Cell-cycle and DNA-replication programs are among the most consistently upregulated signals in HCC. TOP2A, CCNB1, RRM2, AURKA and CDK1 often appear together within mitotic, chromosome-segregation and replication-stress modules (14–19). Their recurrence makes them useful markers of a proliferative state. It does not, however, mean that each gene is a primary driver or a stand-alone therapeutic target.

TOP2A participates in resolving topological stress during DNA replication and chromosome segregation. In HCC and other cancers, high TOP2A expression generally tracks with proliferation and aggressive biology, which provides a rationale for considering topoisomerase-related mechanisms in an appropriate experimental setting (40). CCNB1 and cyclin B2 mark cell-cycle transition and mitotic entry. AURKA is involved in centrosome and spindle regulation, and cyclin-dependent kinase 4/6-related evidence illustrates that cell-cycle perturbation can intersect with immune-related effects in HCC models (41). RRM2 links DNA-replication demand with nucleotide metabolism and replication stress and has been examined as a candidate HCC vulnerability in integrated in silico and in vitro work (42). Read as a module, these genes describe a cycling tumour state more reliably than any single marker.

The risk is familiar in bioinformatics studies: a frequently upregulated cell-cycle gene can be quickly promoted into an independent target, even though high expression may instead reflect tumour purity, the fraction of cycling cells, sampling of an aggressive region or a shared proliferation program. A more defensible use of the signal is to select assays and models. EdU incorporation, colony formation, propidium iodide (PI)-based cell-cycle profiling and protein-level assessment of CCNB1, AURKA, TOP2A or RRM2 provide stronger evidence than expression differences alone (14–19,28–34).

For that reason, cell-cycle genes should first be treated as state markers. Therapeutic language is justified only after functional dependency, pathway engagement and model-consistent effects have been shown.

The other arm of the paired pattern is loss of hepatocyte identity. Downregulated transcripts often include transporters, detoxification enzymes, complement and coagulation genes, metabolic regulators and other liver-enriched functions. Representative examples include SLC22A1, SLCO1B3, CYP-family genes and GLS2. These signals matter because they influence tumour classification, drug uptake, metabolic competence and model interpretation (20–27).

SLC22A1 encodes organic cation transporter 1 (OCT1), which is relevant to hepatic uptake of endogenous metabolites and several drugs (20–22). CYP enzymes and SLCO-family transporters add a second pharmacological layer by linking the expression state to metabolism and liver-specific exposure. GLS2, a glutaminase associated with mitochondrial metabolism and redox balance, may represent a liver-metabolic component that should be interpreted separately from generic proliferation-associated nutrient demand (23–27).

This matters clinically because loss of transporter and metabolic programs can change how a candidate intervention is interpreted. Reduced OCT1/SLC22A1 or SLCO expression may limit intracellular drug exposure, whereas altered CYP and detoxification programs can shift assumptions about dose relevance and toxicity. In HCC, where cirrhosis and impaired liver function are common, transporter competence and metabolic identity are part of the pharmacological context rather than background annotation (20–27).

The same point applies to experimental interpretation. A compound may reduce proliferation in a poorly differentiated monolayer cell line without testing transporter-dependent uptake, hepatocyte metabolism or liver-specific toxicity. Conversely, restoration of one liver-function transcript is not enough to claim restored hepatocyte identity; stronger language requires coordinated changes across transporter, metabolic and functional readouts in models suited to the question (36–39).

Taken together, loss of hepatocyte identity is not only a diagnostic or descriptive signal. It changes the meaning of drug exposure, model suitability and liver-relevant pharmacology. A downregulated signature can be listed in a table, but its practical value lies in deciding what the model can and cannot test.

Proliferation-identity imbalance is a convenient name for a recurring, but not universal, HCC contrast. Replication and mitotic programs tend to rise as differentiated liver functions weaken. The phrase is useful because it puts two findings that are often discussed separately into the same frame and slows the common move from ‘gene is differentially expressed’ to ‘gene is a mechanism or target’.

The concept does not flatten HCC heterogeneity. A tumour with intense cell-cycle activity and marked loss of liver-function transcripts may behave differently from a tumour with moderate proliferation but stronger immune or metabolic remodeling. The model is therefore a guide for comparison, model choice and claim wording, not a new universal class of HCC.

Fig. 1 gives a schematic version of the idea. It should not be read to mean that proliferation and hepatocyte identity always change in lockstep. Its main use is more modest: Examining both arms together is usually more informative than judging either arm in isolation.

Used in that limited sense, the proliferation-identity frame helps organize recurring expression patterns while leaving room for subtype, etiology and microenvironmental variation (11,12,14–19,28–34).

Bulk transcriptomics is useful for detecting reproducible signals, but it averages over cell types. A change seen in a bulk HCC sample may come from malignant hepatocytes, stromal cells, immune cells, endothelial cells or the injured non-tumour liver. Single-cell and spatial approaches can separate some of these sources, which is particularly important for immune, complement, matrix and endothelial programs (28–34).

Recent single-cell and spatial studies make this point concrete. Some prognostic or candidate therapeutic signals are concentrated in malignant-cell subsets, immune interfaces or tumour-host metabolic niches rather than spread evenly through the tumour mass (28–34). For bulk-signature work, this means that the cellular source of a signal should be checked before it is used to justify a target, model or validation assay.

Spatial location also matters. HCC arises in a diseased organ, and the tumour core, invasive margin and cirrhotic liver around the tumour may carry different expression states. A tumour-vs.-adjacent comparison compresses these differences into one average. In translational work, that average can be misleading if the candidate signal is absent from malignant cells or is mainly stromal or immune (28–34).

A more defensible workflow is therefore iterative: start with stable bulk or multi-cohort patterns; use single-cell or spatial evidence to ask where the signal originates; then choose a validation model because it preserves the relevant state rather than because it is convenient (14–19,28–34,36–39). Fig. 2 presents this high-level model-selection logic.

Single-cell and spatial data are valuable when they change the interpretation or the experiment. If they are presented without a link to the cell type, tissue location or assay being tested, their contribution is limited (28–34).

Transcriptomic patterns can suggest candidate vulnerability hypotheses, but the evidence level must remain explicit. Cell-cycle activation points toward cyclin-dependent kinase, Aurora kinase, checkpoint, mitotic-spindle and replication-stress biology. An expression pattern alone is still not a validated target. A vulnerability claim requires a functional phenotype, dose-response behaviour, pathway or protein readout and, preferably, consistency across more than one HCC model (19,41,42).

Metabolic remodeling in HCC encompasses altered glycolysis, lipid synthesis, amino-acid metabolism, redox homeostasis and mitochondrial function, although the contribution of each program varies across tumours and experimental contexts (23–27). These changes can intersect with proliferation because a dividing tumour requires nucleotides, lipids, amino acids and reducing power. Metabolic signals should therefore be interpreted together with proliferative state. A useful translational question is whether the signal implies nutrient dependence, redox vulnerability, transporter-mediated exposure or immune suppression that can be tested (23–27).

Model selection is the third translational axis. Huh7, PLC/PRF/5, Hep3B, MHCC97H and other HCC models differ in differentiation state, TP53 status, viral background, transporter expression and proliferation rate (43). HepG2 is frequently used in liver-cancer experiments, but it has been shown to be a hepatoblastoma-derived cell line rather than a hepatocellular carcinoma cell line (44); it should therefore not be regarded as a representative HCC model. Model choice should follow the expression state under study rather than convenience alone. For example, Table II distinguishes a high cell-cycle module, for which a proliferative model and functional perturbation assays are required, from a low-transporter module, for which transporter-competent models and uptake testing are needed before altered drug exposure can be inferred.

Patient-derived tumour organoids, patient-derived xenograft models and related patient-derived systems can partly bridge this gap because they retain more genomic, histological or treatment-response context than conventional monolayer cultures (36–39,43). They are particularly useful when the question involves transporter competence, metabolic phenotype, intratumoural heterogeneity or drug-sensitivity ranking. Even these models need caution: organoid culture success, stromal and immune loss, medium composition and growth-factor dependence can alter the cell state being tested.

A model suitable for testing a mitotic phenotype may therefore be inadequate for evaluating transporter-dependent drug exposure, hepatocyte-identity restoration or microenvironment-dependent biology (36–39,43).

A practical RT-qPCR panel needs both arms of the proposed state. TOP2A, CCNB1, RRM2 and AURKA can represent proliferation, whereas SLC22A1, SLCO1B3, GLS2 and CYP1A2 can represent hepatocyte identity or metabolic-function loss. The panel does not replace RNA-seq; its purpose is to test, at a manageable scale, whether a perturbation shifts a coordinated expression state rather than one isolated gene (14–27).

Validation should be scaled to the claim. A viability assay alone supports only preliminary growth inhibition. If EdU incorporation and cell-cycle profiling also change, the interpretation can move toward proliferation or cell-cycle modulation. If selected module genes shift in the expected direction by RT-qPCR or RNA-seq, it becomes reasonable to discuss a module-level transcriptional response. Mechanistic or translational wording should wait for protein or pathway readouts, target perturbation, model extension and, where relevant, in vivo support (14–19,36–43).

Fig. 3 and Table III present a validation ladder rather than a mandatory checklist. Fig. 3 provides the stage sequence, whereas Table III specifies the assay categories, the evidence each stage can support and the limitations that remain. The purpose is to prevent a common shortcut: treating altered expression or a short viability experiment as if it already proved a mechanism (14–19,36–43).

In practical terms, a claim should move stepwise. First comes phenotype screening, followed by proliferation and cell-cycle assessment, then a coordinated transcriptional response, and only then protein, pathway or target-level support. A single viability assay, or one gene measured alone, can justify a preliminary observation, not a mechanistic conclusion (14–19,36–43).

A concrete example can be drawn from the TCGA-LIHC cohort, a widely used bulk transcriptomic resource for HCC (7). In this dataset-level setting, the first step is not to rank a single gene as the most important finding, but to ask whether the tumour profile shows a coordinated state shift. Published TCGA-LIHC and integrated HCC analyses consistently identify TOP2A, CCNB1 and RRM2 as components of proliferation-associated programmes (11,12,14–16). In contrast, reduced albumin and transthyretin expression reflects diminished hepatocyte differentiation (14,15), whereas reduced SLC22A1/OCT1 and SLCO-family transporter expression has been documented in HCC (20–22). Under the proposed framework, this pattern is interpreted as a proliferation-identity imbalance rather than as evidence that any one highly ranked gene is automatically a driver or drug target.

The second step is to interrogate cellular origin. A TCGA-LIHC tumour-sample bulk signal averages malignant hepatocytes, immune cells, endothelial cells and fibroblasts, In tumour-versus-adjacent tissue analyses, adjacent diseased liver constitutes a separate comparator and may influence the observed differential signal. Therefore, the proliferation-identity pattern should be cross-checked against single-cell or spatial evidence. If proliferative genes are concentrated in malignant epithelial compartments and hepatocyte-identity transcripts are reduced in those same compartments, the bulk signal supports tumour-cell state remodeling. If the apparent signal is mainly stromal, immune or derived from adjacent cirrhotic tissue, the validation question changes and a conventional HCC cell-line experiment may not test the relevant biology (28–34).

The third step is state-matched model selection. If the TCGA-LIHC-like signal represents a high-proliferation and low-hepatocyte-identity tumour state, the first experimental model should preserve strong cycling activity and allow cell-cycle readouts. Such a model can be used to test EdU incorporation, PI-based cell-cycle distribution, colony formation and protein expression of CCNB1, CDK1, TOP2A, RRM2 or PCNA. By contrast, if the working hypothesis concerns transporter-dependent drug exposure or restoration of hepatocyte metabolism, the model should retain measurable transporter and metabolic competence, such as selected differentiated models, organoids or patient-derived systems (36–39,43). The model is therefore chosen because it matches the state being tested, not simply because it is available.

The fourth step is to apply the validation ladder. For TCGA-LIHC example, a first tier would assess phenotype, including growth or survival. A second tier would examine proliferation and cell-cycle readouts, such as EdU incorporation, PI-based cell-cycle distribution or colony formation assays. A third tier would confirm the coordinated mRNA pattern by RT-qPCR or RNA-seq using both proliferation markers and hepatocyte-identity markers. A fourth tier would examine protein or pathway readouts, such as CCNB1, CDK1, PCNA, TOP2A, RRM2, OCT1/SLC22A1 or selected CYP proteins when appropriate. Stronger mechanistic claims would require perturbation or rescue experiments, and translational claims would require extension across additional models or patient-derived systems (14–19,36–43). This example shows how a bulk TCGA-LIHC-type DEG list can be converted into a state-aware experimental plan rather than a direct therapeutic conclusion.

Interpretation can drift for several reasons. Adjacent liver is not a neutral control in many HCC datasets; it may be cirrhotic, inflamed, steatotic, fibrotic or virally injured. Tumour purity also changes bulk immune, stromal, endothelial, complement and matrix signals. Platform differences add another layer because microarray and RNA-seq studies differ in probe design, normalization, dynamic range and annotation. Original studies can use meta-analysis, robust rank aggregation or direction-consistency filters to reduce noise, but none of these steps removes the underlying biology and sampling issues (7,11,12,14–19,28–34).

Hub genes create another trap. Genes that sit near the centre of protein-interaction or co-expression networks are often essential, abundant or well studied. That does not make them uniquely disease-driving. Similarly, a high-risk prognostic signature may describe aggressive tumours without defining a core HCC mechanism or a general therapeutic vulnerability (14–19).

Validation can be too thin as well. One cell line, one dose and one or two genes may be enough for a first observation, but not for a coordinated mechanism. These limits do not weaken transcriptomics; they mark the point where interpretation should stop (36–39,43).

Expression patterns are most useful when readers can see the cohort context, platform constraints, cell-type origin and validation depth. Without that information, stable-looking signatures can easily harden into therapeutic stories that the data do not yet support (7,14–19,28–34,36–39,43).

Future work should join robust bulk signatures with cell-type resolution more deliberately. Multi-cohort analyses can show which patterns recur, whereas single-cell and spatial studies can show where those patterns arise. Stronger experimental designs will then choose a cell line, organoid, co-culture or xenograft because it retains the state under study, not because it is the nearest available model (14–19,28–34,36–39,43).

A first recommendation is to report transcriptomic claims in evidence tiers. A recurrent DEG module should be described as a state hypothesis. A coordinated RT-qPCR or RNA-seq response can support module-level transcriptional modulation. Protein, pathway and phenotype assays are needed before mechanistic language is used. Perturbation, rescue and model extension are needed before translational vulnerability claims are justified (14–19,36–43).

A second recommendation is to make model-selection rationale explicit. For proliferation-associated modules, the model should retain measurable cell-cycle activity. For hepatocyte-identity or transporter modules, the model should preserve relevant uptake, detoxification or metabolic features. For immune, complement or stromal modules, the model should include the cellular context required to test the signal. This recommendation is especially important because HCC cell lines, organoids and xenografts preserve different aspects of tumour biology (28–34,36–39,43).

For translation, background liver disease and drug handling need to be part of the same discussion. A proliferative module may point to cell-cycle or replication-stress testing, but transporter loss, CYP remodeling and weak hepatocyte identity can affect intracellular exposure and liver-specific toxicity. Future studies should therefore report not only whether tumour cells are inhibited, but also why the chosen model can support the pharmacological interpretation (20–27,36–39,43).

Future translational strategies should remain hypothesis-generating until they are tested in HCC-relevant models with state-matched readouts. Transcriptomic modules can prioritize questions about cell-cycle, replication-stress or metabolic dependencies, but they do not establish therapeutic efficacy by themselves (19,23–27,40–42).

Reporting habits also need to improve. Gene lists, preprocessing choices, ranking methods, cohort composition, model-selection rationale and validation status should be visible to readers. Negative validation results are worth reporting because they show where public transcriptomic predictions fail. The language should follow the evidence: a recurrent cell-cycle signal supports a proliferative-state hypothesis; it does not make every gene in the module a target. Loss of transporter or metabolic transcripts supports hepatocyte-identity erosion; it does not by itself prove drug resistance (7,14–19,28–34,36–39,43).

Across HCC transcriptomes, a recurring paired pattern is evident: Cell-cycle and DNA-replication programs often rise while differentiated hepatocyte transport, metabolic and defense functions decline. The proliferation-identity frame offers a practical way to read that pattern, select experimental models and plan validation without overstating what expression data can prove.

Expression data are most useful when they lead to testable biological questions. They are weakest when treated as proof of mechanism or therapeutic effect. A state-aware reading asks a limited set of questions before stronger claims are made: Which cell state is being represented, which model retains it, what phenotype is expected to change and what evidence would be sufficient for the claim.