Liver cancer, mainly hepatocellular carcinoma (HCC) and intrahepatic cholangiocarcinoma (iCCA), is the leading cause of cancer-related deaths worldwide, with an estimated incidence of over 1 million cases by 2025 (Llovet et al., 2018; Villanueva, 2019; Llovet et al., 2021). Infection with hepatitis B and C viruses, smoking, iron overload, and alcohol-related cirrhosis are well-known risk factors for liver cancer. Recently, metabolic dysregulation in liver cancers has gained more attention, and the associated factors, such as obesity, type 2 diabetes, and non-alcoholic fatty liver disease, have been extensively investigated (Piccinin et al., 2019; Satriano et al., 2019; Faubert et al., 2020).

Owing to the high heterogeneity in liver cancers, systemic therapies, including immune checkpoint inhibitors (ICIs) and tyrosine kinase inhibitors (TKIs), have failed to achieve satisfactory efficacy, especially for patients with advanced stages of the disease (Llovet et al., 2008; Bruix et al., 2017; Kudo et al., 2018; Chen et al., 2019; Bruix et al., 2021; Qin et al., 2021). Accordingly, novel treatment targets must be identified, and patients with liver cancer must be precisely stratified to enable accurate treatment.

Loss-of-function experiments are usually performed during high-throughput screening (HTS), including RNA interference (RNAi) screening (for knockdown) and CRISPR-Cas9 screening (for knockout) for functional genomics and drug discovery (Moffat et al., 2006; Dietzl et al., 2007; Joung et al., 2017). Owing to its high efficiency and specificity, CRISPR-Cas9 screening, which mediates double-strand breaks in the target DNA using guide RNA (gRNA) libraries, is more widely used than RNAi screening (Morgens et al., 2016). There are two main sources of CRISPR screening: the dependency map (DepMap) (Pacini et al., 2021) portal, which is the largest and latest database integrating three large-scale projects, and the BioGRID ORCS (Oughtred et al., 2021), which serves as a warehouse for the published CRISPR screening results. To identify the potential dependency genes that could be used as therapeutic targets, we explored liver cancer datasets deposited in the DepMap database.

Various metabolic alterations occur in liver cancers, such as the upregulation of aerobic glycolysis and nucleotide synthesis, providing energy and biomacromolecules for tumor development and progression (Piccinin et al., 2019; Satriano et al., 2019; Huang et al., 2020). Based on previous studies, metabolic rearrangements, such as the upregulation of nucleotide metabolism and downregulation of lipid metabolism, representing poor prognosis, could also serve as prognostic markers for HCC (Bidkhori et al., 2018; Pascale et al., 2019). Furthermore, the liver, which is the major site for carbohydrate and lipid biosynthesis and amino acid metabolism, is important for maintaining metabolic homeostasis (Gebhardt, 1992; Kietzmann, 2017; Piccinin et al., 2019). As a result, evaluating metabolic rearrangement is essential for understanding liver cancer onset and progression.

Here, we identified fitness genes that could be used as therapeutic targets using data from CRISPR-Cas9 screens and constructed a prognostic model. After evaluating the pathways and biological processes of these genes, we found that they were mainly associated with metabolic rearrangements. Thus, robust targets for liver cancer were identified. Overall, a comprehensive picture of potential fitness genes that are critical for the survival or proliferation of liver cancer cell lines is presented herein. These tumor vulnerabilities could facilitate the development of potential therapeutic targets and ultimately improve patient outcomes.

Given the global burden of liver cancers and the modest outcomes of current therapeutics, there is a critical need for precise stratified strategies, especially novel comprehensive therapeutic options for patients with liver cancers. In this study, we used the ADaM to identify context-specific fitness genes from DepMap and constructed a 25-gene prognostic model that divided patients into high- and low-risk groups. GSVA and GO analyses revealed significant metabolic rearrangements associated with this signature. Furthermore, a drug sensitivity analysis was performed using data from LIMORE. Genes included in this signature are mainly involved in the life cycle of tumor cells rather than normal cells, suggesting their potential as ideal therapeutic targets with minimal side effects. To date, no studies have systemically explored the role of fitness genes in the treatment of liver cancers. Furthermore, our prognostic model had better predictive power than previous models.

Mutations or dysregulation of transcription factors, essential kinases, and signaling receptors represents a unique class of drug targets that mediate aberrant gene expression, including blocking differentiation and cell death gene expression programs as well as hallmark properties of cancer (Bushweller, 2019). Owing to the cascade amplification effect, minimal perturbations of these upstream molecules can induce significant downstream changes and are at the heart of the overall tumor signaling network (Sever and Brugge, 2015). The differential expression of these core proteins cannot be directly captured using high-throughput analysis. In fact, in conventional transcriptome differential analysis, a minimum threshold of 2-fold expression difference is usually used, and most of the core proteins cannot reach such significant differences. Numerous studies have been conducted on large-sample transcriptome cohorts, such as TCGA, to screen for hub genes (Wang et al., 2021). Owing to the reasons mentioned previously, many upstream core proteins were excluded due to their lower expression and insignificant differential expression. Therefore, a better strategy is needed to identify core proteins that play essential roles in tumors.

The emergence of CRISPR screening technology is an ideal solution to these problems. By performing a large-scale genomic-level lethal gene screen within hundreds of cell lines, the DepMap project could identify core proteins dependent on the growth of tumor cell lines (Pacini et al., 2021). Furthermore, the ADaM was used to select context-specific fitness genes with drug target potential. These genes are proteins that have been validated at a practical level and play a central role in tumor survival. By limiting our targets to these genes, we can effectively avoid the effect of significant signals caused by cascade amplification effects. Combined with the TCGA/ICGC database and DepMap data, targets with both prognostic and therapeutic significance can be effectively screened.

Two problems cannot be avoided if the TCGA database alone is used. First, there is no guarantee that every gene has a significant biological function and may be screened for passenger genes: participants with significant expression changes, but not tumor prognostic differences. Second, the TCGA database may include screening for genes that indirectly affect tumor prognosis, such as PDCD1. Although such genes can also be used as therapeutic targets, this study focused on targeting HCC cells themselves as a killing mechanism.

Of note, only targets with a combination of efficacy and low toxicity are ideal drug targets. Genes that are essential for tumor cell survival but not for the survival of normal tissues should be ideal therapeutic targets with high effectiveness and minimal side effects. Thus, it is particularly important to distinguish context-specific fitness genes from core fitness genes. To explore the genomic effect in improving cancer patient clinical outcomes and accelerate the development of new cancer therapies, CRISPR technologies based on gRNAs have been used to study gene function and identify cellular fitness genes (defined as genes essential for cell growth or viability).

This 25-gene signature based on context-specific fitness genes not only serves as a prognostic model but also provides novel treatment strategies for patients with liver cancer. Most of the genes involved in this signature can be used as therapeutic targets with mild toxicity to normal tissues. Centrosomal P4.1-associated protein (CPAP, also called CENPJ), which is positively correlated with recurrence and vascular invasion in HCC and contributes to sorafenib resistance (Chen et al., 2020), could be a therapeutic target for inhibiting angiogenesis and treating metastatic HCC. Some genes are associated with metabolic processes, such as glucose transporter 1 (SLC2A1, also known as GLUT1), which encodes a key rate-limiting factor in glucose transport into cancer cells, and thioredoxin reductase 1 (TXNRD1), which is the cytosolic subunit and key enzyme of the thioredoxin system. According to previous studies, inhibition of GLUT1 could impair the growth and migratory potential of HCC cells and reduce glucose uptake and lactate secretion, whereas inhibition of TXNRD1 hinders the proliferation of HCC cells and induces apoptosis in vitro (Amann et al., 2009; Lee et al., 2019). Auranofin (AUR), a pharmacological inhibitor of TXNRD1, can effectively suppress the growth of HCC tumors in animal models and sensitize HCC cells to sorafenib (Lee et al., 2019). Two genes that were negatively correlated with risk scores, UDP-glucose pyrophosphorylase 2 (UGP2) and argininosuccinate synthase (ASS1), could serve as potential therapeutic targets. In a previous study, the downregulation of UGP2, a key enzyme in glycogen biosynthesis, was revealed to be associated with the occurrence and development of various cancer types and poor survival of HCC, while the downregulation of ASS1, a key enzyme in the conversion of nitrogen from ammonia and aspartate to urea, was confirmed to support cancerous proliferation via the mammalian target of the rapamycin (mTOR) pathway (Rabinovich et al., 2015; Li et al., 2018; Hu et al., 2020).

Although most therapeutic targets are inhibitors of oncogenes, tumor suppressor genes are more frequently mutated in cancers than oncogenes. With the development of functional genomics and pharmacology, strategies targeting tumor suppressor genes or associated pathways have received increasing attention. Thus, we included the suppressor genes in our analysis. This 25-gene signature could reveal more potential treatment targets for patients with liver cancer.

By evaluating drug sensitivity using data from LIMORE in the present study, we found a positive correlation between the risk scores and the efficacy of trametinib, an MEK inhibitor. A previous study confirmed the potent single-agent antitumor activity of MEK inhibitors in animal models (Huynh et al., 2010; Schmieder et al., 2013). Based on the heterogeneity of liver cancers, the ability of our model to identify the benefits of trametinib is important. These findings indicate that high-throughput drug screens in LIMORE could assess the variable effects of drugs and provide opportunities for pharmacogenomic analysis in liver cancers.

Owing to the high cost of molecularly targeted agents and ICIs, systemic chemotherapy is still an indispensable option, despite its well-known modest efficacy in liver cancers. To reduce side effects, this signature provides a strategy to distinguish the beneficiaries of chemotherapy. GEMOX, based on gemcitabine (GEM) (Mini et al., 2006), an anti-metabolic drug, is a common choice for patients with liver cancer. However, metabolic rearrangement not only promotes the growth and metastasis of tumor cells but also induces GEM resistance. Increasing evidence suggests that gemcitabine resistance is related to the metabolism of glucose, amino acids, and lipids. In this study, the upregulation of GLUT1 promoted glucose uptake and increased glycolysis, resulting in resistance to GEM. In this process, increased glycolytic flux converts glucose intermediates into the pentose phosphate pathway (PPP) and increases pyridine biosynthesis to elevate the intrinsic levels of deoxycytidine triphosphate (dCTP), which competes with the effective levels of GEM (Gu et al., 2021). In addition to the upregulation of GLUT1, increased glycolysis and pyridine biosynthesis metabolism have been confirmed via biological analysis. ASS1 is a rate-limiting enzyme involved in arginine synthesis. Owing to a deficiency of ASS1, cancer cells become addicted to external arginine and resistant to GEM (Prudner et al., 2019). Therefore, the high-risk group with significantly higher metabolic rearrangements would be prone to chemoresistance.

Mitogen-activated protein kinases (MAPKs) are protein serine/threonine (Ser/Thr) kinases that convert extracellular stimuli into various cellular responses, including proliferation, migration, differentiation, and metabolism (Yang and Liu, 2017). In the ERK-MAPK signaling pathway, the initiating MAP3Ks are members of the RAF family, which consists of ARARF, BRAF, and CRAF, and are activated by the interaction with active GTP-binding RAS proteins. MAP3K activation leads to the activation of MEK1 and MEK2, which then activate ERK1 and ERK2 through dual phosphorylation of tyrosine and threonine residues (Cargnello and Roux, 2011). After activation, ERK1/2 phosphorylates and triggers a variety of nuclear and non-nuclear proteins that activate proliferative programs and promote the aerobic glycolytic phenotype. Among these transcription factors, c-Myc increases the expression of GLUT1 and several enzymes in the glycolytic pathway, and induces the expression of enzymes involved in nucleotides, fatty acid synthesis, and glutaminolysis (Papa et al., 2019). c-Myc promotes glycolytic intermediates in the PPP, serine, and glycine biosynthesis pathways by inducing the expression of PKM2 (David et al., 2010). These metabolic pathways and serine–threonine kinase activity pathways were found to be upregulated in the high-risk group, indicating a critical role of the ERK–MAPK signaling pathway in HCC.

Sorafenib, initially discovered as a CRAF(BRAF) inhibitor and recently identified as a multikinase inhibitor, has been approved for the treatment of HCC (Llovet et al., 2008). However, the activation of alternative survival pathways leads to drug resistance and limits the effectiveness of sorafenib (Zhou et al., 2019). Of note, the combination of MEK and BRAF inhibitors is better tolerated than their respective monotherapies (Tolcher et al., 2015). The combination of trametinib and dabrafenib, a BRAF inhibitor, was approved for the treatment of BRAF-V600E/K-mutant metastatic melanoma in 2014. In this study, the sensitivity to trametinib gradually increased with increasing risk scores, indicating that direct targeting of MEK to block MAPK signaling is an effective option for the treatment of HCC. Previous studies reported no significant improvement in the efficacy of trametinib in combination with sorafenib in patients with unselected HCC, further suggesting the importance of patient selection (Kim et al., 2020). Altogether, our model may provide valuable information for patient selection.

In general, precision medicine could be the only approach to overcome the heterogeneity of liver cancers. This biomarker-driven identification would not only improve therapeutic decision-making but also provide wide therapeutic targets.