Molecular biomarkers for diagnosis, prognosis, precision therapy, and immunotherapy in cholangiocarcinoma, cHCC-CCA, and HCC Authors: Jingguo Sun, Yingjie Lu, Lu Chang, Ruiqun Gao, Ming Guan, Haoyu Ruan, Xuan Deng Published: 2026 ABSTRACT Primary liver cancers include hepatocellular carcinoma (HCC), intrahepatic cholangiocarcinoma (iCCA), combined hepatocellular–cholangiocarcinoma (cHCC-CCA), and rarer hepatic malignancies. Their biomarker landscapes differ substantially. HCC has many approved systemic therapies, yet most are still selected through clinical context, liver function, tumor burden, and treatment line. By contrast, iCCA and broader biliary tract cancers (BTCs) have fewer patients but multiple treatment-directing genomic alterations, including FGFR2 fusions or rearrangements, IDH1 mutations, HER2 positivity, BRAF V600E, NTRK or RET fusions, and microsatellite instability-high (MSI-H) or deficient mismatch repair (dMMR) status. cHCC-CCA adds a diagnostic layer because lineage classification and sampling adequacy influence whether HCC-based or BTC-based evidence is more relevant. This review organizes biomarkers by the treatment decision they are expected to support. It covers curative intent and locoregional risk stratification, molecularly guided systemic therapy, immune-state assessment, dynamic monitoring, resistance reassessment, and translational readiness. Across these settings, biomarker value depends on clinical action, evidence maturity, assay feasibility, and timing. The review separates decision-ready markers, trial-enrichment tools, monitoring assays, and mechanistic signatures. INTRODUCTION Primary liver cancer remains a major cause of cancer mortality, ranking sixth in incidence and third in cancer-related death worldwide in 2022.1 Although HCC accounts for most cases, iCCA and other BTCs are clinically important because they differ in tumor biology and treatment options.2 This heterogeneity makes biomarker-guided care in liver cancer more complex than in tumors, where a single molecular result can directly determine therapy. In HCC, treatment is still shaped by liver function, tumor burden, vascular invasion, performance status, and prior therapy. In iCCA and selected BTCs, molecular profiling has a more direct therapeutic role, including FGFR2 fusions or rearrangements, IDH1 mutations, HER2 positivity, BRAF V600E, NTRK or RET fusions, and MSI-H or dMMR status.34 The clinical meaning of a biomarker changes across the disease course. AFP may inform transplant selection, identify patients eligible for ramucirumab, or support response monitoring, but these uses require different thresholds and contexts.56 FGFR2 testing can guide targeted therapy in iCCA, whereas secondary FGFR2 alterations become relevant after FGFR inhibitor exposure.7 Spatial markers, such as tertiary lymphoid structures (TLSs), may describe the immune context but require standardized sampling and scoring before treatment selection.8 This review discusses biomarkers according to the clinical decision they support, while keeping the evidence bases for HCC, iCCA, BTC, and cHCC-CCA separate. BIOMARKERS AT CURATIVE-INTENT AND LOCOREGIONAL DECISION POINTS In HCC and iCCA, recurrence after apparently curative therapy may reflect microvascular invasion (MVI), residual viable disease, intrahepatic dissemination, microscopic positive margins, or continued carcinogenesis in an injured liver. Biomarkers should therefore clarify whether resection, transplantation, ablation, or TACE is technically feasible, and whether additional intervention, closer surveillance, or trial entry is justified (Figure 1). Clinical decision landscape of biomarkers in primary liver cancer. In HCC, biomarkers mainly support biological risk assessment, locoregional suitability, limited systemic treatment assignment, and dynamic monitoring. AFP, AFP-L3, and DCP contribute to transplant-waiting and resection risk assessment, whereas ALBI grade and Child–Pugh class function as liver-tolerance indicators for TACE, radioembolization, or SBRT rather than tumor biomarkers. AFP ≥400 ng/mL remains the clearest label-linked systemic marker in HCC through ramucirumab eligibility, while VETC and CTNNB1/β-catenin activation currently support anti-angiogenic enrichment or immune-resistance stratification rather than routine treatment assignment. In iCCA, tissue and molecular profiling have stronger treatment-directing roles, particularly FGFR2 fusion or rearrangement testing for FGFR inhibition, IDH1 mutation for ivosidenib, HER2 positivity for anti-HER2 strategies, and selected tumor-agnostic markers such as BRAF V600E, NTRK or RET fusions, and MSI-H/dMMR. ctDNA has distinct timing-dependent roles: postoperative MRD detection after local therapy and molecular response or resistance reassessment during systemic therapy. cHCC-CCA is shown as a lineage-overlap problem, emphasizing that treatment interpretation depends on adequate histology, immunophenotype, serum marker context, and DNA/RNA profiling. Abbreviations: AFP, alpha-fetoprotein; AFP-L3, Lens culinaris agglutinin-reactive alpha-fetoprotein; ALBI, albumin–bilirubin; BRAF, B-Raf proto-oncogene; cHCC-CCA, combined hepatocellular–cholangiocarcinoma; CTNNB1, catenin beta 1; ctDNA, circulating tumor DNA; DCP, des-gamma-carboxy prothrombin; dMMR, deficient mismatch repair; FGFR2, fibroblast growth factor receptor 2; HER2, human epidermal growth factor receptor 2; HCC, hepatocellular carcinoma; IDH1, isocitrate dehydrogenase 1; iCCA, intrahepatic cholangiocarcinoma; MRD, minimal residual disease; MSI-H, microsatellite instability-high; NTRK, neurotrophic receptor tyrosine kinase; RET, rearranged during transfection; RNA, ribonucleic acid; SBRT, stereotactic body radiation therapy; TACE, transarterial chemoembolization; VETC, vessels encapsulating tumor clusters. Biological risk in HCC transplantation and resection For HCC transplantation, the Milan criteria and expanded or downstaging strategies remain foundational.910 Their limitation is that size and number incompletely capture tumor biology. AFP is the most established serum marker, especially when interpreted dynamically during bridging or downstaging treatment. High AFP or inadequate AFP decline suggests aggressive biology despite acceptable imaging.11 This is particularly relevant during bridging or downstaging, where the direction and speed of AFP change can be more informative than one baseline value. AFP-L3 and DCP/PIVKA-II add complementary information; prospective waitlist data showed that AFP-L3 at least 35% and DCP at least 7.5 ng/mL outperformed AFP thresholds for predicting dropout, and dual positivity marked a high-risk subgroup.12 These markers also predict early post-transplant recurrence and may improve RETREAT-based risk estimation.13 Their strongest use is not isolated thresholding, but serial interpretation with imaging response, treatment delay, and clinical fitness. Interpretation requires assay awareness. AFP-L3 is reported as the Lens culinaris agglutinin-reactive AFP fraction, whereas DCP/PIVKA-II assays differ in design, calibration, and units.1415 Cutoffs from transplant, resection, or monitoring cohorts should not be transferred across platforms without validation.16 MVI remains a strong pathological predictor but is usually confirmed after tumor removal.17 Preoperative models combining serum markers, imaging, radiomics, and clinical variables may estimate MVI risk before treatment and guide bridging intensity, surveillance, and recurrence-prevention trials.18 Treatment response itself is also becoming a biological selection signal. A prospective phase II study of atezolizumab plus bevacizumab downstaging before transplantation suggested that selected patients beyond conventional criteria may reach transplant eligibility, although post-transplant rejection risk, washout interval, and immune monitoring remain unresolved.19 Post-resection risk and adjuvant trial enrichment After resection, serum dynamics and pathology should be interpreted together. Incomplete AFP decline may indicate occult disease in patients with high preoperative AFP, whereas perioperative AFP ratios can retain prognostic value even in AFP-negative disease.2021 In solitary early HCC, MVI or satellitosis may outweigh favorable morphology and identify patients at risk of aggressive recurrence.22 A layered postoperative model should combine serum kinetics, vascular invasion, tumor-border features, and MRD. This approach also allows adjuvant trials to enrich for patients whose recurrence risk is biologically plausible and measurable. In iCCA and extrahepatic cholangiocarcinoma, CA19-9 and CEA are commonly used after resection, but biliary obstruction, cholangitis, and drainage limit specificity.23 FGFR2, IDH1, and HER2 guide advanced disease therapy but do not yet define routine adjuvant selection.24 Postoperative circulating tumor DNA (ctDNA) is therefore better considered a trial-enrichment and MRD-monitoring tool. Biomarkers for liver-directed therapy After ablation, the immediate question is whether radiological complete response reflects biological clearance. GALAD and LAD scores, which combine AFP, AFP-L3, DCP, and demographic variables, may help detect residual viable HCC after local therapy.2526 Declining scores support local control, whereas persistent elevation or rebound should prompt closer imaging or retreatment assessment. Before TACE, biomarkers should distinguish 3 decisions: tumor controllability, liver tolerance, and the need for early transition to combination or systemic therapy. TACE response is influenced by arterial perfusion, hypoxia adaptation, angiogenic signaling, hepatic reserve, and immune context.27 A pretreatment transcriptomic signature composed of ADH1C, CXCL11, EMCN, SPARCL1, and LIN28B distinguished TACE-sensitive from TACE-resistant HCC and was linked to CT radiomics and clinical variables.27 EMERALD-1 showed that adding durvalumab and bevacizumab to TACE improved progression-free survival in unresectable HCC suitable for embolization, increasing the need to identify patients who require upfront combination therapy.28 Microbiome-derived markers, such as reduced Limosilactobacillus reuteri and indole-3-lactic acid, have also been linked to TACE-related liver injury and survival.26 These transcriptomic, radiomic, and microbiome markers remain exploratory enrichment tools. Liver tolerance markers should be separated from tumor-control markers. ALBI grade and Child–Pugh class estimate hepatic reserve and risk of decompensation during repeated TACE, yttrium-90 radioembolization, or SBRT.293031 By contrast, GALAD/LAD dynamics, arterial enhancement, viable enhancing volume, CT or MRI radiomics, and TACE-related signatures are closer to tumor controllability or early response. Preserved liver function does not guarantee local control, and an embolization-responsive tumor may still be unsuitable for repeated TACE if liver reserve is poor. In the era of TACE plus ICI or anti-VEGF therapy, the clinically relevant question is whether a patient has an embolization-sensitive tumor, a TACE-tolerant liver, and a disease state that does not already require systemic intensification because of vascular invasion, high burden, or angiogenic and immune resistance. MOLECULARLY GUIDED SYSTEMIC THERAPY Systemic therapy highlights the difference between HCC and iCCA/BTC. HCC has many approved drugs but few label-linked biomarkers, so selection remains mainly clinical.32 In iCCA and selected BTCs, molecular profiling can assign therapy through FGFR2 fusions or rearrangements, IDH1 mutations, HER2 positivity, BRAF V600E, NTRK or RET fusions, and MSI-H or dMMR status.433 cHCC-CCA requires lineage-aware interpretation because histology, sampling, and genomic profile determine whether HCC-based or BTC-based evidence is more relevant.34 Limited molecular stratification in HCC AFP-guided ramucirumab remains the clearest treatment-assignment example in HCC. REACH-2 prospectively enrolled patients with advanced HCC after sorafenib and AFP at least 400 ng/mL, making this threshold drug-specific.6 Other HCC markers are better viewed as enrichment candidates. FGF19 amplification or overexpression can activate FGFR4 through beta-Klotho and support FGFR4-inhibitor trials, but amplification, overexpression, FGFR4 expression, and beta-Klotho positivity are not equivalent.35 Fisogatinib and newer agents such as irpagratinib illustrate this enrichment approach. MET shows the same need for precision: broad MET overexpression failed as a strategy with tivantinib, whereas high-level MET amplification is rarer but closer to true dependency.3637 EGFR activation has been linked to lenvatinib resistance and supports EGFR-high enrichment in resistance-focused trials, without establishing first-line selection.3839 Prevalence estimates depend on assay and cohort. FGF19 amplification has been reported in about 10%–20% of some Asian HCC cohorts, while FGF19/FGFR4 expression is more frequent and less specific for dependence.4041 High-level MET amplification is uncommon, often around 1%–5%, and EGFR-high status is assay-defined, whereas EGFR mutation or high-level amplification is rare.3742 Vessels that encapsulate tumor clusters (VETC) are a distinctive vascular pattern in which tumor cell clusters are surrounded by endothelial-lined vessels. It captures an organized vascular phenotype and may benefit from anti-angiogenic strategies such as tyrosine kinase inhibitors or bevacizumab-containing regimens.4344 Because VETC assessment depends on histological architecture, it is more vulnerable to sampling bias than serum or genomic biomarkers. Rare tumor-agnostic markers, including MSI-H, dMMR, NTRK or RET fusions, BRAF V600E, and TMB-high status, are uncommon in HCC but can connect selected patients to approved tumor-agnostic indications or basket trials.324546 Broad profiling is therefore most defensible in atypical HCC, young patients, mixed histology, discordant imaging, or progression after standard systemic therapy. Actionable genomic drivers in iCCA iCCA and broader BTCs have a more mature treatment-assignment framework. FGFR2 fusion-positive iCCA is the clearest matched-therapy paradigm.47 Pemigatinib and infigratinib established the feasibility of FGFR inhibition after prior therapy, and futibatinib showed durable activity through irreversible FGFR1-4 inhibition.484950 The report should distinguish FGFR2 fusion or rearrangement from FGFR2 mutation, amplification, or expression, because these findings have different therapeutic meanings and different assay requirements. IDH1-mutant cholangiocarcinoma is another mature biomarker-drug pair; ivosidenib improved progression-free survival and showed an overall survival signal despite crossover.51 Acquired MAPK alterations are most relevant during resistance reassessment and should not be used for baseline treatment assignment. HER2 is clinically important, especially in gallbladder cancer and extrahepatic cholangiocarcinoma, and may occur in iCCA or mixed tumors.5253 Trastuzumab-based therapy, zanidatamab, and trastuzumab deruxtecan have strengthened HER2-directed strategies, but reports should specify IHC 3+, IHC 2+ with ISH positivity, ERBB2 amplification, or HER2-low status in a drug-specific context.5253545556 HER2 prevalence varies by anatomical subtype and assay definition, with higher rates generally reported in gallbladder cancer and extrahepatic cholangiocarcinoma than in iCCA.5758 Apparent geographic differences should therefore be interpreted with attention to tumor-site composition, antibody platform, and drug-specific positivity criteria. PTEN deficiency represents a treatment-enrichment marker because it may predict proteasome-inhibitor or gemcitabine sensitivity.5960 KRAS variants, MDM2 amplification, CLDN18.2, SRC dependency, and co-mutation patterns remain promising but not decision-ready.616263 Their current value is to refine biological heterogeneity within established classes, explain variable response, and support prespecified trial cohorts. cHCC-CCA and lineage-aware profiling cHCC-CCA is difficult because hepatocellular and cholangiocarcinoma components vary, and small biopsies may miss one lineage. AFP and CA19-9 can be discordant, and imaging may resemble either HCC or iCCA.64 Genomic profiling helps because cHCC-CCA is not molecularly identical to conventional HCC; some datasets show higher frequencies of IDH1, IDH2, and FGFR2 alterations, and iCCA-enriched genomic findings can prompt pathological reassessment.53 A pediatric case with FGFR2-PRDM16 fusion responding to pemigatinib plus chemotherapy illustrates why RNA fusion testing may matter in selected mixed tumors, although it is not generalizable.64 A practical cHCC-CCA report should include 4 layers. The first is lineage confidence, based on morphology, hepatocellular and cholangiocytic immunophenotype, AFP, CA19-9, and imaging. The second is treatment-directing alteration, including FGFR2, IDH1, HER2, BRAF V600E, NTRK, or RET fusions and MSI-H/dMMR when a BTC-like component is suspected. The third is discordance management; HCC-like imaging with CA19-9 elevation, cholangiocarcinoma-like morphology with high AFP, or an HCC diagnosis with iCCA-enriched alterations should trigger pathology re-review and expanded DNA/RNA profiling when tissue allows. The fourth is actionability reporting, stating dominant lineage, sampling confidence, assay limits, and the evidence base most applicable to treatment. This is a lineage-aware evidence assignment, not a validated cHCC-CCA companion diagnostic. Overall, HCC biomarkers currently function mainly as enrichment, monitoring, or resistance-assessment tools, whereas iCCA/BTC biomarkers more often assign treatment. The practical workflow should combine histology and immunophenotype, DNA and RNA profiling, and phenotype assignment. Operationally, cHCC-CCA reports should not present an actionable alteration without a lineage statement. A useful report should specify whether the tested tissue is HCC-dominant, cholangiocarcinoma-dominant, genuinely mixed, or indeterminate; whether the biopsy may have under-sampled one component; and whether the alteration is supported by HCC evidence, BTC evidence, a tumor-agnostic indication, or only a trial rationale. This distinction is important because an FGFR2 fusion or IDH1 mutation in a tumor initially diagnosed as HCC should prompt pathology re-review, whereas AFP elevation in a BTC-like tumor should not automatically override morphology. BIOMARKERS PREDICTING IMMUNOTHERAPY RESPONSE Except for MSI-H or dMMR disease, checkpoint regimens in primary liver cancers are still selected mainly by clinical context. Immunotherapy biomarkers should be separated into baseline immune sensitivity, tumor-intrinsic or spatial resistance, BTC combination enrichment, and antigen-defined strategies. HCC immune sensitivity beyond standard checkpoint markers In HCC, programmed death-ligand 1 (PD-L1) expression, TMB, MSI/MMR status, and peripheral inflammatory indices have limited independent value for treatment assignment. PD-L1 depends on the antibody clone, scoring system, compartment, biopsy site, and inflammatory background.65 MSI-H, dMMR, and TMB-high HCC are rare, and many responders lack these features. Regimen-specific signatures are more informative. In the atezolizumab plus bevacizumab setting, translational analyses linked immune genomic signatures with benefit by integrating immune activation, vascular biology, and myeloid suppression.6667 These signatures support immune-state classification and trial enrichment, not routine companion diagnostics. This distinction avoids overstating single immune markers in a disease where cirrhosis, viral hepatitis, metabolic liver disease, and prior locoregional therapy can all reshape baseline inflammation. CTNNB1 or beta-catenin assessment should also be used carefully. DNA sequencing, beta-catenin nuclear staining, and glutamine synthetase immunohistochemistry capture related but non-identical pathway states. These assays may support negative enrichment or trial stratification, especially for ICI monotherapy, but atezolizumab plus bevacizumab data are mixed, so beta-catenin activation should not be used as a simple exclusion rule.68697071 When reported clinically, the result should indicate whether pathway activation was inferred from DNA mutation, nuclear protein localization, glutamine synthetase staining, or transcriptomic state. Tumor morphology may add immune information. Infiltrative HCC treated with first-line atezolizumab plus bevacizumab showed lower response and survival and was associated with TP53/ATM loss, EMT, TGF-beta activation, and immunosuppressive programs.72 Its practical role is therefore triage for deeper immune and stromal profiling. HCC immune resistance strategies via negative enrichment and spatial stratification HCC immune resistance can be grouped by clinical use. Negative-enrichment markers include CTNNB1/beta-catenin activation and should be reported with assay type. Mechanism-based candidates include tumor-intrinsic programs such as YAP activation, BCL9-enhanced Wnt signaling, and lipid remodeling; etiology-related immune dysfunction such as MASH-associated CD8 T-cell dysfunction; and microenvironmental programs such as macrophage polarization, neutrophil reprogramming, CAF-associated exclusion, and galectin-4.7374757677787980 These markers support biological stratification and combination-trial design. Pathway-to-combination examples include MMP9-mediated CD8 T-cell exclusion in CTNNB1-mutant HCC and the PVRL1/PVR/TIGIT axis, both of which support biologically selected combination trials.8182 Spatial organization determines whether infiltration becomes productive. TLS location, maturity, composition, and relation to the invasive front are more informative than a present-or-absent label. In neoadjuvant immunotherapy, intratumoral TLS density was associated with pathological response and relapse-free survival.8 In resected HCC treated with adjuvant PD-1 inhibitors, TLS-positive tumors were associated with longer recurrence-free and overall survival, and an ACKR1, CCR7, and IL7R TLS-related signature was linked to prognosis and possible immunotherapy response.8384 These data suggest a role for TLS in postoperative immunotherapy enrichment, but they do not yet establish a universal cutoff for adjuvant PD-1 treatment. In cHCC-CCA, intratumoral and mature TLS appear more favorable than adjacent-tissue TLS.85 Spatial markers therefore require standardized region selection, maturity scoring, and reproducible image analysis before prospective use. From an implementation perspective, TLS scoring should record whether structures are intratumoral, peritumoral, or adjacent to non-tumor liver; whether they contain B-cell follicles, follicular dendritic cell networks, high endothelial venules, and T-cell zones; and whether immune cells can access tumor nests or remain trapped behind stromal or vascular barriers. This matters because a TLS-rich adjacent liver may reflect chronic inflammation, whereas mature intratumoral TLS more directly supports local antitumor immunity. These details should be captured in trial protocols before TLS is used as an enrichment biomarker. iCCA and BTC immune context and combination enrichment TOPAZ-1 and KEYNOTE-966 established gemcitabine-cisplatin plus durvalumab or pembrolizumab as first-line options for broad BTC populations, but they did not define a predictive biomarker comparable to FGFR2, IDH1, or HER2.8687 PD-L1 combined positive score, TMB, tumor site, and inflammatory indices may be subgroup variables, but no validated cutoff currently selects patients for first-line chemoimmunotherapy.33868788 MSI-H or dMMR remains the clearest immunotherapy-assignment marker in BTC, although it is uncommon and varies by etiology; liver-fluke-associated Thai cholangiocarcinoma has shown higher MSI-H rates than many Japanese or Western non-liver-fluke cohorts.89909192 Baseline molecular testing should therefore retain MSI/MMR assessment even when the expected yield is low, because a positive result has direct therapeutic implications. Other markers are regimen-specific enrichment tools. In IMbrave151, high VEGFA expression was nominated as a potential bevacizumab-benefit signal.93 In ZSAB-TOP, TIGIT and PD-L1 co-expression was proposed to enrich response to tislelizumab plus ociperlimab with chemotherapy.94 In later-line anti-angiogenic and ICI combinations, homologous recombination deficiency (HRD) and inflamed tumor signals may identify patients more likely to benefit from sitravatinib plus tislelizumab.95 These cutoffs should be reported as exploratory thresholds until independent prospective datasets show that selecting by the marker improves outcome. Assay detail is essential because gene-expression thresholds, IHC co-expression, HRD definitions, and inflammatory signatures are not interchangeable across studies. Immune resistance in BTC is shaped by myeloid, CAF/stromal, and tumor-intrinsic programs. Examples include MARCO-positive macrophages, soluble CD109-related CD73-positive macrophage states, KRAS-COX-PGE2-PMN-MDSC biology, GPBAR1-CXCL10 CAF signaling, desmoplastic stroma, PARG-related crosstalk, CMTM4-mediated PD-L1 regulation, and YAP activation.96979899100 These mechanisms can guide combination trials but are not routine treatment-selection biomarkers. Antigen-defined immunotherapy as target selection Some liver cancers fit an antigen-defined model in which the marker is the therapeutic target. Fibrolamellar HCC is driven by DNAJ-PKAc, which can serve as a tumor-specific antigen.101 A phase I peptide-vaccine trial with nivolumab and ipilimumab induced vaccine-specific T-cell responses and clinical activity in some patients.102 HBV-related HCC provides another example: in HLA-A*02:01-positive and HBsAg-positive advanced disease, HBV-specific TCR-T cells showed antiviral and antitumor activity.103 Here, eligibility depends on viral antigen expression, HLA restriction, safety monitoring, and tumor accessibility. Personalized neoantigen vaccination combined with pembrolizumab also induced neoantigen-specific T-cell responses in advanced HCC.104 GPC3-directed mRNA-encoded bispecific T-cell engagers and NKG2D ligand-targeting CAR-macrophages remain earlier strategies in which the biomarker defines an antigen or stress-ligand state.105106 DYNAMIC MONITORING AND ADAPTIVE REASSESSMENT Biomarker-guided care continues after baseline assignment. Apparent radiological control may coexist with MRD, and targeted therapy failure may reflect on-target resistance or bypass signaling. Dynamic monitoring should therefore be framed as adaptive reassessment. ctDNA is used here for postoperative MRD and for on-treatment molecular response or resistance during systemic therapy. The same assay can support different actions depending on timing: closer imaging after local therapy, resistance testing during targeted therapy, or trial entry when an actionable escape clone appears. MRD after curative-intent therapy After resection, ablation, or transplantation, MRD monitoring asks whether local control has produced a molecularly silent state. In HCC, ctDNA MRD is complicated because recurrence may arise from residual tumor clones, intrahepatic dissemination, or de novo carcinogenesis in the diseased liver.107108109110 A single negative result cannot eliminate future risk. Tumor-informed or targeted assays tracking TERT promoter, TP53, CTNNB1, copy-number changes, chromatin-remodeling alterations, or HBV integration-derived viral-host DNA may identify patients at higher molecular risk despite negative imaging.109111112 Their main current role is surveillance refinement and trial enrichment. This is important because adjuvant escalation based only on molecular positivity has not yet shown a survival advantage in HCC. Cholangiocarcinoma provides a clearer serial model. In the STAMP biomarker cohort of resected extrahepatic cholangiocarcinoma, ctDNA positivity before or during adjuvant chemotherapy predicted inferior disease-free survival, while persistent negativity identified the most favorable group.23 Serial change is more informative than a single postoperative value. CA19-9 and CEA remain useful but are confounded by obstruction, cholangitis, drainage, and inflammatory injury.113114 Bile or serum exosomal circRNA panels may add specimen-specific information when bile is accessible.115 Intervention trials are still needed to test whether MRD-guided escalation or de-escalation improves survival. For HCC, MRD studies should also distinguish recurrence patterns. Early, multifocal, same-clone relapse is more consistent with residual disease, whereas late de novo tumors may arise from the cirrhotic field and may not be captured by the original tumor-informed assay. Serial testing and clonal comparison with recurrent tissue are, therefore, more informative than a single postoperative result. This distinction determines whether ctDNA positivity should trigger intensified surveillance, adjuvant trial entry, or molecular reassessment at recurrence. Early response and resistance during systemic therapy During systemic therapy, imaging can lag behind biological response because of necrosis, immune-related changes, mixed response, and cirrhotic parenchymal alterations. In HCC, serial ctDNA during atezolizumab plus bevacizumab showed that mutation persistence was associated with radiological progression, whereas mutation disappearance suggested response.116 Total cfDNA requires caution because ablation, TACE, radiotherapy, inflammation, liver injury, and impaired clearance can all increase circulating DNA.117 PD-L1-positive CTC burden may indicate systemic immune-evasive disease during PD-(L)1 blockade, and targeted therapy may indicate systemic immune-evasive disease and has been associated with poorer outcomes.118 In iCCA, monitoring is most useful when therapy targets a driver. Plasma assays can detect FGFR2 fusions in many tissue-positive cases and can track response or resistance during FGFR inhibition.119120 Progression should be classified as on-target or bypass escape because the next therapeutic question differs: persistent on-target dependence may justify a next-generation inhibitor, whereas bypass signaling supports combination or alternative pathway trials. FGFR2-altered cholangiocarcinoma develops polyclonal kinase-domain, gatekeeper, and molecular-brake mutations as well as RTK-MAPK bypass.121122123 Selective FGFR2 inhibitors such as lirafugratinib may retain activity against some resistance mutations but select alternative profiles.124125 IDH1-mutant disease can acquire MAPK alterations during ivosidenib.126 In HCC, NGF-TrkA activation, ERBB3-mediated regorafenib efflux, AOX1-linked ferroptosis susceptibility, the HLF/c-Jun axis, and ERK/YAP1-CDK6 upregulation remain resistance-enrichment hypotheses for trials.127128129130131 Table 1 summarizes these resistance mechanisms and the corresponding reassessment actions. TABLE 1Resistance mechanisms and adaptive reassessment during systemic therapy Therapy or pathwayResistance mechanism or reassessment signalDetection approachClinical actionFGFR inhibitors in FGFR2-altered iCCA/BTCPolyclonal FGFR2 kinase-domain mutations; gatekeeper and molecular-brake changes; RTK-MAPK bypass; selective FGFR2 inhibition may select new resistance profiles.Baseline and progression tissue or plasma DNA/RNA NGS; ctDNA clone tracking when tumor shedding is adequate.Classify progression as on-target or bypass escape; repeat profiling; consider next-generation FGFR inhibition or mechanism-based trials rather than assuming persistent FGFR dependence.IDH1 inhibitor in IDH1-mutant cholangiocarcinomaMAPK pathway alterations can emerge during ivosidenib treatment and attenuate immune-interferon signaling.Progression tissue or plasma NGS compared with baseline IDH1 status.Renew molecular assessment at progression; consider trial enrollment or rational combinations when bypass escape is detected.HCC tyrosine kinase inhibitors, including lenvatinib, sorafenib, and regorafenibLenvatinib resistance can involve EGFR activation, NGF-TrkA signaling, and CDK6 upregulation; sorafenib resistance has been linked to the HLF/c-Jun axis; regorafenib resistance may involve ERBB3-HIF1A-ABCB1 signaling; AOX1 may mark ferroptosis-related TKI vulnerability.Paired tissue or RNA/IHC studies when feasible; plasma exploratory assays; organoid or functional testing when sequencing suggests uncertain dependence.Use these findings as trial-enrichment and paired-sample study signals. They support combination hypotheses but should not trigger routine drug switching without clinical validation.ICI or ICI-based combinations in HCC and BTCCTNNB1/beta-catenin activation, MMP9-mediated CD8+ T-cell exclusion, PVRL1/PVR/TIGIT signaling, spatial exclusion, TLS context, and PD-L1+ CTCs describe immune resistance or dynamic immune-evasive states.Tissue DNA/RNA sequencing; beta-catenin and glutamine synthetase IHC; spatial pathology; CTC phenotyping; selected plasma monitoring.Report the assay and immune context; avoid excluding atezolizumab plus bevacizumab solely by beta-catenin status; enrich TIGIT, MMP9, spatial, or CTC-informed trials and prompt closer reassessment when dynamic immune-evasive signals appear.Abbreviations: AOX1, aldehyde oxidase 1; ABCB1, ATP-binding cassette subfamily B member 1; BTC, biliary tract cancer; cHCC-CCA, combined hepatocellular–cholangiocarcinoma; CDK6, cyclin-dependent kinase 6; CTNNB1, catenin beta 1; CTC, circulating tumor cell; ctDNA, circulating tumor DNA; ERBB3, erb-b2 receptor tyrosine kinase 3; FGFR, fibroblast growth factor receptor; FGFR2, fibroblast growth factor receptor 2; HCC, hepatocellular carcinoma; HIF1A, hypoxia-inducible factor 1 subunit alpha; HRD, homologous recombination deficiency; ICI, immune checkpoint inhibitor; IDH1, isocitrate dehydrogenase 1; iCCA, intrahepatic cholangiocarcinoma; IHC, immunohistochemistry; MAPK, mitogen-activated protein kinase; MMP9, matrix metallopeptidase 9; NGF, nerve growth factor; NGS, next-generation sequencing; NTRK, neurotrophic receptor tyrosine kinase; PD-L1, programmed death-ligand 1; PVR, poliovirus receptor; PVRL1, poliovirus receptor-related 1; RNA, ribonucleic acid; RTK, receptor tyrosine kinase; TIGIT, T-cell immunoreceptor with Ig and ITIM domains; TLS, tertiary lymphoid structure; TKI, tyrosine kinase inhibitor; VETC, vessels encapsulating tumor clusters. Resistance mechanisms and adaptive reassessment during systemic therapy Abbreviations: AOX1, aldehyde oxidase 1; ABCB1, ATP-binding cassette subfamily B member 1; BTC, biliary tract cancer; cHCC-CCA, combined hepatocellular–cholangiocarcinoma; CDK6, cyclin-dependent kinase 6; CTNNB1, catenin beta 1; CTC, circulating tumor cell; ctDNA, circulating tumor DNA; ERBB3, erb-b2 receptor tyrosine kinase 3; FGFR, fibroblast growth factor receptor; FGFR2, fibroblast growth factor receptor 2; HCC, hepatocellular carcinoma; HIF1A, hypoxia-inducible factor 1 subunit alpha; HRD, homologous recombination deficiency; ICI, immune checkpoint inhibitor; IDH1, isocitrate dehydrogenase 1; iCCA, intrahepatic cholangiocarcinoma; IHC, immunohistochemistry; MAPK, mitogen-activated protein kinase; MMP9, matrix metallopeptidase 9; NGF, nerve growth factor; NGS, next-generation sequencing; NTRK, neurotrophic receptor tyrosine kinase; PD-L1, programmed death-ligand 1; PVR, poliovirus receptor; PVRL1, poliovirus receptor-related 1; RNA, ribonucleic acid; RTK, receptor tyrosine kinase; TIGIT, T-cell immunoreceptor with Ig and ITIM domains; TLS, tertiary lymphoid structure; TKI, tyrosine kinase inhibitor; VETC, vessels encapsulating tumor clusters. Choosing tissue, plasma, or bile A specimen-aware workflow is more realistic than a single liquid-biopsy strategy. Tissue remains the baseline anchor when histology, lineage state, HER2 protein status, spatial architecture, or mixed differentiation affects treatment selection. Plasma ctDNA is the most practical repeatable source for systemic reassessment, and in FGFR2 fusion-positive iCCA, plasma detected FGFR2 fusion events in 16 of 18 tissue-positive patients in one targeted-therapy cohort, while lower ctDNA levels during treatment correlated with clinical benefit.119 Negative plasma results still require caution because low tumor fraction, low shedding, and fusion-detection limits can cause false reassurance. Bile-derived material addresses a different question. In malignant biliary strictures, bile cfDNA sequencing improved diagnostic sensitivity compared with standard pathology, and endoscopic retrograde cholangiopancreatography (ERCP)-obtained DNA/RNA NGS can combine malignancy detection with actionable alteration discovery.132133 Its clinical use depends on access through ERCP or drainage, infection and obstruction status, sampling site, and rapid specimen handling. For cHCC-CCA, tissue is essential for lineage assignment, while plasma or bile may reveal clones missed by a small biopsy when the anatomy and tumor shedding make these assays informative. TRANSLATING BIOMARKERS INTO DECISION-READY TOOLS Biomarker translation fails when markers are biologically plausible but analytically inconsistent, analytically feasible but clinically inactive, or predictive in retrospective models without proof that acting on them improves outcome. The practical framework should combine the decision node, evidence category, and assay feasibility. Table 2 summarizes how the same biomarker class changes meaning across curative, locoregional, systemic, immune, monitoring, and implementation decisions. This decision-node framework is intended to reduce repeated caveats and make the clinical consequence of each assay explicit. TABLE 2Decision-node evidence matrix for biomarker-guided treatment in primary liver cancersDecision nodeHCCiCCA/BTCcHCC-CCAReadiness questionCurative-intent risk Tier: EN/MON Markers: AFP dynamics; AFP-L3; DCP; MVI/radiomics; ctDNA MRD Platform: serum; pathology; imaging; plasma Tier: EN/MON Markers: CA19-9/CEA; postoperative ctDNA; proteogenomic risk Platform: serum; tissue; plasma Tier: CLS/EN Markers: lineage-aware pathology; dominant component; AFP/CA19-9 discordance; sampling adequacy Platform: histology; IHC; serum; DNA/RNA when diagnosis is discordantCan the result identify patients whose recurrence risk changes transplant selection, surveillance intensity, or postoperative trial entry?Locoregional therapy Tier: EN/MON Markers: tolerance markers including ALBI grade and Child–Pugh class; tumor-control markers including arterial enhancement, viable enhancing volume, CT/MRI radiomics, GALAD/LAD dynamics, and TACE transcriptomic signatures Platform: liver-function scores; serum; imaging; selected tissue or transcriptomic assays Tier: ME Markers: limited routine role; local control mainly anatomy driven Platform: imaging; tissue when available Tier: ME/CLS Markers: histologic phenotype and dominant component for HCC-like or BTC-like local strategy Platform: histology; imaging; IHC when neededDoes the result separate liver tolerance from tumor controllability before TACE, radioembolization, SBRT, or combination therapy?Systemic treatment assignment Tier: LB/EN Markers: AFP ≥400 ng/mL for ramucirumab; FGF19-FGFR4; MET amplification; VETC; EGFR-high Platform: serum; tissue; histology Tier: DR/LB Markers: FGFR2 fusion; IDH1 mutation; HER2-positive disease; BRAF V600E; NTRK/RET fusions; MSI-H/dMMR Platform: tissue DNA/RNA NGS; IHC/ISH Tier: CLS/EN; DR/LB if alteration is label-linked Markers: dominant lineage; HCC-like vs. BTC-like evidence assignment; actionable iCCA-like alterations, including FGFR2, IDH1, and HER2; tumor-agnostic markers, including MSI-H/dMMR, BRAF V600E, NTRK or RET fusions Platform: histology; IHC; AFP/CA19-9; tissue DNA/RNA NGS; RNA fusion testing if rearrangement is suspectedIs the alteration linked to an approved drug or label-defined treatment window that is feasible for the patient?Immunotherapy selection Tier: EN/ME Markers: PD-L1/TMB limited; CTNNB1/beta-catenin negative enrichment; TLS and spatial exclusion for trials Platform: IHC; DNA/RNA; spatial pathology Tier: DR for MSI-H/dMMR; EN for others Markers: MSI-H/dMMR; VEGFA; TIGIT/PD-L1 co-expression; HRD Platform: NGS; MMR IHC; RNA/IHC Tier: ME/EN Markers: TLS maturity; lineage context; antigen or stress-ligand expression Platform: tissue and spatial assaysDoes the result assign immunotherapy, enrich a defined combination, or mainly explain immune escape biology?Dynamic monitoring and resistance Tier: MON for ctDNA/CTC; ME for resistance programs Markers: ctDNA response; cfDNA context; PD-L1+ CTCs; NGF-TrkA; ERBB3; AOX1; CDK6 Platform: plasma ctDNA/cfDNA; CTC; paired tissue Tier: MON/EN Markers: FGFR2 resistance mutations; IDH1-MAPK escape; bile or plasma ctDNA; exosomal circRNA Platform: plasma; tissue; bile Tier: ME Markers: repeat tissue for lineage shift; plasma or bile for missed actionable or resistant clones Platform: tissue; plasma; bileCan the result justify earlier imaging, rebiopsy, resistance testing, trial entry, or a validated next therapy?AI and multimodal implementation Tier: MD Markers: MVI radiomics; TACE-response models; organoids; plasma dynamics Platform: CT/MRI; pathomics; organoids; plasma Tier: MD Markers: radiogenomics for FGFR2/IDH1/HER2; tissue adequacy prediction; bile/plasma triage Platform: imaging; pathology; NGS Tier: ME Markers: multimodal lineage classifier Platform: histology; genomics; imagingDoes the model direct confirmatory testing or a clinical action after external validation and calibration?Abbreviations: AFP, alpha-fetoprotein; AFP-L3, Lens culinaris agglutinin-reactive alpha-fetoprotein fraction; ALBI, albumin–bilirubin; BRAF, B-Raf proto-oncogene; BTC, biliary tract cancer; CA19-9, carbohydrate antigen 19-9; cHCC-CCA, combined hepatocellular–cholangiocarcinoma; CEA, carcinoembryonic antigen; CLS, classification; cfDNA, cell-free DNA; CTNNB1, catenin beta 1; CTC, circulating tumor cell; ctDNA, circulating tumor DNA; DCP, des-gamma-carboxy prothrombin; dMMR, deficient mismatch repair; DR, decision-ready; EN, enrichment; FGFR2, fibroblast growth factor receptor 2; GALAD, gender, age, AFP-L3, AFP, and DCP; HCC, hepatocellular carcinoma; HER2, human epidermal growth factor receptor 2; HRD, homologous recombination deficiency; IHC, immunohistochemistry; IDH1, isocitrate dehydrogenase 1; iCCA, intrahepatic cholangiocarcinoma; ISH, in situ hybridization; LAD, Lens culinaris agglutinin-reactive AFP, AFP, and DCP; MD, model-dependent; ME, mechanistic or exploratory; MMR, mismatch repair; MSI-H, microsatellite instability-high; MVI, microvascular invasion; MON, monitoring; NGS, next-generation sequencing; NTRK, neurotrophic receptor tyrosine kinase; PD-L1, programmed death-ligand 1; RET, rearranged during transfection; RNA, ribonucleic acid; SBRT, stereotactic body radiation therapy; TACE, transarterial chemoembolization; TGIТ, T-cell immunoreceptor with Ig and ITIM domains; TLS, tertiary lymphoid structure; TMB, tumor mutational burden; VEGFA, vascular endothelial growth factor A; VETC, vessels encapsulating tumor clusters. Decision-node evidence matrix for biomarker-guided treatment in primary liver cancers Abbreviations: AFP, alpha-fetoprotein; AFP-L3, Lens culinaris agglutinin-reactive alpha-fetoprotein fraction; ALBI, albumin–bilirubin; BRAF, B-Raf proto-oncogene; BTC, biliary tract cancer; CA19-9, carbohydrate antigen 19-9; cHCC-CCA, combined hepatocellular–cholangiocarcinoma; CEA, carcinoembryonic antigen; CLS, classification; cfDNA, cell-free DNA; CTNNB1, catenin beta 1; CTC, circulating tumor cell; ctDNA, circulating tumor DNA; DCP, des-gamma-carboxy prothrombin; dMMR, deficient mismatch repair; DR, decision-ready; EN, enrichment; FGFR2, fibroblast growth factor receptor 2; GALAD, gender, age, AFP-L3, AFP, and DCP; HCC, hepatocellular carcinoma; HER2, human epidermal growth factor receptor 2; HRD, homologous recombination deficiency; IHC, immunohistochemistry; IDH1, isocitrate dehydrogenase 1; iCCA, intrahepatic cholangiocarcinoma; ISH, in situ hybridization; LAD, Lens culinaris agglutinin-reactive AFP, AFP, and DCP; MD, model-dependent; ME, mechanistic or exploratory; MMR, mismatch repair; MSI-H, microsatellite instability-high; MVI, microvascular invasion; MON, monitoring; NGS, next-generation sequencing; NTRK, neurotrophic receptor tyrosine kinase; PD-L1, programmed death-ligand 1; RET, rearranged during transfection; RNA, ribonucleic acid; SBRT, stereotactic body radiation therapy; TACE, transarterial chemoembolization; TGIТ, T-cell immunoreceptor with Ig and ITIM domains; TLS, tertiary lymphoid structure; TMB, tumor mutational burden; VEGFA, vascular endothelial growth factor A; VETC, vessels encapsulating tumor clusters. Assay context and analytical validity Analytical validity should be reported as a decision-specific assay statement. HER2 testing should specify IHC score, ISH, or NGS amplification status, specimen type, and drug context. FGFR2 testing should state whether fusion detection used DNA, RNA, or plasma, and whether the assay captures diverse partners. ctDNA reports should include tumor fraction, assay coverage, fusion-detection capability, and the possibility of false-negative results in low-shedding disease. Spatial markers such as VETC, TLS, invasive-front exclusion, and cHCC-CCA lineage patterns depend on tissue depth, sampling area, staining, image analysis, and compartment definition.43 Validation must define tissue region, scoring unit, minimum sample requirement, reader reproducibility, and the treatment decision linked to the score. Clinical validity and utility Clinical utility should be judged by action. A decision-ready marker assigns an approved therapy or tumor-agnostic indication, such as FGFR2 fusion testing for FGFR inhibitors in iCCA. A label-linked marker defines eligibility for a specific label, such as AFP at least 400 ng/mL for ramucirumab. Enrichment markers identify patients for a strategy or trial without determining routine care. Monitoring markers support surveillance, response assessment, or resistance reassessment. Mechanistic markers nominate combinations. Model-dependent markers require external validation, calibration, and a defined confirmatory action, while classification markers support lineage or diagnosis. This framework prevents AFP-L3, ctDNA MRD, CTNNB1 activation, TLS, VETC, and FGFR2 fusions from being treated as equivalent simply because all are biomarkers. Real-world precision oncology studies show where actionability is lost. The French Genomic Medicine Initiative identified actionable alterations in many advanced primary liver cancers, but only a subset received matched therapy, and the benefit concentrated in ESCAT I–III alterations.32 Actionability, therefore, depends on evidence tier, drug access, tissue adequacy, patient fitness, liver function, and timing. A technically positive result has little value if the report returns after the treatment window has closed or if hepatic reserve prevents the proposed therapy. A biomarker program should be judged by how often results return early enough to guide feasible treatment. Turnaround time is part of clinical utility. A perfect molecular result that returns after first-line therapy has started, after liver function has deteriorated, or after tissue has been exhausted may no longer change care. Implementation studies should therefore report sample failure rate, time from biopsy or blood draw to result, proportion of patients receiving a matched action, and reasons for non-action, including inadequate liver reserve, drug access, competing clinical urgency, or patient preference. These metrics make biomarker translation auditable. Heterogeneity and implementation Primary liver cancers are vulnerable to sampling failure. HCC arises in a chronically injured liver, so circulating markers may reflect tumor burden, liver injury, and clearance capacity. A single biopsy can miss VETC-positive regions, TLS, stromal barriers, immune-excluded fronts, or resistant subclones. cHCC-CCA is especially prone to under-sampling and incomplete lineage assignment. Treatment pressure adds temporal heterogeneity through FGFR2 kinase-domain mutations, IDH1-associated MAPK bypass, and HCC TKI resistance states. Reports should be clinically interpretable. BTC reports should distinguish FGFR2 fusion from FGFR2 mutation or overexpression, HER2 IHC 3+ from ERBB2 amplification, and MSI-H/dMMR from broader immune context. HCC reports should separate AFP-guided ramucirumab eligibility from exploratory FGF19-FGFR4 activation, high-level MET amplification, VETC, and resistance states. Reports should include evidence tier, drug availability, specimen limitations, turnaround time, and next action for positive, negative, or inconclusive results. For example, a negative plasma test with inadequate tissue is not equivalent to a true negative comprehensive tissue result, and an inconclusive lineage assignment in cHCC-CCA should lead to re-review or repeat sampling when clinically feasible. Artificial intelligence and multimodal readiness Multimodal models can integrate serum markers, histology, radiomics, pathomics, genomics, transcriptomics, ctDNA, bile-derived DNA, and spatial features. Their output must be interpretable because predictions may reflect tumor biology, non-tumor liver background, acquisition artifacts, or confounding. Organoid pharmacogenomics can complement classifiers when treatment dependence is uncertain and may help trial selection when feasible.134 Current applications remain decision support: radiomics for preoperative MVI risk, radiogenomics to prioritize FGFR2 or IDH1 confirmation in iCCA, and organoid or plasma-based models to refine ambiguous dependencies.134135136 None should replace confirmatory pathology, molecular testing, or clinical judgment unless prospective action trials demonstrate benefit. TRIPOD+AI and PROBAST+AI provide reporting and bias-assessment standards.137138 A model that predicts MVI is useful only if it changes resection, transplant, or surveillance planning. A TACE-failure model is useful only if it guides earlier combination therapy or systemic transition. A virtual biopsy model should guide confirmatory testing for HER2, FGFR2, CTNNB1, or immune phenotype. For ctDNA, AI may improve variant calling and noise suppression, but cirrhosis, inflammation, treatment necrosis, and impaired clearance are major confounders. AI outputs should therefore include a recommended confirmatory action, such as repeat imaging, tissue confirmation, RNA fusion testing, resistance profiling, or referral to a molecularly matched trial. CONCLUSION The next research agenda should move from marker discovery to action testing. Each study should begin with the decision, assay, time point, threshold, and clinical action that the result will trigger. A biomarker should be considered ready only when acting on it improves selection, surveillance, treatment sequencing, or trial allocation. For HCC, 3 unresolved questions deserve priority. First, CTNNB1 or beta-catenin activation needs prospective evaluation in atezolizumab plus bevacizumab and other combination settings, with sequencing and immunohistochemistry reported together. Second, ctDNA MRD needs intervention trials that test whether positive results justify intensified imaging, adjuvant escalation, or trial entry. Third, locoregional biomarkers should separate liver tolerance from tumor controllability so that TACE, radioembolization, SBRT, and upfront combinations are chosen on distinct evidence. For iCCA and BTC, implementation is the main bottleneck. Trials should test tissue-first workflows with RNA fusion detection, HER2 assay harmonization, plasma fallback when tissue is inadequate, and resistance testing at progression. Enrichment markers such as PTEN deficiency, VEGFA expression, TIGIT and PD-L1 co-expression, and homologous recombination deficiency should move forward only when the companion action and cutoff are defined before enrollment. For cHCC-CCA, the immediate need is better assignment of lineage and dominant biology. Future studies should integrate histology, immunophenotype, DNA and RNA profiling, plasma or bile assays when informative, and spatial context to decide when HCC-based, BTC-based, antigen-defined, or molecularly matched evidence applies. AI, virtual biopsy, organoids, spatial assays, and ctDNA should be treated as decision support. They need calibrated reporting, external validation, turnaround metrics, and a recommended confirmatory action. A useful biomarker report should state evidence tier, specimen limitation, drug access, and the next feasible step for positive, negative, or inconclusive results.