Unmet needs in basic and translational research in Cholangiocarcinoma

Despite the impact of cutting‐edge technologies in providing deep molecular phenotyping of many tumours, management of cholangiocarcinoma (CCA), a rare and insufficiently studied cancer with marked heterogeneity (including intrahepatic and extrahepatic variants), has remained limited and it has poor prognosis. Renewed interest in this enigmatic disease has been fostered in the last decade. Here, we will give an overview of the most important gaps in knowledge of the basic and translational research of CCA that must be prioritized to improve the CCA management in the future.


| INTRODUC TI ON
Cholangiocarcinoma (CCA) represents one of the biggest challenges that the scientific community, at multiple levels, beyond hepatologists and oncologists, is currently coping with. 1 Indeed, because of its major aggressiveness and usual late diagnosis at an advanced stage, since the new millennium improvement in clinical management of CCA has been extremely limited, resulting in a 5-year survival that has remained dismal (7%-20%), 2 reflecting a long and unfair neglected topic ( Figure 1). Although in the last few years there has been some renewed interest in this enigmatic disease and its several controversial facets, CCA is still a sort of 'child of a lesser God' if compared with the 'giants' of the oncology research, breast and lung cancer above all, and even the major type of liver cancer, that is, hepatocellular carcinoma (HCC). Besides being epidemiologically rare, at least in Western countries, CCA is also an understudied cancer, with less than 10% of published works compared to breast and lung cancer, and about 17% as compared to the 'closest relative' HCC (Table 1). Moreover, the main issue making investigational approaches to CCA even more difficult and likely reluctant has been the high degree of heterogeneity of this cancer at both intertumoural and intratumoural levels. 1,3 CCAs encompass different clinical entities, which reflect the variable anatomical site of origin within the biliary tree (intrahepatic, peri-hilar and distal CCA, the latter two considered as extrahepatic) and indeed, well-established classification criteria have been applied in the study design only recently. 1,4,5 Heterogeneity is further complicated at the cellular level by a multifaceted microenvironment, containing a variety of stromal and immune cell elements, whose composition has only recently begun to be described. [6][7][8] Against this background, in this review we will give an overview of the most important gaps in knowledge of the basic and translational research of CCA that must not be delayed further.
The aim is to provide the reader with a series of research priorities that, from our point of view, should be raised to improve the CCA management in the next decade.

| ME THODS
To perform this study, the initial search had been based on the Consensus Statement on Cholangiocarcinoma 2020. 1 to which the authors herein involved contributed (RIRM, MS, JJGM, LF). In this consensus paper, priorities of either basic/translational or clinical research were proposed, selected and approved after careful discussion among experts in different fields. Starting from these outlines, a PubMed search was conducted by combining the term 'cholangiocarcinoma' with the following key words: 'genetics', 'genome-wide association study (GWAS)', 'molecular classification', 'chemoresistance/ pharmaco-resistance', 'biomarkers', 'microenvironment', 'cancerassociated fibroblasts', 'cancer stem cells', 'tumour-infiltrating lymphocytes', 'tumour-associated neutrophils', 'tumour-associated lymphangiogenesis', 'extracellular vesicles', 'miRNAs', 'circRNAs', 'lncRNAs' and 'in-vitro and in-vivo models'. No specific search dates were used. Sections were assigned according to the expertise of each author, and once pooled the whole document was thoroughly discussed and extensively revised by all co-authors.

| G ENE TI C S , MOLECUL AR PROFILING AND CL A SS IFI C ATI ON
In recent years, the application of innovative technologies, such as next-generation sequencing (NGS) or single cell analysis, to the study of CCA has allowed an extensive profiling of the genetic and molecular alterations of this neoplasm. Thanks to these approaches, it has become clear that CCA is an extremely diverse cancer that besides different histopathological variants, includes a wide number of mutations with related signalling perturbations that underpin its pathophysiology. This in-depth analysis has led to the classification by  Containing Protein 1 (SAV1). Clinically, the proliferative class showed a poorer survival rate and greater tendency to recurrence compared with the inflammatory class. Other studies performed in different iCCA cohorts gave substantially confirmatory results. 10,11 Some years later, the molecular landscape of the extrahepatic CCA (eCCA) was similarly dissected, 12 appearing much more complex than iCCA, with the description of four subclasses: proliferation, mesenchymal, metabolic and immune, each harbouring specific genetic and signalling alterations ( Figure 2). Among these subclasses, the mesenchymal type was the most aggressive. According to OncoKB targets, 13 only 25% of eCCAs were characterized by the presence of actionable targets and less than half with respect to iCCA (50%-60%), highlighting the concept that eCCA is an even less attractive disease for targeted therapies. 12,14 In particular, potentially targetable genetic alterations, such as Fibroblast Growth Factor Receptor (FGFR) aberrations (fusions/mutations/amplifications) and Isocitrate dehydrogenase 1 (IDH) mutant-enriched subtypes, while quite common in iCCA (20% for FGFR, 15% for IDH), 15 are instead rare in eCCA (1% and 4.7% respectively). 12 Based on these observations, only iCCAs are currently included in clinical trials assessing efficacy of FGFR and IDH inhibitors. 16 Among genetic mutations, KRAS shows a quite relevant proportion in both iCCA ((20%-54% 17 and eCCA (36.7%) 12 ), but unfortunately targeting this pathway is troublesome, since no approved drugs capable of targeting mutated KRAS proteins directly are available. Additionally targeting KRAS indirectly at the level of its downstream effectors is unsuccessful as shown in pancreatic ductal adenocarcinoma (PDAC), a similarly aggressive malignancy frequently demonstrating activation of KRAS. 18,19 The analysis of a cohort of mixed HCC-CCA 20  Altogether, studies devoted to genetic and molecular characterization are technically sound and highly refined, but the deliverables are still preliminary since they lack a consensus and leave open several discrepancies between the histopathological and the molecular classifications, which need to be analysed and standardized. Moreover, given the low prevalence of actionable driving pathways, in particular in eCCA, 12 further efforts aimed at dissecting the intimate mechanisms of tumour invasiveness, from epigenetic modifications, to 3D chromatin conformation, post-translational modifications (SUMOylation, NEDDylation), and secretome and transcriptomic alterations, will be necessary to improve the success of therapies. Another aspect of molecular profiling with therapeutic potential is the assessment of mismatch repair (MMR) deficiency and microsatellite instability (MSI) as well as programmed death-ligand 1 (PD-L1) expression, which support the use of immunotherapy.
Unfortunately, these features can be found in only 8% of iCCA 22 and 2% of eCCA, 12 thereby making this treatment option unlikely to be useful for CCA.
Although CCA is classically recognized as a rare cancer, there are considerable geographical variations in age-standardized incidence. Indeed, CCA is not rare in the South-East Asia (China, South Korea and Thailand), where the incidence is greater than six cases per 100,000 people. 21 Moreover, CCA mortality is higher in men than in women worldwide. 1 Thus, it is tempting to speculate on the one hand, that beyond local environmental factors, genetic predisposition is different across ethnic groups, and on the other, that gender-related differences may sustain diverse aggressive tumour phenotypes. Today, data on the inherited factors involved in CCA pathogenesis are very limited. 'Omics' studies aimed at addressing these differences should help to bridge this gap.
Overall, these refined molecular classifications, although methodologically sound and relevant as role models for experimental protocols applied to the study of heterogeneous cancers, Pharmacoresistance can be primary (intrinsic) or secondary (acquired) triggered by the exposure to drugs. Unfortunately, MPRs usually induce resistance to multiple drugs (for a complete recent review, see 25 ). MPRs are based on the existence of a dynamic resistome, 26,27 consisting of more than one hundred genes classified into seven groups (MRP-1/7). Those included in MRP-1 are involved in lowering the intracellular concentrations of active drugs by hindering their uptake or increasing their release, which in CCA markedly determines the efficacy of gemcitabine, 5′-FU, and platinum-derived drugs. Their uptake through equilibrative nucleoside transporter 1 (ENT1) and copper transporter 1 (CTR1) is usually hampered in chemoresistant CCAs. 28,29 Changes in the expression and function of other plasma membrane transport proteins, such as organic cation transporters (OCTs), affect the response of CCA to cationic drugs, such as several tyrosine kinase inhibitors (TKIs). 30 On the other hand, the high expression of ABC proteins, mainly MRP1, 31 and MRP3, 32 also plays an important role in the sensitivity of CCA to pharmacotherapy. Besides, the proportion of active drugs can be lowered in resistant CCA (MPR-2). 33 Other genes of the resistome are related to altered interaction with molecular targets (MPR-3). For instance, the response of CCA to pemigatinib has been associated with the presence of mutations in FGFR2. 34 Moreover, IDH inhibitors can lose their efficacy in fighting CCA if mutations altering the IDH1 or IDH2 sequence are found in these tumours. 35 To overcome this problem, several drugs, such as ivosidenib, are being developed. 36 The efficacy of DNA-damaging drugs, such as cisplatin and oxaliplatin, depends on DNA repair machinery (MPR-4), which can be enhanced by up-regulation of the genes involved, such as P53R2 and KPNA2. 37 A considerable part of the resistome is associated with genes included in MPR-5, which alter death-related signalling pathways favouring survival over apoptosis. An example is the up-regulation of the anti-apoptotic protein Bcl-2 and down-regulation of the pro-apoptotic Bax, which occurs in gemcitabine-resistant CCA cells. 38 Recently, extracellular mechanisms that involve crosstalk between tumour cells and environmental factors affecting drug response (MPR-6), and those that favour phenotypic transitions, such as epithelial-to-mesenchymal or development of stemness characteristics, have been recognized as critical elements of CCA resistome (MPR-7). These are promoted by HMGA1, highly expressed in resistant CCA. 39 Regarding modern immunotherapy, the appearance of epigenetic alterations in CCA constitutes one important mechanism contributing to drug resistance by helping tumour cells to escape host immune surveillance, which has prompted the development of novel drugs that target DNA methylation or histone modifications. 40 Further research efforts should be made into treatments that are currently showing promising results in clinical trials, especially on targeted therapies and immunotherapies. 41,42 Targeted therapies aimed at MPRs can be 'on-target' when the resistance occurs because of mutations in the primary molecular target, which results in poor or no response to the specific drug or it may be 'off-target' when the resistance occurs through activation of signalling pathways parallel to that in which the target of interest is involved. Furthermore, since most molecular targets are localized inside tumour cells, insufficient intracellular drugs levels because of MPR-1 and MPR-2 markedly affect the response to these drugs.

| B IOMARKER S
Given the difficulty in obtaining a quality biopsy to confirm the diagnosis of CCA, in many cases, the availability of minimally invasive markers is an important unmet need in the management of these patients. 48 Currently, the only serum biomarker recommended for routine clinical practice in CCA is carbohydrate antigen 19-9 (CA19-9). 49 Although for the diagnosis of CCA its sensitivity and specificity are far from ideal, especially in the early stages, the determination of this glycoprotein in serum is helpful in the follow-up after surgery to detect tumour recurrence and evaluate the response during pharmacological and radiological treatments. 50 During the last few years, alternative serum biomarkers have been actively sought, 50 especially using 'omics' technologies, combinations of proteins and metabolites, [51][52][53][54] and RNA profiles determined in extracellular vesicles (EVs). 55 Despite the fact that some combinations have shown potential usefulness in CCA diagnosis, at present, none of these biomarkers has reached the clinical setting although, some validation studies are underway.
Recent advances have identified genomic alterations characteristic of CCA associated with the anatomical origin of the tumour. [56][57][58][59][60] Liquid biopsy to evaluate circulating tumour cells (CTCs) and circulating cell-free tumour DNA (ctDNA) released by primary or metastatic sites is an alternative diagnostic approach to overcome the limitation in obtaining biopsies to evaluate tumour progression.
Although monitoring CTCs is showing promising results for other types of cancers, such as HCC, and breast and colorectal cancer, [61][62][63] the findings in CCA are still preliminary. 64,65 It has been suggested that the presence of viable CTCs in portal blood after resection and their interaction with immune cells and myeloid fibroblasts (cell elements actively engaged in the tumour microenvironment), may be responsible for the metastatic spread of CCA. 64 In addition, the presence of CTCs has been shown to correlate with tumour extent and reduced overall survival in CCA patients. 65 Despite the marked inter-tumoural and intra-tumoural genetic heterogeneity, several studies have described a correlation between genetic and epigenetic alterations in tissue-based tumour DNA and ctDNA, 66,67 supporting the usefulness of the latter in the diagnosis of CCA. It has been demonstrated that using NGS, ctDNA analysis is feasible and accurate, although its sensitivity may be low in earlystage CCA tumours. 68,69 Interestingly, more significant amounts of ctDNA have been associated with advanced disease and worse prognosis. 69 The ability to repeat blood analysis more frequently than other traditional techniques (biopsy, imaging) can make it easier to follow the evolution of this malignancy. In addition, this approach F I G U R E 2 Graph representation of the molecular subclasses in intrahepatic (iCCA) and extrahepatic cholangiocarcinoma (eCCA) as defined by comprehensive multi-platform molecular profiling studies. In iCCA, two main subclassess, proliferation and inflammation, have been identified, whereas in eCCA, the proliferation class is accompanied by metabolic, mensenchymal and immune classes. For each sublcass, the most relevant molecular signatures along with the relative proportion are indicated. Data for iCCA and eCCA are reported according to 9-11 and 12 respectively may permit the identification of druggable targets as well as resistant mutations to select the best treatments at each moment.

| TUMOUR MICROENVIRONMENT ( TME )
As already mentioned, regardless of the anatomical subtype, a defin- with the neoplastic bile ducts, which as with the cellular counterpart may also influence tumour cell behaviour. 70 Since TME is crucial in regulating the invasive functions of the tumoural ducts, in both a promoting and restraining fashion, a more detailed understanding of the complex interplay between tumour and stroma/immune milieu is a prerequisite to uncover novel and effective therapeutic targets.
Targeting cancer-associated fibroblasts (CAF) by pro-apoptotic agents (ie navitoclax) is paradigmatic, and in rodent models of CCA it has significant anti-tumour effects, with impaired growth, invasiveness and lymphatic spreading. 71,72 However, this observation stands in stark conflict to that reported in other tumour contexts, that is, PDAC, whereby CAF depletion yielded undifferentiated tumours, with enhanced epithelial-to-mesenchymal transition (EMT), increased tumour cell proliferation and more invasive phenotype. 73,74 Dual effects of targeting CAF may depend upon a pronounced CAF heterogeneity, which in iCCA has been recently analysed at single cell resolution, with identification of five CAF subtypes with distinctive functions. These include vascular, matrix, inflammatory, antigen-presenting, and EMT-like, each variably affecting the tumour phenotype. 75  (specifically miRNAs, circular RNAs, and long non-coding RNAs), 80,81 reactive oxygen species (ROS) 82 and energy-rich metabolites, which are mainly mediated by EVs. 83 This is a field of research awaiting discovery with interesting perspectives and therapeutic implications.
Indeed, selective cargo delivery via EVs provides an attractive tool to potentially induce more effective responses in the target cells with fewer side effects or harmful reactions in other cells. However, the underlying methodology needs to be refined before its applicability and efficacy to be confirmed. 84 Furthermore, given the multicellular composition of the TME, other ways of communicating are likely to operate and in particular, those that engage immune cells have been largely ignored and may provide hints for therapeutic intervention. 85,86 In this scenario, tumour-associated macrophages (TAM) displaying M2 features are classically recognized as tumoursupportive players, but like CAF, they encompass a highly heterogeneous population with a range of phenotypes endowed with a wide tumour-related functionality (including angiogenesis, ECM remodelling and T cell inhibition) that cannot be reduced to a simple M1 and M2 polarization. 87,88 Studies of phenotype-function correlation, as well as deciphering TAM interplay with CAF, vascular cells and other immune cells populating the TME are sorely needed to better appreciate their culture systems, such as spheroids and organoids. 108,113 Historically, the first in-vitro models used were the 2D cultures, because of the fast growth, the low maintenance costs and the high experimental reproducibility. However, these culture types tended to accumulate mutations, did not mimic interactions with TME and nor create polarized monolayers, factors that were detrimental for oncology research. In addition, because of the presence of serum in the culture media, they usually did not contain CSCs, which were unable to remain viable in these conditions. Furthermore, purification procedures provided a selection of tumour cell clones, thereby missing the intratumoural variability typically reported in CCAs. 108 To circumvent some of these limitations, 3D culture systems, divided into spheroids and organoids, have been proposed. 113 Spheroids are cellular aggregates of broad origin (tumoural, embryonic, etc) that grow from a single cell in suspension, while organoids are cultures that grow on matrix, such as hydrogel or matrigel, and derive from embryonic stem cells, induced pluripotent stem cells, progenitor cells or primary cells. 113 The spheroids are easy to generate and handle, and their culture is relatively cheap. Spheroids are genetically stable and form polarized cell layers but do not reproduce the histological and morphological characteristics of the tissue, from which they derive. Organoids are genetically very stable and, given the absence of serum in the culture medium, they maintain the CSC phenotype.
They form polarized cell structures with stable cell-to-cell junctions and preservation of the native tissue architecture, but without reproducing the interactions with the ECM, which is a prerequisite for TME studies. Of note, the organoid technique can be extended to in-vivo settings to generate primary tumours that, once excised, may generate secondary tumours in a second recipient animal. 114 Unfortunately, the high costs of cell maintenance, and the duration of generating such cultures (taking several weeks), are major deterrents to their widespread use even in skilled labs. 108,113 A recap of pros and cons of experimental models of CCA is given in Table 2.
Conceivably, future efforts will be required to develop in-vivo and in-vitro models that besides being compliant with the molecular categorization of CCAs, will mimic better the liver background whereby the malignant biliary transformation arises.

| AC TI ON S TO TACK LE CHALLENG E S OF CC A RE S E ARCH
As previously outlined, a major drawback of CCA research particularly that devoted to basic and translational studies has epidemiological grounds, since the low prevalence of the disease limits the availability of biological samples and tissues. 1,21 This disadvantage is accentuated further by the difficulties related to the generation of animal models able to faithfully recapitulate a markedly heterogeneous disease. 108,111 Given these issues, research proposals supported by an overt inter-

Implantation
Highly reproducible. Useful for studying the mechanisms of proliferation and metastasis. Use of patients-derived xenografts allows studies for personalized therapy. Syngeneic models allow studies on immunotherapy due to the use of immunocompetent animals.
Do not mimic the mechanisms of tumour development.
If not transplanted into the liver, do not mimic the normal tumour location. Do not develop on background of chronic inflammation.

GEMM
Reproduces specific genetic mutations. Develops in an immunocompetent environment allowing studies on immunotherapy. Partially mimics tumour development and transformation. Enables the study of interactions with the TME.
Expensive and laborious. Possible differences in the activation of the genetic mutation and consequent experimental variability. Do not respect the intratumoural variability of the CCA. Do not develop on background of chronic inflammation. Spheroids: Short-term maintenance. Difficult to mimic interactions with TME. Do not reproduce the histomorphology of the tumour of origin. Organoids: Expensive. Long time to generate cell cultures (weeks). Partial reproduction of the interactions within the TME.
Note: CCA, cholangiocarcinoma; GEMM, Genetically engineered mouse models; HCC, hepatocellular carcinoma; TME, tumour microenvironment; 2D, bi-dimensional; 3D, tri-dimensional. the disease, unravelling the complexity of the tumour microenvironment and its dual interplay with the tumour counterpart additionally with the aid of more reliable experimental models, these are some examples of matters to put on the CCA research agenda. In this context, prioritization of the key objectives, with definitions of timescales and evaluation of the cost-benefit ratio, are the next steps for future combined efforts. We believe that this approach may result in considerable progress in the field, and turn the multilevel heterogeneity of CCA into an exciting opportunity for personalized medicine.
Data sharing is not applicable as no new data were created or analysed in this study.