Recurrence Is the Biggest Problem in HCC Surgery Surgical resection offers the best chance of cure for early-stage hepatocellular carcinoma (HCC), but recurrence rates are high - over 50% within five years. Identifying which patients will recur quickly would allow oncologists to consider adjuvant therapies, shorten surveillance intervals, or select patients for transplantation over resection.
Published Signatures Have Not Been Validated Head-to-Head Dozens of gene expression signatures predicting HCC recurrence have been published, but each was derived and tested in a different cohort, making it impossible to know which - if any - were truly superior. This study directly compared 22 published signatures in a unified analysis.
The Study Design Using gene expression data from 287 HCC patients, this study evaluated 22 published prognostic signatures and two multi-signatures (G3 and a poor-survival combination) using the Nearest Template Prediction (NTP) method and random survival forests, asking which signatures independently predicted recurrence.
A Path to Clinical Translation Rather than proposing yet another signature, the study's goal was to determine which existing signatures have the strongest and most reproducible evidence - the kind of rigorous comparative analysis needed before any signature can be recommended for clinical use.
287 Patients from Multiple Institutions The cohort included 287 patients who underwent surgical resection for HCC, with gene expression data from tumor samples. Patients came from multiple institutions, increasing the heterogeneity and real-world applicability of the findings.
The Nearest Template Prediction Framework Each of the 22 published signatures was applied to individual patient samples using NTP, which assigns each sample to 'high risk' or 'low risk' based on how closely its expression profile resembles the signature template. This single-sample approach avoids the need to re-batch samples across cohorts.
Multivariable Survival Analysis Signatures that showed univariate prognostic value were then tested in multivariable Cox regression models that included established clinical prognostic factors (vascular invasion, tumor size, AFP, cirrhosis). Only signatures that remained significant after adjusting for clinical factors were considered clinically additive.
Random Survival Forests In addition to Cox regression, random survival forests - a machine learning approach that can capture non-linear interactions - were used to assess signature performance. Consistency of results across these two methodologies increased confidence in the final findings.
Most Signatures Fail in Independent Validation Of the 22 published signatures tested, most showed weak or no independent prognostic value in this cohort. This is a sobering finding - many published signatures that appeared promising in their original studies do not replicate when applied to an independent dataset with a different patient population and analytical framework.
G3 Signature Is an Independent Predictor The G3 signature - originally defined by Boyault and colleagues as a molecular subclass of HCC characterized by chromosomal instability and aggressive biology - showed significant independent prognostic value (hazard ratio 1.75, p=0.003). Patients assigned to the G3 class had dramatically higher recurrence rates.
Poor-Survival Signature Also Validates A second signature, the 'poor-survival' signature derived from a meta-analysis of HCC prognostic studies, also independently predicted recurrence (hazard ratio 1.74, p=0.004) after adjusting for clinical covariates. Both signatures added information beyond what clinical variables alone could predict.
Combining Both Signatures Interestingly, combining the G3 and poor-survival signatures into a joint prediction model provided additional stratification, identifying a very high-risk group of patients with markedly worse outcomes. This suggests the two signatures capture partially distinct biological programs.
G3 Reflects Chromosomal Instability The G3 molecular class was originally defined by transcriptomic features associated with chromosomal instability, TP53 mutation, and loss of hepatic differentiation markers. G3 HCCs tend to be poorly differentiated, AFP-secreting tumors that behave aggressively - making biological sense that this class would predict early recurrence.
The Poor-Survival Signature Captures Tumor Proliferation The poor-survival meta-signature largely reflects tumor proliferative activity, capturing the expression of cell-cycle genes that are upregulated in rapidly growing tumors. High proliferative activity is one of the strongest predictors of poor outcome across many cancer types.
Non-Tumor Liver Gene Expression An intriguing component of the poor-survival signature is that it partly reflects gene expression in the surrounding non-tumor liver tissue, not just in the tumor itself. This suggests that the state of the cirrhotic liver microenvironment - already known to influence HCC prognosis - is captured in expression data from tumor samples that include some contaminating normal tissue.
Overlap and Complementarity The partial overlap and partial complementarity between G3 and poor-survival signatures support a model in which HCC prognosis is determined by multiple distinct biological axes - chromosomal instability and TP53 dysfunction on one hand, and proliferative activity and microenvironment on the other.
Molecular Profiling Adds to Clinical Staging The finding that G3 and poor-survival signatures predict recurrence independently of clinical variables means that molecular profiling provides information that current staging systems (BCLC, TNM) do not capture. This supports the potential clinical utility of gene expression profiling in resected HCC.
Identifying Patients for Intensified Surveillance Patients classified as high-risk by both signatures have dramatically elevated recurrence rates and could benefit from more frequent imaging surveillance, allowing earlier detection of recurrence when salvage treatments remain possible.
Selecting Patients for Adjuvant Trials Patients identified as high-risk by molecular profiling represent an ideal population for enrollment in adjuvant therapy trials, where an effective treatment would produce the largest absolute benefit. Molecular stratification could improve the efficiency and statistical power of such trials.
Transplant vs. Resection Decision For patients who meet criteria for both liver resection and transplantation, molecular profiling could help guide the choice. High-risk patients may derive greater benefit from transplantation (which removes the entire diseased liver) than from resection alone.
FFPE Tissue Compatibility The study used fresh-frozen tissue, but most routine clinical specimens are formalin-fixed, paraffin-embedded (FFPE). Demonstrating that the G3 and poor-survival signatures perform equally well in FFPE material would be essential for routine clinical adoption.
Prospective Validation Required This was a retrospective analysis. Prospective validation in a dedicated clinical trial - where samples are collected and processed according to a standardized protocol before outcome is known - is needed before these signatures can be used to guide treatment decisions.
Integration with Liquid Biopsy The logistical challenges of performing gene expression profiling on tumor tissue could be bypassed if the prognostic information captured in these signatures were also detectable in blood (e.g., from circulating tumor DNA or circulating RNA). This represents a major future research direction.
Functional Targets Within the Signatures Understanding which specific genes within the G3 and poor-survival signatures are functionally important - drivers of the aggressive phenotype rather than merely associated with it - could reveal therapeutic targets for adjuvant treatment.