Liquid Biopsy in Hepatocellular Carcinoma: Where Are We Now?

Simple Summary Hepatocellular carcinoma (HCC) is one of the mostly lethal cancers, with a prognosis which is still very poor. Novel reliable biomarkers, useful in early diagnosis and prognosis assessment, are urgently needed in order to improve HCC patient survival. In recent years, several studies focused on liquid biopsy, the molecular analysis of circulating cancer by-products, as a source of novel biomarkers. Extracellular vesicles, circulating tumor cells, cell-free DNA and non-coding RNA provided very interesting results in a large number of studies published recently, but none of them has entered the clinical routine. In this review we will summarize the available evidence on these novel circulating biomarkers as diagnostic, prognostic, and predictive tools. Liquid biopsy proved to be a very useful source of biomarkers, some of which will probably be applied soon in clinical practice. Abstract Hepatocellular carcinoma (HCC) is one of the leading causes of cancer related death worldwide. Diagnostic, prognostic, and predictive biomarkers are urgently needed in order to improve patient survival. Indeed, the most widely used biomarkers, such as alpha-fetoprotein (AFP), have limited accuracy as both diagnostic and prognostic tests. Liver biopsy provides an insight on the biology of the tumor, but it is an invasive procedure, not routinely used, and not representative of the whole neoplasia due to the demonstrated intra-tumoral heterogeneity. In recent years, liquid biopsy, defined as the molecular analysis of cancer by-products, released by the tumor in the bloodstream, emerged as an appealing source of new biomarkers. Several studies focused on evaluating extracellular vesicles, circulating tumor cells, cell-free DNA and non-coding RNA as novel reliable biomarkers. In this review, we aimed to provide a comprehensive overview on the most relevant available evidence on novel circulating biomarkers for early diagnosis, prognostic stratification, and therapeutic monitoring. Liquid biopsy seems to be a very promising instrument and, in the near future, some of these new non-invasive tools will probably change the clinical management of HCC patients.


Introduction
According to the International Agency for Research on Cancer, in 2018 primary liver tumors ranked as the sixth most common cancer and the fourth leading cause of cancerrelated death worldwide [1]. These figures are predicted to increase in the coming decades and it is estimated that more than 1 million people will die due to liver cancer in 2030 [2]. Hepatocellular carcinoma (HCC) account for 85% of all primary hepatic malignancies. The majority of HCC cases occur in patients with underlying liver diseases, mainly due to chronic hepatitis B or C virus (HBV and HCV) infections, alcohol abuse, aflatoxin exposure, or non-alcoholic liver disease (NAFLD) [3]. Despite the recommendation of all available guidelines to apply a regular surveillance in patients at risk, HCC is often diagnosed in advanced stages when curative therapies are no longer feasible. As a consequence, despite the remarkable progresses in therapy, the prognosis of HCC patients remains dismal, with a 5-years survival rate ranging around 20% [4].
Currently, according to guidelines, liver biopsy has a limited role in the management of HCC patients. This is due to the fact that, in patients with liver cirrhosis, a non-invasive diagnosis in the presence of typical imaging features (hypervascularity in the arterial phase and wash-out in portal venous and/or delayed phases) has high specificity. On the other hand, biopsy is indicated for patients without cirrhosis or for cirrhotics with lesions not showing the peculiar and specific radiologic appearance [5]. In most cases liver biopsy, which is associated with a small but still present risk of bleeding and tumor seeding, is unnecessary. Nevertheless, the debate on a more widespread use of liver biopsy is still open [6], with the expansion in recent years of therapeutic possibilities and in consideration of the identification of molecular markers of susceptibility to available systemic treatments, in an attempt of tailoring first and subsequent lines of therapy [7]. However, a high degree of spatial and temporal heterogeneity is present in HCC. Some somatic mutations occur early during tumorigenesis and propagate in many clones, whereas later mutations are present only in some clones (spatial heterogeneity) [8]. Moreover, different therapies select rare mutants and treatment-resistant clones, leading to the development of several genetic backgrounds at different times (temporal heterogeneity) [9,10]. Therefore, a single biopsy is unlikely to represent the entire biology of the tumor, thus limiting the utility of tissue sampling, beyond confirming the diagnosis [11].
The European Association for the Study of the Liver (EASL) recognizes as an urgent unmet need the identification of reliable biomarkers, for risk stratification and early HCC detection, prediction of prognosis, and of response to therapy (in particular to systemic treatments) [5]. Despite its unsatisfactory performance in early diagnosis and prognostication [12][13][14][15][16], alpha-fetoprotein (AFP) is still the most widely used biomarker in the clinical management of patients with HCC. Other protein biomarkers, such as des-λcarboxyprothrombin [17], glypican-3 [18], osteopontin [19], Golgi protein-73 [20], and squamous cell carcinoma antigen [21][22][23] have been evaluated, with erratic results. In the spectrum of circulating molecules derived from the primary tumor ("HCC circulome"), other biomarkers emerged as appealing tools in overcoming the limitations of conventional biomarkers and of tissue biopsy in diagnosis and prognosis. Liquid biopsy is defined as the molecular analysis of circulating cancer by-products, such as extracellular vesicles (EVs), circulating tumor cells (CTCs), and circulating tumor nucleic acids ( Figure 1). In recent years, a large evidence has been published, paving the way for the use of liquid biopsy as a source of reliable biomarkers for early tumor detection, prognostic stratification, disease monitoring and evaluation of response to treatment. Considering that these non-invasive biomarkers will probably revolutionize the management of patients with HCC in the near future, with this review we aimed to provide a comprehensive overview of the most relevant available data on the role of liquid biopsy in HCC. (Adapted from Labgaa et al. [24]).

Circulating Nucleic Acids
Circulating nucleic acids, released in the bloodstream through active secretion or following apoptosis, necrosis or lysis of tumor cells and circulating tumor cells, can be subgrouped in "cell-free DNA" (cfDNA) and "cell-free RNA" (cfRNA). cfDNA can be found in circulation as short nucleosome-associated fragments or long fragments incapsulated in EVs, while cfRNA is usually detected in association with proteins, proteolipid complexes, and EVs due to its relative instability [25].
The analysis of circulating nucleic acids represents a very promising liquid biopsy strategy for getting information on liver tumors. Beyond the utility in risk prediction, early detection, and monitoring treatment response, cfDNA and cfRNA are optimal candidates for tumor molecular profiling. Unlike tumor biopsy, their ability to mirror tumor heterogeneity represents a powerful tool to identify point mutations, aberrant methylation and chromosomal aberrations conferring drug resistance and guiding molecular target therapy [26].

Cell-Free DNA
The original discovery of cfDNA from sera of healthy individuals dates back to 1948. Following the demonstration of high serum concentration of cfDNA in patients with gastrointestinal cancers [27], its potential role as tumor marker emerged when KRAS mutations were identified in cfDNA from patients with colorectal and pancreatic cancers [28][29][30]. From this starting point, a large number of studies has been conducted focusing on the utility of cfDNA analysis also in HCC (Table 1).

Cell-Free DNA Amount and Integrity
The easiest way to use circulating DNA as a biomarker is through the evaluation of its total amount, since a high level of cfDNA in blood reflect cancer growth and tumor burden [31][32][33][34][35][36][37][38]. In 2006, Iizuka et al. [31] demonstrated that cfDNA was able to identify HCC in a cohort of HCV positive patients with a sensitivity of 69.2% and a specificity of 93.3% (AUC = 0.90), both higher than those of AFP. These early results are in line with previous data from our research group: the total amount of cfDNA achieved a sensitivity of 91%, a specificity of 43%, and an AUC of 0.69 in discriminating HCC from CLD and cirrhotic patients [34]. Since cfDNA is not specific for liver cancer, several studies reported (Adapted from Labgaa et al. [24]).

Circulating Nucleic Acids
Circulating nucleic acids, released in the bloodstream through active secretion or following apoptosis, necrosis or lysis of tumor cells and circulating tumor cells, can be subgrouped in "cell-free DNA" (cfDNA) and "cell-free RNA" (cfRNA). cfDNA can be found in circulation as short nucleosome-associated fragments or long fragments incapsulated in EVs, while cfRNA is usually detected in association with proteins, proteolipid complexes, and EVs due to its relative instability [25].
The analysis of circulating nucleic acids represents a very promising liquid biopsy strategy for getting information on liver tumors. Beyond the utility in risk prediction, early detection, and monitoring treatment response, cfDNA and cfRNA are optimal candidates for tumor molecular profiling. Unlike tumor biopsy, their ability to mirror tumor heterogeneity represents a powerful tool to identify point mutations, aberrant methylation and chromosomal aberrations conferring drug resistance and guiding molecular target therapy [26].

Cell-Free DNA
The original discovery of cfDNA from sera of healthy individuals dates back to 1948. Following the demonstration of high serum concentration of cfDNA in patients with gastrointestinal cancers [27], its potential role as tumor marker emerged when KRAS mutations were identified in cfDNA from patients with colorectal and pancreatic cancers [28][29][30]. From this starting point, a large number of studies has been conducted focusing on the utility of cfDNA analysis also in HCC (Table 1).

Cell-Free DNA Amount and Integrity
The easiest way to use circulating DNA as a biomarker is through the evaluation of its total amount, since a high level of cfDNA in blood reflect cancer growth and tumor burden [31][32][33][34][35][36][37][38]. In 2006, Iizuka et al. [31] demonstrated that cfDNA was able to identify HCC in a cohort of HCV positive patients with a sensitivity of 69.2% and a specificity of 93.3% (AUC = 0.90), both higher than those of AFP. These early results are in line with previous data from our research group: the total amount of cfDNA achieved a sensitivity of 91%, a specificity of 43%, and an AUC of 0.69 in discriminating HCC from CLD and cirrhotic patients [34]. Since cfDNA is not specific for liver cancer, several studies reported an increased diagnostic accuracy when its determination was combined with other biomarkers (i.e., AFP) [32,33,38]. cfDNA have an average size of~180 base pairs and its fragmentation is a nonrandom process, since liver cfDNA has been found to end at specific genomic coordinates [39]. Interestingly, shorter cfDNA was found in HCC patients compared to non-cancer patients, probably reflecting that not only apoptosis, but also necrosis of tumor cells contributes to the pool of circulating DNA [40,41]. Some researchers demonstrated that the evaluation of length and integrity of cfDNA achieved a diagnostic accuracy comparable to that of AFP [36,42]. The measure of cfDNA total amount or integrity may also be useful as a prognostic biomarker. In their seminal study, Tokuhisa et al. [43] demonstrated that higher levels of cfDNA after liver resection in patients with HCV-related HCC were associated with an increased risk of metastases (adjusted hazard ratio [HR] = 4.5, 95% CI 1.3-14.9) and poorer overall survival (OS) (adjusted HR = 3.4, 95% CI 1. 5-7.6). Several other subsequent studies confirmed that patients with high levels of cfDNA had a worse prognosis after different treatments (liver transplantation, liver resection and sorafenib) [34,44,45]. Moreover, a poorer OS was also demonstrated in patients with decreased cfDNA integrity (adjusted HR = 1.86, 95% CI 1.20-2.88) in the study by El-Shazly et al. [36].
When dealing with cfDNA amount as a cancer biomarker, it should be noted that the circulating DNA does not derive only from tumor cells. More precisely, the fraction of cfDNA directly attributable to the presence of cancer is named circulating tumor DNA (ctDNA) [46]. Although patients with cancer have higher cfDNA levels compared to healthy subjects, ctDNA represent a small proportion of the total amount and its level depend on disease burden, stage, cellular turnover and treatment response [47]. Moreover, high quantities of cfDNA are not cancer specific, being also elevated in inflammatory and autoimmune diseases (cirrhosis, chronic hepatitis, systemic lupus erythematous, and rheumatoid arthritis), in pregnancy, and after physical exercise [27,47]. This low specificity may scale back the role of whole cfDNA quantification as diagnostic biomarkers. Nevertheless, a remarkable study demonstrated that the cell and tissue of origin of cfDNA could be inferred by the analysis of the position of nucleosomes [48]. Snyder et al. demonstrated that since nucleosomes, the basic unit of chromatin, are placed in different positions depending on the cell type, nucleosome footprint in cfDNA could be useful to determine the relative contribution of cancer cells to the total circulating DNA pool [48].

Mutations
The majority of studies on cfDNA focused on mutational analysis and epigenetic characteristics, such as its methylation signature. HCC, when compared to other solid tumors, has a lower mutational burden [49]. The main driver somatic mutations affect telomere integrity (TERT promoter, 44%), cell cycle (TP53, 31%), and WNT signaling (CTNNB1, 27%) [50]. Less commonly AXIN1, ARID1A, ARID2, BAP1, RB1, and KEAP1 are mutated (5-10%) [50]. In addition, genetic alterations may be present, including broad chromosome gains and losses with high-level DNA amplifications of chromosomes 6p21 and 11q13, loci corresponding to VEGFA and CCND1/FGF19, respectively [49]. A relevant proportion of the mutations found in HCC biopsies are also detectable in cfDNA (43-83%) [45,51]. According to Howell et al. [52], all the mutations found in the plasma cfDNA matched with tissue mutations, while only 71% of mutations on tumor tissue were found in circulating DNA. When dealing with mutational analysis of cfDNA, we must keep in mind that mutations are more easily identified in advanced disease. In a recent study, at least one mutation in cfDNA was found in almost all (6/7) patients with a tumor ≥5 cm or with metastases, while only 9% of mutations were detected in the cfDNA of patients with smaller, not metastatic HCC [53]. Others reported that, in 48 patients, at least one type of mutation among TP53 (c.747G > T), CTNNB1 (c.121A > G, c.133T > C), or TERT (c.1-124C > T) was documented in 56.3% of patients; only 22.2% of patients had matched mutations in HCC tissue, while none of these mutations were found in non-tumoral liver tissue or in peripheral mononuclear cells [54]. In parallel to what was found in HCC tissue, TP53 is the most commonly mutated gene in cfDNA [55]. In particular, TP53 c.747G > T (p.R249S) mutation appears to be highly specific, since Cohen et al. [56] found it in approximately 20% of HCC blood samples and, conversely, in only 3-4% of pancreatic and stomach cancer samples and in none of more than 800 healthy controls. Although confirming a very high specificity (100%), another study showed a very poor sensitivity (7.6%) for the analysis of TP53 R249S mutation alone in cfDNA [57]. In order to overcome this limitation, the accuracy of TP53 mutation in association with other mutations in a diagnostic panel was evaluated [58][59][60]. Qu et al. demonstrated that a score including several cfDNA mutations (TP53, TERT, CTNNB1 and AXIN1, and HBV integrations), in combination with protein biomarkers (AFP and DCP), age and gender efficiently identified early-stage HCC in a high-risk HBsAg-seropositive population [60]. Sensitivity and specificity, 85% and 93% in the training cohort, were even better in the validation cohort (100% and 94%, respectively) [60]. Moreover, the positivity of TP53 R249S mutation in cfDNA proved to be useful also as prognostic biomarker in a large study involving 895 HCC patients, being a predictor of poorer OS and shorter progression-free survival (PFS) in patients with or without liver resection [61].
The human telomerase reverse transcriptase (TERT) gene encodes for the catalytic subunit of telomerase, which acts together with multiple molecules to maintain telomere homeostasis and chromosomal integrity [62]. The mutations found in TERT promoter lead to TERT reactivation and cell immortalization. Male patients with HCV and/or alcoholic related cirrhosis have a higher prevalence of TERT promoter mutations both in tumor tissue and in cfDNA [63], providing the rationale for TERT promoter mutations analysis in cfDNA for early detection in some populations at risk of developing HCC. In addition, presence of TERT promoter mutation in cfDNA has been associated with poor prognosis after different treatments [58,[63][64][65].

Methylation/Epigenetics
Changes in DNA methylation, particularly in the CpG islands of tumor suppressor genes, have been demonstrated to be pivotal in HCC development [66]. Analysis of the methylation pattern of cfDNA may have a value as diagnostic and prognostic biomarker, and might reveal information about tumor size, risk of metastatic spread, and recurrence [67]. Alterations in DNA methylation patterns in HCC tumor tissue after liver resection have been described for many genes. In particular, hypermethylation was found in p15, CDKN2A (encoding for p16), glutathione S-transferase (GSTP1), Ras association domain family 1A (RASSF1A), APC, SOCS1, SOCS3, TIMP3, blood vessel epicardial substance (BVES), and Homeobox A9 (HOXA9) genes, while hypomethylation in long interspersed element-1 (LINE-1) repetitive sequence [67][68][69][70][71][72][73]. However, only a proportion of cfDNA carried the same methylation patterns: hypermethylation of GSTP1 and RASSF1A was found in 50% and in 70-93% of cases respectively, while hypomethylation of LINE-1 in approximately 67% of cases [71,72]. Nevertheless, a large number of studies investigated the diagnostic accuracy of the methylation patterns in several different genes, demonstrating a diagnostic accuracy comparable or even superior to that of AFP. A very high diagnostic accuracy could be obtained with methylation scores, which combine methylation patterns in different genes. Wen et al. [74] demonstrated that a methylation score derived from the analysis of more than 10 genes achieves a sensitivity of 94% and a specificity of 89%. Lu et al. [75] obtained an AUC of 0.87 analyzing the methylation of APC, COX2, RASSF1A, and miR-203, compared to an AUC of 0.56 for AFP. In another study, the methylation of RASSF1A, BVES, and HOXA9 achieved a 73.5% sensitivity and a 91.1%, specificity, with an AUC of 0.83 [70]. A very high diagnostic accuracy in distinguish HCC patients from cirrhotics (sensitivity/specificity 95%/86%, AUC = 0.93) was reported by Kiesel et al. for a score composed by the analysis of HOXA1, EMX1, ECE1, AK055957, PFKP, and CLEC11A methylation in a discovery, phase I pilot and phase II clinical validation cohort study [76]. Cai et al. developed and validated a non-invasive diagnostic model based on Genome-wide mapping of 5-hydroxymethylcytosines in cfDNA achieving an AUC of 0.85 in distinguish early HCC from chronic liver disease (CLD), thus outperforming AFP (AUC = 0.69) [77]. The methylation analysis of cfDNA demonstrated to be useful also in predicting prognosis. RASSF1A methylation was positively correlated with tumor size, while LINE-1 hypomethylation was associated with HCC progression and patients' survival. The combination of these two genes methylation status was able to predict tumor recurrence after liver resection [71]. The role of LINE-1 hypomethylation in predicting poor prognosis was also confirmed by other researchers [78,79].
In a very interesting recent study including 1098 HCC patients and 835 controls, the authors constructed a diagnostic model with 10 methylation markers in cfDNA, achieving a sensitivity of 85.7% and a specificity of 94.3% in the training cohort (560 normal samples and 715 HCC) [80]. In the validation cohort (275 normal samples and 383 HCC) the model demonstrated a sensitivity of 83.3% and a specificity of 90.5%, thereby differentiating HCC patients from normal controls with an AUC of 0.966 [80]. In the same study, the prognostic score, which was based on the evaluation of the methylation profile of 8 different genes, was associated with higher mortality both in the training (HR = 2.41, 95% CI 1.90-3.03) and in the validation cohort (HR = 1.55, 95% CI 1.25-1.92) [80].
In the chapter of epigenetic biomarkers, nucleosomes and extracellular histones are also emerging. Nucleosomes, beyond being fundamental for genome compaction in the nucleus, may regulate genes expression through their composition and post-translational modifications [81]. Their circulating levels are increased in stroke, trauma, and sepsis [82].
In addition, circulating nucleosome demonstrated a remarkable diagnostic and prognostic performance in several human malignancies, including pancreatic [83], lung [84], colorectal [85], and breast cancers [86]. Moreover, circulating histones have been demonstrated to be key mediators of lethal sepsis [87] and liver inflammatory injury [88]. Some studies demonstrated an involvement of macro histone variants (in particular macroH2A1) in modulating HCC progression and stem cell differentiation [89,90]. There is still poor evidence about circulating nucleosomes and cell-free histones/histone complexes as liquid biopsy biomarkers in HCC. Nevertheless, some interesting results have been achieved in obesity and metabolic fatty liver disease (MAFLD), both risk factors for HCC development. A strong correlation between fatty liver index (a predictor of MAFLD based on BMI, waist circumference, triglycerides, and GGT) and high levels of circulating nucleosomes have been found in obese patients with MAFLD [91]. Moreover, a circulating histone signature (depletion of histone variants macroH2A1.1 and macroH2A1.2, individually or in complex with H2B) identified the severity of steatosis in subjects with lean MAFLD [92]. These encouraging results, together with the simple methodology of the determination (ELISA), could pave the way to the evaluation of circulating nucleosomes and cell-free histones/histone complexes as diagnostic and prognostic biomarkers in HCC.
Overall, a large body of evidence has been produced supporting the great potential of cfDNA as diagnostic and prognostic biomarker in HCC. However, it should be considered that current data largely derive from proof-of-concept retrospective studies, lacking adequate controls (not always including patients at risk of developing HCC, i.e., cirrhotics) and including only a minority of cases with early-stage HCC, which would be candidates for curative treatment options. Moreover, an additional concern regards the lack of standardized protocols for pre-analytical sample preparation, purification, and analysis. Although the use of cfDNA as a liquid biopsy currently presents several limitations in the early detection of HCC, due to the very low amount of cfDNA in the early stages, these approaches may probably dramatically change HCC surveillance. Indeed, a study published more than 10 years ago demonstrated that aberrant methylation of cfDNA fragments was detected up to 9 years before the diagnosis achieved with standard methodology [93]. Methylation score: RGS10, ST8SIA6, RUNX2, VIM, CACNA1C, TBX2, SOX9
Among short non-coding RNAs, which are generally~28 base pairs long, microRNAs (miRNAs) are the most extensively studied biomarkers in HCC in recent years, with a role in the diagnosis and in prognosis prediction. miRNAs generally bind to 3'UTR of the target mRNA, resulting in down-regulation of gene expression through translational repression and/or mRNA degradation. More than 70 miRNAs have already been proposed as candidate biomarkers [25]. Table S1 summarizes the most relevant studies on miRNAs as HCC biomarkers.
In the diagnostic setting, highly expressed miRNAs (miR-21, miR-199 and miR-122) seem to be the most promising for the diagnosis of HCC when considered individually, due to their elevated specificity [131]. For instance, Tomimaru et al. [132] demonstrated that miR-21 yielded an AUC of 0.773 with 61.1% sensitivity and 83.3% specificity in differentiating HCC from chronic hepatitis, and an AUC of 0.953 with 87.3% sensitivity and 92.0% specificity in differentiating HCC from healthy volunteers (in both cases superior to AFP). However, in the long run the diagnostic power of a single miRNA turned out to be limited and various panels consisting of more than one circulating miRNA have been evaluated. Lin et al. [133] demonstrated that a seven miRNAs classifier (miR-29a, miR-29c, miR-133a, miR-143, miR-145, miR-192, and miR-505) had a greater AUC compared to AFP in identifying small size and early-stage HCC, detecting also AFP-negative tumors. Another panel consisting of miR-122, miR-192, miR-21, miR-223, miR-26a, miR-27a, and miR-801 was able to distinguish with high accuracy between HCC and healthy controls (AUC = 0.941), chronic hepatitis B (AUC = 0.842), and liver cirrhosis (AUC = 0.884) [134]. Interestingly enough, in a recent study the determination of eight miRNAs showed a sensitivity of 97.7% and a specificity of 94.7% in identifying the presence of HCC among patients at risk, with almost all cancers (98%) diagnosed in early stages [135].

Extracellular vesicles (EVs) are small membrane vesicles released by cells in extracellular environment in normal physiology and in pathological conditions
EVs are generally classified in small (exosomes) and large EVs (ectosomes, also called microparticles (MPs) or microvesicles) [160]. Although small and large EVs may be distinguished by some of the expressed markers, such as CD63, HSP70, CD9, CD81, and integrins [161,162], the border between these two entities is not sharp [25]. The growing number of studies providing evidence for a key pathophysiological role of EVs in various aspects of liver diseases and the fact that EVs are released in the systemic circulation, where they are remarkably stable, provide the background to consider their assessment and quantification in blood as a novel form of liquid liver biopsy [66]. Several studies demonstrated a potential role of EVs as biomarkers in HCC patients ( Table 2).
First reports showed that HCC patients had a higher level of circulating EVs compared to controls [163] and the determination of total amount of EVs provided slightly better sensitivity and specificity compared to alpha-fetoprotein (AFP) in HCC diagnosis [164]. A specific form of large EVs expressing surface AnnexinV, EpCAM, ASGPR1, and CD133 was identified by Julich-Haertel et al. [165] as a marker able to distinguish HCC and cholangiocarcinoma from other cancer types, cancer-free cirrhotic patients, and healthy subjects. Sensitivity, positive predictive value, and area under the curve (AUC) in the distinction between HCC and cirrhosis were 80%, 73%, and 0.744, respectively [165].
A lower number of studies investigating EVs in the prognostic field are available, and most of them focused on the evaluation of exosomal miRNAs, in particular after surgical therapies (liver resection or liver transplantation) [177][178][179][180][181][182][183]. The only miRNA included in more than one study was miR-21, and its high levels have been repetitively associated with increased risk of disease progression and poorer survival [182,184,185]. Other studies demonstrated that low levels of exosomal miR-718, miR-125b, miR-638 and miR-320a [177][178][179]181] and high exosomal miR-665 and miR-10b [180,182] were associated with worse prognosis.
EVs and their content are promising candidate biomarkers in patients with HCC for diagnosis and prognosis prediction. Nevertheless, additional larger prospective studies should be conducted to definitely establish their role as liquid biopsy.

Circulating Tumor Cells
Metastatization is a complex and largely unknown process requiring the ability for cancer cells to escape from the primary tumor, survive in the circulation, and then settle in a new organ. Circulating tumor cells (CTCs) are key players in cancer dissemination. Considering that CTCs are present in the order of one per billion of blood cells in patients with metastatic disease, there have been some initial obstacles in their study [186]. Nevertheless, technical and methodological advances in the last years led to a significant expansion of publications aimed at investigating their role as candidate biomarkers (Table 3).
Platforms for the detection of CTCs are based on their known biological and physical properties, and can grossly be divided in immunoaffinity-based and biophysical propertybased enrichment [187]. Immunoaffinity-based CTCs enrichment techniques use antibodies against cell surface markers bounded to the device or a magnetic substance. The enrichment can be positive when CTCs are captured with antibodies against tumor specific antigens expressed on CTC surface, or negative when hematopoietic cells in the background are depleted by using antibodies against CD45 [188]. The CellSearch™ system (Veridex) captures CTCs using an immunomagnetic separation system with antibodies against EpCAM and cytokeratin coated onto ferrofluid beads and has been approved by the US Food and Drug Administration for use in patients with breast, prostate and colorectal cancers [189,190]. Other developed detection systems include CTC-Chip™ [191], CTC-iChip™ [192], and NanoVelcro™ [193]. These methods rely on tumor expression of the target proteins and their role is limited for cancers that do not typically express them. Only about one third of CTCs in HCC are positive for EpCAM and cytokeratin [194,195], and even if CellSearch™ became the most popular detection system, it could be of limited application in HCC. Moreover, given that epithelial markers such as EpCAM are often downregulated or lost during epithelial-to-mesenchymal transition (EMT) [196], CTC with EMT phenotype which have strong metastatic potential could not be detected by positive enrichment methods that target epithelial markers. Therefore, strategies targeting stem cell markers (CD133), mesenchymal markers (vimentin), and cancer specific antigens (such as HER2, PSMA, ASGPR, Hepar 1, and carbamoyl phosphate synthetase 1) have been developed [197,198]. The biophysical methods to isolate CTCs rely on their typical features such as large size, mechanical plasticity, and dielectric mobility properties, employing centrifugation and filters or flow devices with channels of varying size or nature. Although the advantage of avoiding the challenges of targeting numerous tumor specific epitopes, these methods may be less cancer-specific.
As far as the diagnostic value of CTCs analysis is considered, published studies showed that CTCs may have a role in differentiating HCC from controls. A major concern when dealing with CTCs analysis as diagnostic biomarker is the fact that, since their levels correlate with tumor burden [199], the sensitivity in early-stage disease may be too low. Nevertheless, Guo et al. [200] in a large study investigating a CTC-derived PCR score (quantifying the expression of cancer-related genes in blood), demonstrated a sensitivity of 72.5%, a specificity of 95%, and an AUC of 0.88 (compared with 57%, 90%, and 0.77 of AFP at a cut-off of 20 ng/mL). In addition, this score performed well also in patients with early-stage HCC (AUC of 0.92 in BCLC stage 0 and 0.86 in BCLC stage A).
CTCs evaluation combined with AFP provided incremental performance with respect to AFP alone in identifying HCC patients. In a study, the AUC in the discrimination of CLD and HCC patients was 0.67 for AFP (cut-off 400 ng/mL), 0.77 for CTCs (detected with CanPatrol™), and 0.82 for the combination of both [201]. Guo et al. reported that CTCs, defined by the expression of EpCAM mRNA, had a sensitivity of 42.6% and an AUC of 0.70 in discriminating HCC from CLD and healthy controls, while AFP (cut-off 400 ng/mL) demonstrated a lower sensitivity (39.5%; AUC not reported); the combination of CTCs and AFP increased sensitivity to 73% and the AUC to 0.86 [202].
Considering that CTCs are extremely rare in the circulation and that their number tends to be proportional to tumor volume, which make their detection in early-stage disease challenging [199], they are probably more useful for prognostication rather than for early diagnosis. Indeed, several evidences emerged linking CTCs enumeration with prognosis of HCC patients. A landmark study in 2004 demonstrated that the presence and number of CTCs, identified and enumerated based on their morphology, were associated with shorter survival [203]. Subsequent studies using CellSearch™ showed that the detection of EpCAM positive CTCs was associated with an higher tumor recurrence rate after liver resection [204] and with a worse prognosis [205,206]. The independent prognostic value of CTCs amount was confirmed even with other CTCs enrichment technologies, such as ImageStream flow cytometry, which uses a panel of markers and generates high resolution images of isolated CTCs [195,207]. Beyond simple enumeration, other reports have investigated the prognostic impact of subgroups of CTCs, divided according to cell surface markers, RNA expression, or genomic aberrations. The identification of CTCs with cancer stem cell-like or mesenchymal surface markers is useful to predict tumor recurrence [208][209][210][211].
Other studies demonstrated that CTCs with detectable AFP mRNA were associated with a higher risk of metastatic dissemination [212], whereas CTCs with aneuploid chromosome 8 predicted shorter survival in patients treated with surgical resection [213]. The interesting study by Ha et al. [214] introduced the concept of ∆CTC, which is defined as the variation in CTCs enumeration after surgery, and is an independent factor of lower survival and recurrence after hepatectomy.
Cancer cell dissemination seems to be promoted by treatment, in particular surgical therapies. Liver manipulation is associated with a release of CTCs [215] and the anterior as compared to the conventional surgical approach is associated with a lower release of CTCs as well as better survival [216]. In liver transplantation for HCC, five steps to minimize CTCs dissemination and thereby reduce the risk of recurrence have been described [217]. This approach in transplantation assumes even more importance as an association between CTCs detection and recurrence after transplantation has been demonstrated [218,219]. Overall, data consistently reported that the number of CTCs is a surrogate of poor prognosis, predicting higher recurrence and lower survival. A recent metanalysis and data from experimental models led to the same conclusions [220,221].
Considered that CTCs detection methods are costly and time consuming, the application of CTCs enumeration in clinical practice requires a clear advantage to be established. Probably, this is an unrealistic goal and therefore phenotypic characterization of CTCs may be more useful, since tissue-based biomarkers that could be of help in treatment choice and monitoring are currently lacking. Moreover, it is clear from several studies that CTCs are a heterogeneous population and that they may reflect tumoral heterogeneity better than a tissue biopsy [187,195]. The CTC pERK/pAkt phenotype has recently been reported to predict sensitivity to sorafenib [222], while the presence of CTCs positive for PD-L1 is associated with response to checkpoint inhibitors [223]. Considered that result, it could be imagined that phenotyped CTCs will be useful surrogates for guiding enrichment trials with molecular targeted therapies. Moreover, methods for collecting living CTCs from HCC patients and culture them into three-dimensions spheroid-like structures have also been developed, with the possibility to bring personalized medicine to a new level. In this scenario, Zhang et al. [224] explored individual sensitivity to sorafenib and oxaliplatin after collecting and culturing CTCs, and the evaluation of multiple therapeutic candidates in patients' CTC-derived xenografts may become a future reality [66].
Even if the use of CTCs analyses as biomarkers in guiding clinical decisions has huge potential, perhaps the most innovative and relevant contribution of CTC studies will be in advancing our understanding of the biology of metastatic disease as well as the development of treatment strategies. The analysis of CTCs at a molecular level, facilitated by the advancements in sequencing technologies, may lead to the identification of new mutations responsible for tumor metastatization and resistance to drugs [225]. Moreover, other insights in metastatic spread have been achieved analyzing the spatial distribution of CTCs in the bloodstream. An interesting study analyzed and compared CTCs collected in HCC patients from different vessels (peripheral veins and arteries, portal vein, and hepatic veins). The greatest number of CTCs was demonstrated in hepatic veins, with a dramatic reduction in peripheral vessels after passage through the lungs. Moreover, there was a phenotypic heterogeneity in CTCs isolated from different sites, being predominantly epithelial into the hepatic vein and EMT-transformed when isolated in peripheral vessels [226]. The CTC burden and the presence of CTC clusters in both hepatic and peripheral veins predicted lung and liver metastases.
Although the rapid evolution in technologies supporting CTCs detection, isolation and characterization and the very promising results in the studies so far published, the clinical application of CTCs as biomarkers is hindered by the different methodologies applied by single researchers. Indeed, few studies have been reproduced by more than one research group. Before the incorporation of CTCs evaluation in trials and clinical practice, standardized protocols with reproducible results, currently lacking in HCC, are needed.  [194] CellSearch™ 20 HCC 10 CLD AFP (400 ng/mL) CTC detection in 7 of 20 (35%) HCC patients and 0 of 9 CLD (p = 0.04). AFP ≥ 400 ng/mL: sensitivity 70% AFP < 400 ng/mL: sensitivity 10% (p = 0.008)     HCV, hepatitis C virus; HR, hazard ratio; IAT, intra-arterial therapies; LR, liver resection; LT, liver transplantation; OS, overall survival; OR, odds ratio; NS, not significant; NR, not reported; PFS, progression-free survival; PVT, portal vein thrombosis; RFS, recurrence-free survival; RT, radiotherapy; SIRT, selective internal radiation therapy; TACE, transarterial chemoembolization; TTP, time to progression; TTR, time to recurrence.

Conclusions and Future Perspectives
The identification of reliable non-invasive biomarkers that could allow a personalized management of HCC patients has become a key priority in the last years. Circulating markers that can integrate or eventually replace percutaneous liver biopsy, overcoming its limitations, are crucial. In addition, HCC detection at early stages, when it is susceptible to potentially curative treatments, and prediction of response to therapy are critical to improve patient survival. Although fewer data are available for HCC compared to other malignancies, numerous recent publications demonstrated very interesting and promising results regarding liquid biopsy role in diagnosis, prognosis, and prediction of response to treatment. cfDNA, cfRNA, EVs, and CTCs emerged as attractive liquid biopsy candidates because they fulfil many of the major characteristics of an ideal biomarker. To date, the closest approach to reaching the introduction in clinical practice, after the necessary large and prospective studies, is cfDNA methylation profiling for early detection of HCC in patients at risk. Mutational profiling of cfDNA and CTCs analyses are dependent on tumor burden and therefore likely more useful in intermediate or advanced settings as prognostic and predictive tools. Even tough fewer data are currently available, the analysis of EVs could provide biomarkers at every HCC stage and has the advantage to provide functional information (e.g., interactions between cancer cells and tumor microenvironment or distant cells).
Although the large amount of encouraging data collected in recent years predict a bright future for liquid biopsy in HCC, its widespread clinical application is yet not on the horizon. The majority of data supporting its utility derives from proof-of-concept studies, mainly retrospective, and not validated by different researchers. The main limitation that hinders the routine application of liquid biopsy is the lack of standardization, absence of accepted standard operating procedures, and the lack of comparability between existing approaches [47]. The standardization of pre-analytical, analytical, and post-analytical variables should be addressed. Considering for instance cfDNA analysis, the avoidance of white blood cells (WBC) lysis during blood collection and processing is important to prevent dilution of tumor circulating fragments with non-tumoral DNA (pre-analytical phase). Moreover, transportation, processing and storage temperature are also critical, impacting on WBC stability and cfDNA degradation. Since cfDNA has a short half-life and there are time-dependent changes of DNA in blood collection tubes (because of the degradation from DNase activity), plasma should be isolated within an hour from collection (analytical phase). Considered the relevance of these and other variables on the final results, the standardization of methodological protocols is an essential step to take in order to integrate liquid biopsy in the everyday clinical practice.
With the aim of identifying clinically useful diagnostic biomarkers, studies should include as controls only patients at risk of developing HCC (i.e., cirrhotics or high risk chronic hepatitis patients), who represent the ideal target for surveillance [5]. This is not trivial, also considering that it could make more difficult the identification of specific diagnostic biomarkers. In fact, chronic hepatitis and cirrhosis are pre-cancerous conditions in which some of the molecular modifications found in overt HCC are already in place. For instance, during the progression of liver damage the pattern of DNA methylation changes over time in multiple hepatic cell types, and the release of methylated cfDNA from dying hepatocytes has been demonstrated to be a useful approach to evaluate fibrosis grade [247,248]. In order to have a chance of being introduced in clinical practice, liquid biopsy biomarkers should be specific enough to distinguish early-stage HCC from simple cirrhosis, a condition in which the molecular pathways leading to cancer may be already at least in part activated. In addition, when tumor burden is low, highly sensitive tests are necessary to overcome the limitation posed by the small amount of circulating cancer byproducts. Even though these new liquid biopsy strategies represent very promising tools, another not negligible consideration should be done about their costs. While currently used biomarkers (AFP) are measured with unexpensive and simple methods, EVs isolation and analysis, cfDNA mutational profiling and epigenetic analysis, and CTCs enrichment methods require devoted personnel and are all costly and time consuming. Nevertheless, such limitations will likely be overcome by advances in technology that will make these determinations easier and accessible to most laboratories.
Once these new generation reliable biomarkers will be developed and validated, the final step will be to determine the optimal way to integrate them in the clinical management of patients with HCC. The replacement of currently used tools in the management of HCC patients by liquid biopsy biomarkers is unrealistic, but they will likely be integrated in the process, providing a stronger predictive power. An interesting approach in surveillance, which remains to be evaluated in ad hoc studies, could be the combined evaluation of liquid biopsy biomarkers with the currently used periodic liver ultrasonography. Given the possibility of minimally invasive repeated sampling, liquid biopsies can enable realtime monitoring of disease during therapy and could supplement imaging information to provide a more careful assessment of the tumor. Hopefully, in the future, the analysis of circulating HCC by-products will also allow personalized molecular targeted therapy. In order to achieve these important goals, not only prospective observational trials should be conducted, to correlate liquid biopsy biomarkers with clinical outcome, but also interventional studies, in which cfDNA, EVs, and CTCs analysis will prompt therapeutic decisions, are necessary.
Funding: This research received no external funding.