promote unbalanced transcription and gene expression that impair cell homeostasis and sustain malignant

transformation. Growing evidence supports deregulated methylation motifs in CCA cells compared with their normal counterparts, with a prevalent hypermethylation of multiple CpG sites

occurring in CCA132,133. One of the largest studies of integrative genetic and epigenetic analyses in CCA, including 489 CCAs from ten countries/regions, has shown how the molecular make-up

of CCA goes beyond the differentiation according to anatomical site116. Indeed, by combining DNA sequencing with transcriptomic and DNA methylation analyses, four clusters of CCA with

different clinical outcomes were identified. Two sets of hypermethylated CCAs stood out, with an interesting association between CpG island hypermethylation and liver fluke-related tumours,

increased mutation rate, downregulation of the DNA demethylation enzyme TET1, upregulation of the histone methyltransferase EZH2 and an increased level of deamination events. Conversely, the

subgroup of iCCAs with enrichment in IDH1/2 and BAP1 mutations, as well as FGFR translocations, showed hypermethylation of the CpG shores (the regions immediately flanking CpG islands, up

to 2 kb away). This different pattern suggests how early epigenetic deregulation caused by external carcinogenic events (for example, liver flukes) are at the basis of CCA development in the

first cluster, whereas in the second cluster, epigenetic aberrations probably arise as a downstream consequence of somatic mutations (IDH) that produce oncometabolites responsible for the

DNA hypermethylation. These differences have remarkable clinical implications, because on the one hand early epigenetic events might be used for early detection of tumours in the first

cluster (by using quantitative DNA methylation markers in the bile of individuals at risk)80 and on the other hand, the tumour clonal mutations might be a marker of effective targeted

therapies (such as IDH inhibitors).

Methylome data can also provide insights into the cells of origin of CCA. Tumours with high genetic and epigenetic occurrence seem to have an enrichment of events within embryonic stem

cell-related bivalent regulation134,135,136. IDH-mutated tumours instead seem to resemble the profile of cholangiocellular CCAs that show gene expression traits of epithelial–mesenchymal

transition (EMT)136. Histone modifications have been less studied in CCA. Histone deacetylase (HDAC) enzymes are responsible for regulation of histone acetylation that ultimately affects

chromatin organization. HDAC were found to be upregulated in CCA in vitro137, and are being investigated as targets of treatment. Evidence also suggests that HDAC inhibitors, as well as

dasatinib, might be particularly active in IDH-mutated tumour cells129,130. Non-coding RNAs account for around 98% human RNAs and include microRNAs (miRNAs) and long non-coding RNAs, among

others. These non-coding RNAs regulate the expression of a plethora of target genes affecting all the hallmarks of the cancer phenotype from cell proliferation and migration to EMT and the

regulation of the primary cilium in cholangiocytes138,139,140,141,142 (Fig. 4).

Non-coding RNAs (ncRNAs) that have been found to be dysregulated (up or down) in cholangiocarcinoma and that have key roles in the regulation of cellular processes, such as proliferation,

cell cycle, ciliogenesis, epigenetics, inflammation, chemoresistance, survival, epithelial to mesenchymal transition (EMT), migration and invasion are shown.

CCA often arises in the setting of prolonged biliary inflammation and/or cholestasis, which contribute to carcinogenesis. According to transcriptomic profiles, the ‘inflammation’ (38%) and

‘proliferation’ (62%) subtypes of iCCA were previously identified and reported to be differentially enriched with activation of the pro-inflammatory and oncogenic pathways, respectively110.

The inflammation subclass of tumours was characterized by induction of immune-related signalling pathways. By contrast, the proliferation subclass was enriched in classic oncogenic pathways,

including deregulated receptor tyrosine kinase (RTK) signalling, RAS–RAF–ERK, PI3K–AKT–mTOR, insulin growth factor receptor 1, MET, polo-like kinase 1, aurora kinase A, KRAS mutations and

stem-like genomic traits as well as a focal deletion in the Hippo pathway (SAV1)8,110,143. Notably, patients with the proliferation subtype of iCCA displayed decreased OS (median 24.3 months

versus 47.2 months for those with the inflammation subtype; P = 0.048).

Cholangiocarcinogenesis is orchestrated by a complex interplay of extracellular ligands (such as pro-inflammatory cytokines, growth factors and bile acids, among others), which are present

in the tumour microenvironment (TME), and increased expression and/or aberrant activation of cell surface receptors and the deregulation of intracellular signalling pathways, finally leading

to cell proliferation, survival and genetic and/or epigenetic alterations (Fig. 5).

The process of cholangiocarcinogenesis, and further tumour evolution and growth, involves complex and heterogeneous processes that include the interplay of extracellular ligands (such as

pro-inflammatory cytokines, growth factors and bile acids, among others), which are present in the tumour microenvironment, and increased expression and/or aberrant activation of cell

surface receptors and the deregulation of intracellular signalling pathways, finally leading to cell proliferation, survival and migration or invasion. The most common genes that might be

mutated or amplified resulting in the overactivation of some of these pathways are KRAS, BRAF, ARID1, PBRM1, BAP1, IDH1 and IDH2. The activation of these signalling pathways might also occur

as a result of the interaction between the tumour epithelia and the tumour reactive stroma. 2-HG, 2-hydroxyglutarate; ECM, extracellular matrix; RTK, receptor tyrosine kinase.

Chronic inflammation and fibrosis facilitate cholangiocyte transformation in a multistep manner, providing extracellular ligands that modulate several signalling pathways. In particular,

sustained IL-6–STAT3 signalling was shown to contribute to mitogenesis by upregulating myeloid cell leukaemia 1 (MCL1) or altering EGFR promoter methylation144,145. Similarly, bile acids are

not genotoxic but might also promote cholangiocarcinogenesis through a mechanism involving the activation of EGFR, induction of COX2, MCL1 and IL-6, and downregulation of farnesoid X

receptor (FXR)146,147. Of note, FXR expression was reported to be decreased in human CCA tumours compared with surrounding normal liver tissue, correlating with tumour differentiation140. By

contrast, the levels of TGR5, another bile acid receptor, were found to be increased in CCA tumours and to be correlated with a worse prognosis (perineural invasion)140. CCA tumours, and

particularly iCCAs and pCCAs, are characterized by a reactive desmoplastic stroma containing cancer-associated fibroblasts (CAFs) that crosstalk with CCA cells secreting paracrine factors

such as heparin-binding EGF-like growth factor, stromal-cell derived factor 1 (SDF1), platelet-derived growth factor (PDGF)-B and extracellular matrix (ECM) proteins148.

Although there are marked differences in the genomic features depending on the anatomical location and risk factors, activation of the RTK signalling pathway is a common event in CCA across

subtypes. In this regard, aberrant EGFR, ERBB2 and MET RTK expression has been found in different CCA subclasses that are associated with worse prognosis8,110. RTK signalling mainly triggers

the activation of the RAS–MAPK and PI3K–AKT–mTOR pathways. Furthermore, RAS–MAPK pathway activation due to KRAS-activating mutations is found in all CCAs without distinction, whereas BRAF

mutations are more prevalent in iCCA149. Interestingly, chromosomal oncogenic gene fusion rearrangements involving FGFR2 RTK occur almost exclusively in iCCA50,52,56,112,113. Besides FGFR2

fusions, ROS1 kinase protein fusions have also been identified in iCCA150. Thus, RTK signalling pathways present actionable molecular alterations that are amenable for therapeutic targeting

at multiple levels. IDH1 and IDH2 encode metabolic enzymes that interconvert isocitrate and α-ketoglutarate51,53,113,117,151. Mutations in IDH1 and IDH2 lead to the production of high levels

of 2-hydroxyglutarate, an oncometabolite that interferes with histone and DNA demethylases and inhibits the mitochondrial electron transport chain. Indeed, IDH-mutant CCAs were shown to

exhibit high levels of mitochondrial and low levels of chromatin modifier gene expression, such as low ARID1A expression due to DNA hypermethylation121. Besides epigenetic silencing,

inactivating mutations in multiple chromatin-remodelling genes (including BAP1, ARID1A and PBRM1) are common in iCCA151.

Developmental pathways, including Notch, WNT and transforming growth factor-β (TGFβ) signalling pathways are prominently active in iCCA compared with HCC, as shown by integrated microarray

analysis152. During liver repair and in inflammatory conditions (known risk factors for iCCA), signalling pathways involved in biliary development are activated in ductular reactive cells,

including Notch, WNT, Hippo–YAP and Hedgehog. The Notch pathway is known to be involved in biliary repair, growth, tubulogenesis, fibrosis and maintenance of the stem cell niche; defective

Notch function due to JAG1 or NOTCH2 mutations causes impaired regeneration and Alagille syndrome153, whereas increased Notch activity has been associated with primary liver tumours154.

Overexpression or aberrant Notch receptor expression has been reported both in iCCAs and eCCA, including pCCA and dCCA155,156,157. Activation of Notch signalling was shown to mediate

transdifferentiation of hepatocytes into cholangiocytes during carcinogenesis79,80,81,158. In this regard, experimental overexpression of the intracellular domain of NOTCH1 receptor (NICD1)

in hepatocytes has been associated with the development of iCCA in mouse models79,80,158. Similarly, inhibition of NOTCH2, the expression of which has been shown to be related to

well-differentiated iCCA155, markedly reduced tumour burden in various mouse models of liver cancer (including iCCA)81,159, whereas overexpression of NOTCH3 was associated with the

development and progression of iCCA, promoting cell survival via PI3K–AKT signalling160. Several Notch inhibitors are being developed, and their availability increases interest in this

pathway161.

The WNT–β-catenin signalling pathway is also known to be activated in most CCAs, in part as an effect of the release of Wnt ligands by inflammatory macrophages infiltrating the

stroma162,163, but also as a consequence of DNA methylation alterations targeting this pathway133 and/or mutations encoding key components of the canonical WNT–β-catenin signalling

pathway164. Notably, the promoter of the WNT–β-catenin pathway inhibitor SOX17 was hypermethylated in CCA tumour tissue compared with healthy tissue, correlating with a worse prognosis after

tumour resection132. Noteworthy, SOX17 was shown to regulate cholangiocyte differentiation and to act as a tumour suppressor in CCA in vitro132. WNT inhibitors successfully inhibit tumour

growth in experimental models163 and clinical trials with agents targeting this pathway are currently being explored164. The Hippo–YAP signalling pathway regulates organ size and cell

proliferation, among other functions165. YAP is a transcriptional co-activator that is usually inhibited by Hippo (MST1 or MST2), but can be activated by Hippo-independent signals, such as

inflammation and changes in ECM composition and stiffness166. Several groups have reported increased nuclear expression of YAP in CCA specimens and correlation with a worse

prognosis167,168,169. In vitro studies on CCA cell lines have shown that YAP can be activated by IL-6, PDGF and fibroblast growth factor170,171. PDGF and fibroblast growth factor form a

feed-forward loop activating YAP; YAP transcriptional targets are genes of these signalling pathways, such as FGFR1, FGFR2 and FGFR4 (refs170,171,172). Genetic alteration of the YAP pathway

seems to be uncommon in CCA, according to an integrative genomic analysis of CCA specimens121. However, mutations in ARID1A have been reported in up to 14% of CCAs149. ARID1A encodes a

subunit of the SWI–SNF chromatin-remodelling complex that among other functions reduces YAP transcriptional activity173.

EMT is a cell plasticity-promoting phenomenon initially reported to occur during embryogenesis, but that also takes place in cancer, enabling epithelial cancer cells to acquire mesenchymal

features with invasive properties that lead to metastatic colonization174. The prototype inducer of EMT is the TGFβ-dependent pathway, whose signature has been identified in iCCA

stroma8,175. In CCA, TGFβ induces EMT directly or cooperates with other major EMT inducer pathways such as EGFR176,177. During this plastic EMT programme, tumour cells lose their epithelial

traits and gain mesenchymal features178. Although initially considered as a binary process, it is now well established that epithelial cells undergoing EMT become mesenchymal in a gradual

manner, known as partial EMT178,179. Thus, EMT is a dynamic process that gives rise to intermediate cellular states with both epithelial and mesenchymal traits, contributing to cell

heterogeneity and a broad range of functions from cancer initiation to progression178,179. Notably, EMT is orchestrated by transcription factors (EMT-TFs), comprising SNAIL, ZEB and TWIST

family, that regulate the expression of epithelial and mesenchymal genes180. CCAs express EMT-TFs, which are associated with poor prognosis regardless of anatomical localization181. Beyond

the EMT programme, EMT-TFs display pleiotropic roles linking EMT to stemness, metabolic reprogramming, immune evasion and drug resistance178,182,183.

Increasing evidence suggests associations between EMT and acquisition of cancer stem cell (CSC) properties in different cancer types65,184, and this might also contribute to CCA

heterogeneity as well as resistance to anticancer drugs. Importantly, CSCs represent a peculiar subcompartment of the tumour cell population crucially involved in recurrence, metastasis and

drug resistance185,186,187. A growing body of evidence indicates that CSCs express EMT traits in human CCAs65,187,188,189. Interestingly, CCA emerging in patients with PSC are characterized

by EMT features and high expression of stem and/or progenitor cell markers in peribiliary glands, suggesting a connection between EMT and stemness in tumour initiation49. Indeed, EMT-TFs,

such as ZEB1, regulate expression of CSC markers by inhibiting miR-200 family members, well-known potent stemness repressors190. In stem-like iCCA, a signature linking miR-200c with EMT

regulators such as ZEB1 and TGFβ has been identified191. Besides EMT, TGFβ is known to promote stemness in CCA cells in vitro (human CCA cell line TFK-1)192. A statistically significant

correlation between TGFβ1 and aldehyde dehydrogenase 1 (ALDH1), a functional CSC marker, has been found in both iCCA and eCCA192. Furthermore, TGFβ-induced EMT resulted in acquisition of

mesenchymal traits, ALDH expression and resistance to 5-fluorouracil (5-FU) in vitro192. Moreover, new evidence suggests that cell plasticity promoted by the EMT programme confers

immunosuppressive effects on carcinoma cells by mechanisms not completely understood178; one mechanism identified so far is the regulation of the immune checkpoint PD1 ligand (PDL1) by ZEB1

in breast cancer cells193.

CCA tumours contain a diverse range of cellular types (Fig. 6). Although the tumour epithelium is considered as the coordinator of tumour growth, the importance of the TME cannot be

understated. Histopathologically, CCA is typified by an extensive cellular and acellular stroma that can comprise the bulk of the tumour194. CCA shares many characteristics with scars that

form around bile ducts in premalignant disease, as usually found in PSC and congenital hepatic fibrosis, suggesting that the origin of the tumour stroma can be found in the regenerative

microenvironment during bile duct repair195. The CCA stroma consists of cancer-associated endothelial cells, CAFs and a complex group of inflammatory cells, including macrophages,

neutrophils, natural killer (NK) and T cells196. In addition to this complex cellular microenvironment, the tumour stroma also contains an extensive network of ECM proteins such as

collagens, laminin and fibronectin197,198. The TME directly interacts with the cancer epithelium to support epithelial proliferation and tumour growth, among which CAFs have been the most

extensively investigated.

Cancer-associated fibroblasts (CAFs) are recruited and persistently activated by cholangiocarcinoma (CCA) cells, in response to the effects of PDGF-D, and of FGF and TGFβ1, also released by

tumour-associated macrophages (TAMs). In turn, CAFs enhance cell proliferation and the invasive ability of CCA cells directly, or by influencing the activity of other cells in the tumour

microenvironment. CAFs stimulate tumour-associated lymphangiogenesis (lymphatic endothelial cell (LEC)), support M2 polarization of TAMs and the activation of regulatory T (Treg) cells,

while dampening the activity of CD8+ T cells, natural killer (NK) and dendritic cells. CAFs also induce heavy remodelling of the extracellular matrix (ECM), which becomes stiffer and affects

mechanotransduction of CCA cells, leading to activation of intracellular pathways, including YAP–TAZ. Soluble factors mediating each cell–cell interplay are shown in boxes of different

colours according to their origin (orange from CAFs, green from CCA cells, light blue from TAMs, red from ECM). Mediators in bold are those with proven effects, the rest are putative

signalling molecules. CAF-derived short-range (Hedgehog (Hh)) and direct (NOTCH3) cell–cell developmental cues also underlie interactions with CCA cells (lower right corner).TH2 cell, T

helper 2 cell.

CAFs are a heterogeneous population of spindle-shaped cells with mesenchymal origin that contribute to tumour progression in many human cancers199. In CCA, the abundance of CAFs positively

correlates with tumour growth and poor survival200. CAFs most likely originate from several different cells types, namely tissue-resident portal fibroblasts, hepatic stellate cells,

pericytes, bone marrow-derived mesenchymal stem cells and monocyte precursor-derived fibrocytes via transdifferentiation and activation181,201,202. This activation process also results in a

metabolic reprogramming that enhances proliferation, cellular motility, as well as secretion of regulatory molecules and components of the ECM. Importantly, although CCA cells express

mesenchymal markers they do not transdifferentiate into CAFs, but they do secrete PDGF-D to stimulate fibroblast migration203. In CCA, the persistent activation of fibroblasts is induced

primarily by TGFβ, fibroblast growth factor and PDGF, which are released from tumour-associated macrophages and CCA cells204. TGFβ was reported to be pivotal in promoting an

iCCA-desmoplastic phenotype in a 3D rat organotypic culture model205, and targeting the TGFβ pathway in thioacetamide-treated rats improved fibrosis and reduced CCA burden206. CAFs secrete a

multitude of signalling molecules (such as IL-1β, PDGF-B, heparin-binding EGF-like growth factor and SDF1) that promote cancer progression by enhancing proliferation, survival, chemotaxis

and angiogenesis148. Furthermore, CAFs have also been shown to promote CCA growth through short-range and direct cell–cell morphogenetic signals, such as NOTCH3 (ref.160) and Hedgehog207. By

secreting immunomodulatory factors, CAFs can also promote an immunosuppressive TME208: they regulate innate immunity by supporting M2 macrophages, and decreasing NK cell activation.

Regarding adaptive immunity, CAFs promote regulatory T cells and T helper 2 cells, and disable dendritic cells and cytotoxic T cells208. Data support the ability of CAFs to interact with

lymphatic endothelial cells209. Following stimulation by PDGF-D originated from the tumoural cholangiocytes in vitro, CAFs secrete VEGF-A and VEGF-C, which recruit and assemble lymphatic

endothelial cells in vascular structures susceptible to tumour cell intravasation209.

Cell interactions within the TME are favoured by the ECM, which is gradually ‘transformed’ into a compact and stiff scaffold, enabling mutual communications and exchange of paracrine factors

between the different cell elements210. The ECM is continuously remodelled by deposition of newly synthesized matricellular proteins, including tenascin C, osteopontin and periostin, in

concert with an intensive degradation by matrix metalloproteinases (MMPs; MMP1, MMP2, MMP3 and MMP9) that are copiously released by CAFs, tumour-associated macrophages and malignant

cholangiocytes210. Thanks to these phenotypic changes, ECM boosts key pro-invasive functions of tumour cells. In cooperation with collagen I, tenascin C and integrins (α5β1, α5β3, α5β5 and

α6β4), periostin stimulates cell proliferation of malignant cholangiocytes in a PI3K–AKT-dependent manner in vitro211. Increased ECM stiffening is also instrumental in the activation of

intracellular mechanosensors, such as YAP–TAZ, involved in tumour initiation and progression. Whereas soft ECM inhibits YAP–TAZ activity by favouring its sequestration by the SWI–SNF

chromatin-remodelling complex through ARID1A, stiff ECM induces YAP–TAZ to detach from SWI–SNF and to bind to TEAD, unfolding a transcriptional programme and promoting cell proliferation,

CSC traits, plasticity and reprogramming173,212. Overall, the multifaceted interplay of CAFs with tumour cells, immune cells, lymphatic endothelial cells and ECM is continuously evolving

(Fig. 6) and could offer potential therapeutic targets. Importantly, selective pro-apoptotic targeting of CAFs with subsequent reduction in tumour growth and lymph node metastases has been

demonstrated in a CCA rat model213. Overall, the signalling networks that govern CCA tumours are the result of the intrinsic genomic and epigenetic alterations of tumour cholangiocytes, as

well as their interplay with CAFs, immune cells and ECM. The secretion of proinflammatory, oncogenic and fibrogenic factors from CCA cells could contribute to the recruitment of other cells

to the TME, which in turn will activate and sustain specific signalling pathways in cancer cells, thus perpetuating CCA growth and progression.

Transcriptomic sequencing of CCA tumours has demonstrated that the subset of patients with the poorest prognosis have an elevated tumour mutational load and enhanced expression of immune

checkpoint molecules56. Importantly, the presence of T cell-infiltrated TMEs, characterized by infiltration of CD8+ T cells, chemokines and molecules responsible for T cell priming and

immune infiltration, is associated with higher response to immune checkpoint blockade, whereas non-T cell-infiltrated TMEs have poorer responses214,215.

Regarding innate immune responses, activated or ‘M2-like’ tumour-associated macrophages are anti-inflammatory and immunosuppressive. M2-like macrophages stimulate WNT signalling with

consequent CCA progression163, and are associated with inferior patient outcomes216,217. High numbers of M2 tumour-associated macrophages are linked to poor disease-free survival in patients

with iCCA217. Similarly, in a retrospective study of patients with pCCA who had undergone surgical resection, high density of tumour-associated macrophages in the tumour invasive front

correlated with increased local and tumour recurrence216. Myeloid-derived suppressor cells are another immunosuppressive element in the TME. Fibroblast activation protein-positive (FAP+)

CAFs promote myeloid-derived suppressor cell infiltration in desmoplastic tumours218. Moreover, increased stromal FAP expression in human resected CCA specimens has been linked to poor

patient outcomes218. The presence of CD83+ dendritic cells in human resected CCA specimens was associated with better outcomes219. Although NK cells comprise 30–40% of all hepatic

lymphocytes220, current knowledge on the role of these cells in CCA is limited. Culture of CCA cells (human CCA cell lines, Hucct1 cells and OZ cells) with the anti-EGFR monoclonal antibody

cetuximab augmented CCA cell death via NK cell-induced antibody-dependent cellular cytotoxicity221. Similarly, infusion of ex vivo-expanded human NK cells in CCA mouse xenograft models

resulted in tumour regression222.

CCA progression has been associated with a decrease in the components of the adaptive immune response223. Immunohistochemical analyses have demonstrated a preponderance of CD8+ T cells

within the tumour and CD4+ T cells in the tumour–liver interface224, as well as an association with longer OS and the presence of tumour-infiltrating CD4+ or CD8+ T cells223,225,226,227.

Similarly, the presence of B cells has been linked to a favourable prognosis in CCA223,224. Factors associated with a higher likelihood of response to immune checkpoint blockade include the

presence of biomarkers such as PDL1, genetic aberrations such as DNA mismatch repair (MMR) deficiency and/or microsatellite instability (MSI), and the cumulative tumour mutational burden2.

On the basis of small cohorts of patients with CCA (range 41–104 patients across the studies), PDL1 is expressed in 42–72% of tumours228,229,230, and seems to be present primarily on immune

cells228,229. MMR deficiency has been reported in 5% of pCCA or dCCAs and 10% of iCCAs231. Of note, hypermutation was found in 6% of CCAs and MMR deficiency and/or MSI was present in 36% of

these hypermutated tumours56. MSI-high tumours are generally ‘hot’ tumours with an increased number of neoepitopes, CD8+ T cell infiltration, and improved responses to immune checkpoint

blockade in cancer generally215. In a cohort of 86 patients with MMR-deficient tumours, including four patients with CCA, immune checkpoint blockade with the anti-PDL1 monoclonal antibody

pembrolizumab resulted in a complete response in one of the patients with CCA and stabilization of disease in the other three232. These data indicate that immune-directed therapies including

immune checkpoint blockade are a promising approach for, at least, this subset of patients with CCA.

Over the past decade, a number of in vitro and in vivo models of cholangiocarcinogenesis have been generated to clarify the phenotypic, biochemical and biological events occurring during the

transformation of normal cells into fully malignant cholangiocytes (Table 3). In vitro cell lines, mainly derived from human CCA specimens, have been used widely as a tool to study this

disease76,233,234,235. Cell lines exhibit various advantages over animal models: they are free from non-tumourous and necrotic tissues, their growth can be synchronized, relatively high

numbers of cells can be produced, cell proliferation and apoptosis can be accurately determined, and they can be molecularly modified (that is, by overexpression or silencing of genes, using

antisense oligonucleotides, small interfering RNAs, CRISPR–Cas, and so on), therefore enabling the study of single genes or signal transduction pathways. Furthermore, cell lines can be

subjected to drug administration. However, in vitro passaging renders cell lines increasingly different from the original tumours. Primary cultures of CCA cells from tumour tissue are used

shortly after derivation and grown under serum-free growth factor-enhanced conditions; therefore, more closely resembling the in vivo situation236,237. Unfortunately, important shortcomings

also apply to this system; in particular, primary cultures are time-consuming and elimination of non-tumour cells can be complicated. Furthermore, primary cultures can only be established

from surgically resected specimens, limiting the applicability to a subset of patients with CCA who have undergone surgery. Also, primary culture cells lack realistic intercellular and

cell–matrix interactions236,237. Importantly, preneoplastic (for example, PSC-derived cholangiocytes) and/or normal cholangiocyte primary cultures should be used as controls132,140,238,239.

To recapitulate more adequately the in vivo tumour tissue structure and to investigate the interaction between CCA and the TME, 3D model systems, known as tumour spheroids and organoids,

were developed240,241. Tumour spheroids are self-assembled cultures of cancer cells in the presence or absence of stromal cells within a hydrogel, mimicking the basement membrane, where

cell–cell interactions predominate over cell–substrate interactions241. By contrast, tumour organoids are self-organizing stem cell-like structures cultured and expanded in a

hydrogel76,242,243. Organoids are successfully established from resected tissue biopsy and needle biopsy samples, faithfully recapitulating the patient tumour at the histopathological level,

both in culture and as xenografts in immune-deficient mice242,244. However, to recapitulate CCA tumours in vivo, organoids should be co-cultured with stromal cells. Importantly, whole-exome

sequencing revealed that the vast majority of the mutations are retained in liver cancer organoids derived from resected tissues, whereas mutation retention is heterogeneous in

biopsy-derived liver cancer organoids76,242,243. Furthermore, CCA organoids have been shown to be a reliable system for drug testing and personalized medicine applications, and possess an

almost negligible capacity to differentiate into hepatocytes76,245. As an alternative method, CCA organoids can be established by inducing genetic mutations in healthy organoids via viral

transduction and CRISPR–Cas9 genome editing approaches, thus enabling the characterization and elucidation of the roles of oncogenes and/or tumour suppressor genes, either alone or in

combination, in cholangiocarcinogenesis76,246.

Mouse models of CCA enable the investigation of the pathobiology of the disease and treatment response in a context that more closely recapitulates the human disease76,247,248,249. Multiple

approaches have been used to induce CCA formation in mice and the principal mouse models can be classified into four major groups: chemically induced models, in which a chemotoxic drug is

responsible for the oncogenic insult(s); genetically engineered mouse models (GEMM); implantation models; and transposon-based models. As human CCA can develop in the setting of a diseased

liver, various methods have been developed to mimic liver alterations, such as those induced in humans by viral hepatitis, chronic inflammation and cholestasis, further increasing the

similarity with the human situation. Another major advantage of in vivo models is that they enable the study of CCA starting from early pre-neoplastic to fully progressed lesions, meaning

researchers can dissect the specific molecular events occurring at various stages of cholangiocarcinogenesis. In addition, in vivo models enable real-time monitoring of tumour development

and response to therapies using imaging modalities such as CT or MRI, or other techniques involving bioluminescence.

The current ‘omics’ era is enabling the discovery of new and promising biomarkers in biofluids (serum, urine, bile, saliva) and tumour tissue that could change the paradigm in disease

diagnosis and management in the upcoming years (Supplementary Figure 1).

Circulating nucleic acids found in biofluids after active transport or resulting from dying cells are promising diagnostic and prognostic tools for human disorders250,251,252. Cell-free DNA

(cfDNA) has been envisaged as mirroring changes in tumour aggressiveness and size, being found both in tumour tissue and plasma from patients with CCA253. Detection of cfDNA in plasma

samples could also guide potential mutational-based therapeutic interventions as de novo multiple point mutations in FGFR2 kinase domain were detected in cfDNA, primary tumours and

metastases from patients with CCA with acquired resistance to the pan-FGFR inhibitor BGJ398 (ref.254). On the other hand, miRNAs have received special attention due to their increased

stability and abundance in biofluids. Two meta-analyses have evaluated their diagnostic value for CCA, and found a pooled area under the receiver operator curve (AUC) of ~0.9 (refs255,256).

Notably, bile represented the biological fluid with the highest diagnostic capacity, followed by serum, tissue and urine (AUC 0.95, 0.913, 0.846 and 0.745, respectively)256. In this regard,

some bile miRNAs have already been shown to display increased diagnostic capacity for CCA, in comparison with healthy individuals (miR-9, miR-145)257 and also when comparing patients with

PSC-derived CCA and isolated PSC (miR-412, miR-640, miR-1537, miR-3189)258. Importantly, combining miR-1537 with CA19-9 resulted in higher diagnostic values than CA19-9 alone (AUC 0.91

versus 0.88; P > 0.05)258. In serum, the levels of miR-21 (refs259,260,261), a well-known onco-miR, were found to be increased in patients with CCA, compared with healthy individuals,

positively correlating with clinical stage and poor survival, although the translation of this miRNA into clinics should be performed carefully since it is usually increased in serum and/or

plasma of patients with HCC and other liver diseases and cancers262,263. Other miRNAs were also differentially found in the serum and/or plasma of patients with CCA compared with control

individuals264,265,266,267,268,269,270, but with some inconsistency, which underscores the necessity for conducting further studies for validation in large, biopsy-proven and

well-characterized cohorts of patients and adequate controls.

Proteins and cytokines are now regarded as potential diagnostic and/or prognostic biomarkers. A soluble fragment of cytokeratin-19 (CYFRA 21-1), MMP-7, osteopontin, periostin and IL-6, among

others, were shown to be enriched in the serum of patients with CCA, when compared with healthy individuals as controls and/or patients with benign biliary diseases (such as

PSC)271,272,273,274,275,276,277,278,279,280,281,282. Among these biomarkers, increased CYFRA 21-1 and osteopontin levels showed superior diagnostic capacity for identifying CCA compared with

CA19-9 and CEA271,275, and also showed prognostic value. Of note, increased serum periostin levels were also associated with decreased OS, and the serum periostin level was an independent

prognostic factor (HR 3.197)282. As cancer cells display marked metabolic alterations, measuring metabolites in distinct biological samples is now regarded as an encouraging alternative to

find diagnostic and/or prognostic biomarkers. Up to now, only a limited number of studies have addressed this issue. Bile acids and phospholipids have been highlighted as promising

metabolites in bile for the diagnosis of CCA, as their levels are increased in patients with CCA compared with healthy individuals and patients with HCC283,284,285,286,287. Serum

metabolomics has also revealed promising diagnostic biomarkers288,289. An international collaborative study including patients with biopsy-proven iCCA, HCC or PSC and healthy individuals

found that several metabolites had higher diagnostic capacity for iCCA than CA19-9, and the authors proposed an algorithm containing six metabolites that was able to differentially diagnose

iCCA and HCC (AUC 0.9) in discovery (n = 20 per group) and validation phases (independent cohorts of 14–15 patients per group)289. Interestingly, proteomic analysis of serum extracellular

vesicles from patients with CCA, HCC or PSC and healthy individuals as controls revealed candidate proteins with high accuracy for the differential diagnosis of these liver diseases, having

higher AUC values than either CA19-9 or α-fetoprotein levels239. Furthermore, another study identified an extracellular vesicle-derived miRNA panel in bile (miRNAs miR-191, miR-486-3p,

miR-1274b and miR-484) for the discrimination of CCA from non-malignant biliary diseases290. Although few studies have addressed the potential role of circulating tumour cells (CTCs) as

diagnostic and/or prognostic biomarkers in CCA, in a study investigating the associations between numbers of CTCs, patient and tumour characteristics and survival in patients with biliary

tract cancer, 17–25% of the patients showed elevated numbers of CTCs (two or more per 7.5 mL of blood)291,292,293, and