Hepatocellular Carcinoma: Molecular Diagnosis and Perspectives for Therapy

Hepatocellular carcinoma (HCC) has an increasing incidence over the last 2 decades and more than a sixth have advanced HCC at presentation.^1,2 In such cases, systemic therapy remains the treatment of choice with options including tyrosine kinase inhibitors and immunotherapy. There is rapidly evolving knowledge in the field of systemic therapy with a number of medications approved over the last 4 years. Thus, a multidisciplinary approach with a focus on reliable biomarkers to predict treatment response and stratify patients at risk of progression is needed.^3,4 In this issue of the Journal of Clinical and Experimental Hepatology, Marboli et al. describe molecular signature-based machine learning model for hepatocellular carcinoma (HCC). The model includes a laboratory panel for etiology based on hepatitis B virus (HBV), hepatitis C virus (HCV), or non-viral cause (NBNC) associated HCC markers, and alpha fetoprotein as a standard tumor biomarker in combination with micro-RNA-related markers. These novel molecular and clinical-based machine learning models may improve diagnostic algorithms for HCC. The study utilized 5 machine-learning classifiers on the training dataset to predict the presence of HCC.⁵ The k-nearest neighbors (KNN) classifier was initialized with four neighbors and utilized the Euclidean distance metric for proximity calculation.² Random Forest (RF) classifier, an ensemble of 100 decision trees with a maximum depth of five levels.³ The support vector machine (SVM) classifier with a polynomial kernel (‘poly’) and regularization parameter C = 1 captured complex data relationships.⁴ Light gradient boosting machine (LGBM) classifier with gradient-boosting algorithm and⁵ deep neural network (DNN) model with a 3-layer architecture: a dense layer with 128 neurons and ‘relu’ activation, followed by another dense layer with 64 neurons and ‘relu’ activation. The LGBM classifier performed better than all other models in terms of diagnostic accuracy, precision, recall, and discriminant power. Although machine learning is in an early stage, and the exponential growth in the field may soon be available on a mass scale for accurate analysis of HCC prediction. However, the predictive models are only as accurate as the training dataset. There is considerable tumor heterogeneity in this analysis as the model had just 102 patients with HCC, which was compared with 98 healthy controls and 67 patients with non-malignant liver disease. The micro-RNA signatures selected are only useful in some population subsets. The model also does not account for tumor biology, incorporating only AFP in the component analysis. Although encouraging, the machine learning algorithms need to be standardized for real world practice.

In another study by Patkar et al., 10 patients with HCC underwent transarterial embolization (TAE). The diagnosis was based on characteristic imaging in 6 and on biopsy in 4 individuals. They were assessed for another molecular test for HCC diagnosis-circulating tumor cells, and three miRNAs, i.e. miR-885-5p, miR-22-3pmiR-885-5p, and miR-22-3p. Changes in signature were assessed prior to TAE and reassessed 30 days after the procedure. In this small study, the presence of CTC postoperatively did not have any impact on survival or recurrence.⁶ Therefore, larger studies are required to make any inference on this test, while keeping cost concerns in context.

Another clinicopathological study on HCC tumor biology evaluated 35 HCCs from Indonesia and 41 HCCs from Japan. The authors assessed biliary/stem cell (B/S) markers (cytokeratin 19, sal-like protein 4, epithelial cell adhesion molecule) and Wnt/β-catenin molecules (β-catenin, glutamine synthetase) with evaluation of T and B lymphocytes in tumor tissue. Of interest were vessels that encapsulated tumor clusters, no significant differences were found in the proportions of biliary or stem cells Wnt/β-catenin-based subgroups. Immune-low tumors were more common than in Japanese cases, which may affect response to systemic therapy.⁷ The importance of tumor biology in HCC treatment response cannot be emphasized enough.⁸

These data from 3 different global regions suggest that HCC diagnosis and management require personalized assessment, but the same molecular model will not fit all. As per current data, up to 25% of all HCCs have actionable mutations, which can determine the type of systemic therapy. For example, lenvatinib acts against the fibroblast growth factor receptor, platelet-derived growth factor receptor, and vascular endothelial growth factor pathways. Vascular endothelial growth factor A (VEGFA) gene mutations have a 5% prevalence rate in HCC, and drugs like bevacizumab (VEGFA) and ramucirumab (VEGFR2) can be selected accordingly.⁹ HCC have telomerase reverse transcriptase (TERT) promoter mutations in about 60%, which results in increased telomerase expression. Other commonly affected genes are TP53 and CTNNB1, found in 25%–30% of HCC patients. Unfortunately, TERT, CTNNB1, and TP53 have no potential therapeutic targets currently. Besides these, identification of AXIN1, ARID1A, ARID2, TSC1/TSC2, RPS6KA3, KEAP1, and MLL2 mutations indicate the core pathways affected in HCC and inform the prognosis.¹⁰

Particularly, cost concerns and lack of standardization of these techniques highlighted in this issue of JCEH, such as machine-learning based on composite clinical and molecular signatures, histopathological assessment of inflammatory type HCC, and microRNA or non-coding RNA biomarkers, has prevented these tests from entering the clinical domain.¹¹ The relevance of these techniques of personalized treatment allocation strategies must be validated in global datasets with sufficient accuracy and a reasonable cost-diagnostic yield ratio. Specifically in India, we have limited availability of systemic therapy; therefore, the molecular tests used should either be useful for accurate surveillance or be used for tailoring systemic therapy.⁴ The rising incidence of HCC in India, with an age adjusted incidence rate in men between 0.7 and 7.5 and for women between 0.2 and 2.2 per 100,000 population annually, is a cause for alarm.¹² The INASL guidelines for management of HCC have incorporated newer therapeutics like immunotherapy as first-line treatment of advanced HCC,^3,13 and the ability of molecular signatures to guide treatment decisions can impact the outcomes of patients in India.¹⁴ The experience from India on management of HCC with drugs like atezolizumab and bevacizumab is relatively new.¹⁴ All three studies included in this issue of JCEH represent preliminary data or pilot studies. Much larger datasets are required to further validate the arguments or clinical applications of newer molecular tests as adjunct to current imaging and biomarker-based treatment algorithms.

Credit authorship contribution statement

MP and YC were both involved in the manuscript preparation. Both the authors have read and approved the manuscript.

Funding

None.

Disclosures

Neither of the authors have potential conflicts (financial, professional, or personal) which are relevant to this manuscript.

Declaration of competing interest

Nothing to report.