Targeted Inhibition of Purine Metabolism Is Effective in Suppressing Hepatocellular Carcinoma Progression

Abstract

Abbreviations ADSL adenylosuccinate lyase ADSS adenylosuccinate synthetase AICAR 5‐aminoimidazole‐4‐carboxamide‐1‐β‐d‐ribofuranoside AKT protein kinase B AMP adenosine monophosphate ATIC 5‐aminoimidazole‐4‐carboxamide ribonucleotide formyltransferase c cyclic CREB cyclic adenosine monophosphate responsive element binding protein Ctrl control DEN diethylnitrosamine DMSO dimethyl sulfoxide E2F1 E2F transcription factor 1 GART phosphoribosylglycinamide formyltransferase, phosphoribosylglycinamide synthetase, phosphoribosylaminoimidazole synthetase GMP guanosine monophosphate GMPS guanine monophosphate synthase HCC hepatocellular carcinoma HSP90 heat shock protein 90 IC50 median inhibitory concentration IMP inosine monophosphate IMPDH inosine‐5′‐monophosphate dehydrogenase LIHC liver hepatocellular carcinoma MAPK mitogen‐activated protein kinase MMF mycophenolate mofetil MPA mycophenolic acid mRNA messenger RNA mTOR mammalian target of rapamycin MYC myelocytomatosis NSG NOD scid gamma PAICS phosphoribosylaminoimidazole carboxylase and phosphoribosylaminoimidazolesuccinocarboxamide synthase PDX patient‐derived xenograft PFAS phosphoribosylformylglycinamidine synthase PI3K phosphoinositide 3‐kinase PPAT phosphoribosyl pyrophosphate amidotransferase PRPS1 phosphoribosyl pyrophosphate synthetase 1 qPCR quantitative polymerase chain reaction RB RB transcriptional corepressor SGH Singapore General Hospital sh short hairpin TCGA The Cancer Genome Atlas TF transcription factor XMP xanthosine monophosphate Hepatocellular carcinoma (HCC) is the most prevalent form of primary liver cancer and accounts for the second highest number of cancer mortalities worldwide. Large‐scale HCC cohort studies, such as The Cancer Genome Atlas (TCGA), highlight extensive interpatient tumor heterogeneity in terms of gene expression levels and oncogenic mutations, indicating a patient‐specific tumor profile and illustrating the complexities of a one‐size‐fits‐all approach. TCGA‐liver hepatocellular carcinoma (LIHC) gene expression data and clinical parameters were downloaded from the University of California Santa Cruz Cancer Genomics Browser (https://xenabrowser.net/) and cBioPortal for Cancer Genomics (https://www.cbioportal.org/). ( 23,24) To identify metabolic pathways essential for HCC tumorigenesis, we assessed the gene expression pattern of key enzymes involved in cellular metabolism in patients from the TCGA‐LIHC data set.

Abbreviations
• ADSL
• adenylosuccinate lyase
• ADSS
• adenylosuccinate synthetase
• AICAR
• 5‐aminoimidazole‐4‐carboxamide‐1‐β‐d‐ribofuranoside
• AKT
• protein kinase B
• AMP
• adenosine monophosphate
• ATIC
• 5‐aminoimidazole‐4‐carboxamide ribonucleotide formyltransferase
• c
• cyclic
• CREB
• cyclic adenosine monophosphate responsive element binding protein
• Ctrl
• control
• DEN
• diethylnitrosamine
• DMSO
• dimethyl sulfoxide
• E2F1
• E2F transcription factor 1
• GART
• phosphoribosylglycinamide formyltransferase, phosphoribosylglycinamide synthetase, phosphoribosylaminoimidazole synthetase
• GMP
• guanosine monophosphate
• GMPS
• guanine monophosphate synthase
• HCC
• hepatocellular carcinoma
• HSP90
• heat shock protein 90
• IC50
• median inhibitory concentration
• IMP
• inosine monophosphate
• IMPDH
• inosine‐5′‐monophosphate dehydrogenase
• LIHC
• liver hepatocellular carcinoma
• MAPK
• mitogen‐activated protein kinase
• MMF
• mycophenolate mofetil
• MPA
• mycophenolic acid
• mRNA
• messenger RNA
• mTOR
• mammalian target of rapamycin
• MYC
• myelocytomatosis
• NSG
• NOD scid gamma
• PAICS
• phosphoribosylaminoimidazole carboxylase and phosphoribosylaminoimidazolesuccinocarboxamide synthase
• PDX
• patient‐derived xenograft
• PFAS
• phosphoribosylformylglycinamidine synthase
• PI3K
• phosphoinositide 3‐kinase
• PPAT
• phosphoribosyl pyrophosphate amidotransferase
• PRPS1
• phosphoribosyl pyrophosphate synthetase 1
• qPCR
• quantitative polymerase chain reaction
• RB
• RB transcriptional corepressor
• SGH
• Singapore General Hospital
• sh
• short hairpin
• TCGA
• The Cancer Genome Atlas
• TF
• transcription factor
• XMP
• xanthosine monophosphate

Hepatocellular carcinoma (HCC) is the most prevalent form of primary liver cancer and accounts for the second highest number of cancer mortalities worldwide.⁽¹⁾ Despite the global burden, surgical resection and liver transplantation remain the only curative treatments,⁽²⁾ and existing therapies for advanced unresected HCC offer modest survival benefits.^(3,4) Several inhibitors that have demonstrated clinical benefits in other cancers, such as sunitinib and erlotinib, as well as yttrium‐90 radioembolization fail to show positive effects in patients,⁽⁵⁾ highlighting the dearth of available treatments and a need to identify new modalities for clinical intervention in HCC. Cancer metabolism offers an attractive target to selectively antagonize tumors given that tumor cells exhibit altered metabolism to gain a survival advantage over nontransformed cells.⁽⁶⁾ Metabolic reprogramming of tumors to consume glucose and divert these metabolic intermediates into lactate production, a phenomenon known as aerobic glycolysis or the Warburg effect, was one of the first reported distinctions between cancer and normal tissues.⁽⁷⁾ Rewiring of amino acid, lipid, and nucleotide metabolism has also been shown to be essential for sustaining tumorigenesis in various cancers,⁽⁶⁾ suggesting a benefit of exploiting metabolic liabilities for cancer therapy. Deregulation of tumor metabolism is often achieved by alterations in the expression of metabolic enzymes either by gene amplification or deletion or as a consequence of changes in the activation status of upstream growth signaling pathways.⁽⁸⁾ The purine biosynthesis pathway has been demonstrated to be under the transcriptional control of myelocytomatosis (MYC) signaling in cancers, such as glioma, melanoma, lymphoma, and prostate cancer.^(9‐12) Insofar as purine metabolism has been implicated in maintaining tumor‐initiating cells and the pathogenesis of various malignancies through enhancing nucleic acid synthesis to sustain rapid proliferation,^(9,13) a role for purine biosynthesis in HCC has remained inconclusive.

Inosine‐5′‐monophosphate dehydrogenase (IMPDH) is a critical enzyme that catalyzes the conversion of inosine monophosphate (IMP) to xanthosine monophosphate (XMP), an irreversible and rate‐limiting process during purine biosynthesis that establishes the intracellular pool of guanine nucleotides.⁽¹⁴⁾ IMPDH activity is enhanced in tumor cells, with IMPDH a potential chemotherapy target.^(15,16) Two IMPDH inhibitors, mycophenolic mofetil (CellCept) and mycophenolate sodium (Myfortic), are used as immunosuppressants in organ transplants and autoimmune diseases.^(17,18) Clinical trials to assess the efficacy of CellCept in cancers have been confined to hematologic malignancies, and investigations into the therapeutic benefit of inhibiting purine biosynthesis in solid cancers are lacking.

Large‐scale HCC cohort studies, such as The Cancer Genome Atlas (TCGA), highlight extensive interpatient tumor heterogeneity in terms of gene expression levels and oncogenic mutations, indicating a patient‐specific tumor profile and illustrating the complexities of a one‐size‐fits‐all approach.⁽¹⁹⁾ Although common mutations in telomerase reverse transcriptase (TERT), tumor protein p53 (TP53), and catenin beta 1 (CTNNB1) exist in most patients, these molecular alterations are not clinically druggable. The promise of precision medicine in oncology has resulted in tailor‐made treatments based on tumor molecular profile instead of tumor origin.⁽²⁰⁾ Despite advances across tumor types, the development of molecularly targeted therapeutics is missing in HCC due to the absence of validated biomarkers to predict treatment outcome.⁽²¹⁾ Here, we provide evidence that inhibition of purine biosynthesis by targeting IMPDH in HCCs distinguished by high purine biosynthetic activity reduced the tumor burden. In combination with an antagonist of phosphoinositide 3‐kinase (PI3K), an upstream regulator of purine biosynthetic enzyme expression, to potently suppress purine metabolism, we observed an enhanced reduction in tumorigenesis, offering a novel combination for precision therapy of HCC.

Materials and Methods

Human Subjects and Data

Protocols involving human patients were approved by the SingHealth Institutional Review Board at the National Cancer Center Singapore (NCCS). Tumor samples were surgically removed at the NCCS with informed consent from the patients. TCGA‐liver hepatocellular carcinoma (LIHC) gene expression data and clinical parameters were downloaded from the University of California Santa Cruz Cancer Genomics Browser (https://xenabrowser.net/) and cBioPortal for Cancer Genomics (https://www.cbioportal.org/). Data from the Roessler Liver 2 data set were downloaded from the Oncomine platform (https://www.oncomine.org/resource/login.html) with default thresholds set for P values and gene ranks. Gene expression data of HCC cell lines were downloaded from the Cancer Cell Line Encyclopedia (CCLE) portal (https://portals.broadinstitute.org/ccle).

Animals and Tumor Xenograft Model

All animal studies were approved by the institutional animal care and use committees at A*STAR and the National University of Singapore. We purchased 6‐8‐week‐old male C57BL/6, female CrTac:NCr‐Foxn1^{athymic nude (nu)}, and female NOD.Cg‐Prkdc^{severe combined immunodeficiency (scid)}Il2rg^tm1Wjl/SzJ (NOD scid gamma [NSG]) mice from InVivos, Singapore. All mice were maintained under pathogen‐free conditions on a 12‐hour light/dark cycle with free access to water and standard chow diet (1324; Altromin). To chemically induce HCC, 5 mg/mL diethylnitrosamine (DEN) (N0258; Sigma) was injected intraperitoneally into 2‐week‐old C57BL/6 mice at a dose of 20 mg/kg body weight, followed by eight consecutive administrations of 1,4‐Bis[2‐(3,5‐dichloropyridyloxy)]benzene (TCPOBOP) (T1443; Sigma) at a dose of 3 mg/kg body weight every 14 days starting at 4 weeks of age. At 7 months, tumor size and multiplicity were evaluated by counting the number of visible tumor nodules and determining the largest diameter of each nodule with a caliper. For the tumor xenograft model, 10⁶ cells in 100 µL Dulbecco’s modified Eagle’s medium (DMEM) were mixed with an equal volume of Matrigel (354230; Corning) and injected subcutaneously into the hind flank of 6‐8‐week‐old nude/NSG mice. Tumor growth was allowed to proceed for a month. Mice with no tumors were not used for analysis. The tumor burden was determined by measuring tumor weight and volume (V), where V = (tumor length × tumor width [W] × W)/2. For patient‐derived xenograft (PDX) experiments, freshly resected human HCC tissues were transplanted subcutaneously into NSG mice. Mice were euthanized when tumors reached 1.5 cm in dimension. PDX tumors were subsequently harvested and processed into PDX cells, as described.⁽²²⁾ PDX tumors were maintained as a continuous source of tumor cells using NSG mice. For drug treatment of tumors, mycophenolate mofetil (MMF) was resuspended in a vehicle comprising 0.9% sodium chloride, 0.5% methyl cellulose, and 0.4% TWEEN 80, while the PI3K inhibitor PI‐103 was dissolved in dimethyl sulfoxide (DMSO). One week after tumor cell injection, mice were administered a regimen of 120 mg/kg MMF by oral gavage twice a day and 10 mg/kg PI‐103 by intraperitoneal injection daily for 3 weeks.

Results

De Novo Purine Biosynthesis Is Up‐Regulated in HCC

Previous studies demonstrated that dysregulation of hepatic metabolism is crucial for the development and progression of liver disease.^(23,24) To identify metabolic pathways essential for HCC tumorigenesis, we assessed the gene expression pattern of key enzymes involved in cellular metabolism in patients from the TCGA‐LIHC data set. We found that enzymes in the de novo purine biosynthesis pathway showed a significant increase across 360 HCC tumor samples compared to nontransformed liver (Fig. 1A). In contrast, we did not observe a consistent differential up‐regulation of enzymes in four other core metabolic pathways: glycolysis, glutathione, one carbon metabolism, and pyrimidine (Supporting Fig. S1A). Increased expression of enzymes in the pentose phosphate pathway (PPP) in tumors, however, suggested enhanced levels of ribose 5‐phosphate (R5P), a core product and intermediate of the PPP that is also a critical precursor for purine biosynthesis. Purine metabolism is initiated by the sequential activities of several enzymes, such as phosphoribosyl pyrophosphate amidotransferase (PPAT), phosphoribosylformylglycinamidine synthase (PFAS), and 5‐aminoimidazole‐4‐carboxamide ribonucleotide formyltransferase (ATIC)/IMP cyclohydrolase, functioning in the main branch leading to stepwise formation of IMP from R5P. IMP can then be converted into adenosine monophosphate (AMP) or guanosine monophosphate (GMP) through two distinct reactions catalyzed by adenylosuccinate lyase (ADSL), and IMPDH and guanine monophosphate synthase (GMPS), respectively (Supporting Fig. S1B). Higher grade tumors (G2 and G3), pathologically defined as poorly differentiated and highly metastatic, exhibit progressively higher levels of PFAS, ATIC, IMPDH1, IMPDH2, GMPS, and ADSL compared to low‐grade tumors (G1) or normal liver tissue (Supporting Fig. S1C). We also determined that stage T2 and T3 tumors, clinically manifested as large tumors with high multiplicity, were characterized by up‐regulated expression of these enzymes (Supporting Fig. S1D). Consistent with a critical role for enhanced purine biosynthesis in the pathogenesis of aggressive HCC, high expression of these enzymes correlated with poor patient survival outcome (Fig. 1B). Further interrogation of transcriptomic data from tumor and matched normal tissue of a cohort of 48 patients with HCC from the Singapore General Hospital (SGH) revealed a corresponding enrichment of purine biosynthetic enzymes in tumors (Fig. 1C). Analysis of the Roessler human HCC data set comprising 220 normal and 225 tumor samples on the Oncomine platform^(25,26) also confirmed that expression of purine biosynthetic genes was significantly up‐regulated in tumors (Fig. 1D). Surprisingly, we did not find any correlation between overall patient survival and tumor levels of IMPDH1 or IMPDH2 (Supporting Fig. S2) in 11 other gastrointestinal cancers in the TCGA, suggesting a specific dependence on purine biosynthesis for HCC growth and sustenance.

IMPDH Activity Is Necessary for HCC Proliferation

Elevated expression of purine biosynthetic enzymes in HCC suggests that HCC may be sensitive to inhibition of purine metabolism. To this end, we targeted the first committed and rate‐limiting step in guanine nucleotide biosynthesis by using MMF, a U.S. Food and Drug Administration‐approved prodrug of mycophenolic acid (MPA) that inhibits IMPDH activity, in a chemically induced mouse model of HCC. DEN‐induced tumors expressed elevated levels of both IMPDH isoforms compared to adjacent nontransformed liver (Supporting Fig. S3A). Tumors were visibly smaller after a 3‐week regimen of MMF (Fig. 2A). Despite a decrease in tumor size (Fig. 2B), we did not observe an effect of MMF on the frequency of tumor formation (Fig. 2C). We next turned to an HCC cell line model to elucidate the mechanisms regulating purine biosynthesis. To select an optimal HCC line that demonstrates maximal response to MPA, we mined the CCLE database to identify cell lines with highest expression levels of purine biosynthetic enzymes.⁽²⁷⁾ Clustering analysis resulted in the segregation of 25 HCC cell lines into two distinct groups with contrasting expression profiles (Supporting Fig. S3B). Because MPA directly targets both IMPDH isoforms, we validated by quantitative polymerase chain reaction (qPCR) that Huh7 cells expressed highest levels of both IMPDH1 and IMPDH2 (Supporting Fig. S3C,D). Treatment of Huh7 cells with MPA completely attenuated the proliferation of these cells (Fig. 2D), a phenomenon reversed by exogenous addition of guanine or guanosine (Fig. 2E), consistent with studies demonstrating that the only nucleotides that had intracellular levels reduced by MPA treatment were guanylates.⁽²⁸⁾

We generated stable knockdowns of both IMPDH isoforms in Huh7 cells by using three independent short hairpin (sh)RNA sequences (Supporting Fig. S3E‐G). Reduction of either IMPDH1 or IMPDH2 decreased cellular proliferative capacity (Fig. 2F), a phenomenon reversed by exogenous supplementation of guanine or guanosine (Supporting Fig. S3H). Treatment of IMPDH knockdown cells by MPA further attenuated their proliferation (Supporting Fig. S3I). These cells also displayed impaired tumorigenic activity when xenografted into nude mice (Fig. 2G,H). Similarly, IMPDH1 or IMPDH2 knockdown in SNU398 cells impacted their proliferation (Supporting Fig. S3J‐M). Treatment of SNU398 cells with MPA also abolished their proliferative capacity, a phenomenon that was reversed by guanine or guanosine (Supporting Fig. S3N). In support of purine synthetic activity being critical for the growth and proliferation of HCC cells with elevated expression of IMPDH, HepG2 cells, which express comparatively low levels of both IMPDH1 and IMPDH2, were resistant to MPA (Supporting Fig. S3O). In addition to its catalytic role in the conversion of IMP to XMP, IMPDH is also capable of binding nucleic acids and regulating transcription, a function exclusively mediated by its Bateman domain, a moiety dispensable for the enzymatic activity of IMPDH.^(29,30) To explore a contribution by this noncatalytic feature toward the tumorigenic potential of IMPDH, we constructed IMPDH1 mutants lacking the Bateman domain (Supporting Fig. S4A) and overexpressed them in IMPDH1‐knockdown Huh7 cells (Supporting Fig. S4B). Restoration of cellular proliferation was observed in cells transduced with Bateman‐deficient IMPDH1 (Fig. 2I). The lack of a functional role for the Bateman domain in tumorigenesis was confirmed using two IMPDH1 isoforms: canonical IMPDH1 expressed ubiquitously (isoform E) and a longer isoform containing an additional exon at the amino terminus (isoform B). Both isoforms exhibited a cytoplasmic localization, although isoform B was also found concentrated in rod/donut‐like structures devoid of actin or microtubules (Supporting Fig. S4C). Collectively, these results demonstrate that the participation of IMPDH in the catalytic biosynthesis of purines is necessary for HCC growth.

Intertumoral Heterogeneity in Purine Synthetic Pathway Activity Determines Tumor Response to MPA

HCCs exhibit extensive heterogeneity with regards to proteomic, metabolomics, and immune profiles as well as responses to anticancer drugs, contributing to different metabolic requirements across tumors.^(31,32) We hypothesized that every tumor would thus have a unique growth dependency on IMPDH‐driven purine biosynthesis and demonstrate varying sensitivities and tolerance toward MPA. PDX mouse models offer a clinically relevant system to test this hypothesis, with drug studies being able to be performed in vivo as well as in vitro on PDX organoids in the context of heterogeneous molecular alterations in patients. HCC cells from 13 patients were implanted into immunodeficient mice, and the characteristics of these distinct PDX lines were interrogated with respect to the status of purine metabolic activity (Fig. 3A). Immunoblotting performed on tumors validated the heterogeneous expression of purine biosynthetic enzymes (Fig. 3B,C). Analysis of transcript levels yielded a tight correlation between messenger RNA (mRNA) and protein expression (Fig. 3D). We next measured the levels of metabolites in the purine metabolic pathway as a functional readout of pathway activity. In agreement with protein and mRNA expression, the levels of metabolites in both the AMP and GMP branches showed comprehensive variation among PDX tumors, suggesting differential purine requirements for sustenance of tumor proliferation (Fig. 3E). A pairwise correlation analysis revealed that the purine metabolic signature correlated positively with genes involved in cellular proliferation (Fig. 3F), consistent with an oncogenic role for purine metabolism. Treatment with MPA of PDX tumors cultured as organoids dramatically inhibited the proliferation of some tumors (PDX #1 and #11) while exhibiting a negligible effect on others (PDX #12 and #19) (Fig. 3G). MPA treatment also resulted in elevation of the upstream metabolite 5‐aminoimidazole‐4‐carboxamide‐1‐β‐D‐ribofuranoside (AICAR) in MPA‐sensitive tumors (Fig. 3H), suggesting that AICAR levels could be a useful biomarker to monitor the effectiveness of MPA therapy. Intriguingly, PDX tumors that were MPA sensitive were characterized by two distinct features that implied tumor‐specific vulnerabilities to IMPDH inhibition: elevated expression of either IMPDH1 or IMPDH2, and high levels of guanosine nucleosides.

PI3K Signaling Regulates Purine Synthetic Enzyme Expression

Metabolic reprogramming of tumor cells is regulated by oncogenic pathways, activation of which enables tumor adaptation in an altered nutritional microenvironment to sustain rapid proliferation and evade apoptosis.⁽³³⁾ In this context, we tested a panel of drugs targeting oncogenic pathways previously implicated in cancer metabolic reprogramming on Huh7 shIMPDH1 and shIMPDH2 cells to identify signaling pathways that drive purine metabolism. The rationale for this was that down‐regulated expression of IMPDH1 or IMPDH2 would lead to decreased purine synthetic activity, rendering increased susceptibilities to a drug that inhibits a protein directly regulating the expression levels of purine synthetic enzymes. Given that purine metabolic activity is necessary for proliferation of Huh7 cells, a drug‐induced reduction in enzyme levels would result in enhanced sensitivities of shIMPDH1 and shIMPDH2 cells in the same manner that these cells exhibited increased vulnerability to MPA, which has a higher affinity for IMPDH2 than IMPDH1 (3‐fold and 13‐fold decrease in median inhibitory concentration [IC₅₀] in shIMPDH1 and shIMPDH2 cells, respectively) (Fig. 4A). Indeed, only the PI3K inhibitor PI‐103 conferred added vulnerability of Huh7 shIMPDH1 and shIMPDH2 cells (1.5‐fold and 1.8‐fold decrease in IC₅₀, respectively) (Fig. 4B). In contrast, IMPDH down‐regulation did not alter Huh7 sensitivity to drugs targeting proteins previously shown to regulate HCC tumorigenesis (Supporting Fig. S5A‐H). Importantly, Huh7 shIMPDH1 and shIMPDH2 cells were also resistant to mammalian target of rapamycin (mTOR) inhibition by rapamycin (Supporting Fig. S5I), suggesting that a PI3K‐induced mTOR‐independent signaling mechanism regulates purine synthesis in Huh7 cells.

We next examined the effect of PI‐103 and protein kinase B (AKT) inhibitor MK‐2206 on the expression of purine synthetic enzymes. Protein levels of ATIC, phosphoribosyl pyrophosphate synthetase 1 (PRPS1; functioning in the main branch leading to IMP generation), IMPDH1, IMPDH2 (involved in converting IMP to GMP), and ADSL (involved in converting IMP to AMP) were decreased in a dose‐dependent fashion in Huh7 cells (Fig. 4C). This decrease was also observed at the mRNA level for all 11 purine synthetic enzymes (Supporting Fig. S6A). We observed a similar dependency in PDX #11 organoids (Fig. 4C), a patient‐derived line most vulnerable to IMPDH inhibition (Fig. 3G). To mimic nutrient deprivation and insulin resistance in the tumor environment, we grew serum‐starved Huh7 cells with 1 g/L glucose, which is one fifth the glucose concentration in standard cell culture conditions, in the presence of insulin. Under these conditions, elevated expression of purine synthetic enzymes was decreased to near baseline levels in cells treated with PI‐103 (Fig. 4D). Importantly, overexpression of phosphatase and tensin homolog (PTEN), a negative regulator of PI3K signaling, in Huh7 cells also resulted in attenuated expression of purine synthetic enzymes (Supporting Fig. S6B). The tight association between PI3K signaling and purine synthesis prompted us to interrogate their relationship in the TCGA‐LIHC cohort. Mutations in genes encoding PI3K pathway components, which alter the activation status or expression levels of the enzymes, are a frequent occurrence in cancers.⁽³⁴⁾ A pairwise correlation analysis revealed a strong correlation between the purine metabolic signature and PI3K pathway components (Fig. 4E). In addition, deleterious mutations in PI3K pathway components were enriched in tumors overexpressing purine synthetic enzymes (Supporting Fig. S6C), linking aberrant PI3K pathway activation to elevated expression of these enzymes. Decreased expression of the tumor suppressor PTEN and up‐regulation of the oncogenic phosphoinositide‐3‐kinase regulatory subunit 6 (PIK3R6) also correlate with poor prognosis in HCC (Supporting Fig. S6D). Transcriptomic analysis of the SGH‐HCC cohort revealed a similar correlation between purine biosynthetic and PI3K pathway genes (Supporting Fig. S6E). We next performed drug combination studies to determine if inhibitors of these pathways synergize in inhibiting Huh7 viability. Using the Chou‐Talalay method,⁽³⁵⁾ the combination index values for the combination of PI‐103 and MPA were 0.21 and 0.76 in Huh7 cells (Fig. 4F) and PDX #11 organoids (Fig. 4G), respectively, indicating strong synergy. Taken together, these results establish the nonredundant requirement of purine synthesis on PI3K signaling in maintaining HCC proliferation and tumorigenesis.

E2F Transcription Factor 1 Is a Downstream Effector of PI3K Signaling in the Regulation of Purine Synthesis

The dispensability of mTOR for purine synthesis‐dependent survival of Huh7 cells implied an mTOR‐independent branch of PI3K signaling functioning to manipulate the expression of purine synthetic enzymes. To identify the PI3K effector activating gene transcription, we analyzed the gene promoters of all purine synthetic enzymes to identify top candidate transcription factor (TF) occupancy. We focused our analysis to only include sites that had published evidence of TF binding so as to select for TFs that demonstrate highest functional binding potential. mRNA down‐regulation of all purine synthetic enzymes in response to PI3K inhibition (Supporting Fig. S6A) suggested a common TF, prompting us to compare across all 11 gene promoters, yielding cyclic adenosine monophosphate response element‐binding protein (CREB), E2F transcription factor 1 (E2F1), and MYC as candidate TFs (Fig. 5A). MYC inhibition had no effect on the mRNA levels of genes encoding purine synthetic enzymes (Supporting Fig. S7A), whereas CREB inhibitor 666‐15 resulted in a modest increase in transcript levels (Supporting Fig. S7B), suggesting that MYC and CREB binding to these loci occurred in a manner distinct and independent of PI3K regulation of purine synthesis pathway gene expression. In contrast, expression levels of these genes were down‐regulated in a dose‐dependent fashion in response to E2F1 inhibition by HLM006474 (Fig. 5B). E2F1 knockdown using two independent shRNAs also resulted in decreased expression of purine biosynthetic enzymes (Supporting Fig. S7C). PI‐103 and HLM006474 treatment led to a decrease of E2F1 occupancy on the promoters of these genes (Fig. 5C,D). Combinatorial targeting of E2F1 and IMPDH using HLM006474 and MPA yielded synergism in both Huh7 cells and PDX #11 organoids (Fig. 5E,F) with combination index values of 0.60 and 0.67, respectively. A combination of HLM006474 and PI‐103 was also synergistic in inhibiting the proliferation of Huh7 and PDX #11 cells (Fig. 5G,H), with combination index values of 0.51 and 0.60, respectively. E2F1 activity is regulated by interaction with the RB transcriptional corepressor (RB), whereby hyperphosphorylated RB dissociates from E2F1, allowing for E2F1‐dependent transcription of target genes.^(36,37) Importantly, PI‐103 treatment attenuated RB phosphorylation at four distinct sites in Huh7 and PDX #11 cells (Fig. 5I), supporting a role for PI3K signaling as an upstream driver in hyperphosphorylating RB and subsequent E2F1‐mediated activation of genes in the purine synthesis pathway.

Down‐Regulation of IMPDH Activity Leads to Preferential Inhibition of Mitogen‐Activated Protein Kinase/RAS Signaling

To understand how purine biosynthesis promotes HCC tumorigenesis, we performed RNA‐sequencing on Huh7 shIMPDH1 and shIMPDH2 cells. Principal component analysis comparing the transcriptomes indicated marked differences in gene expression profiles despite similar proliferative impairment (Fig. 6A). Pairwise comparison of short hairpin control (shCtrl) cells versus shIMPDH1 (Supporting Fig. S8A) or shIMPDH2 (Supporting Fig. S8B) cells showed that depletion of either IMPDH isoform was sufficient to induce large‐scale alterations in gene expression, despite both isoforms having similar substrate kinetics for the conversion of IMP to XMP. Because IMPDH1 or IMPDH2 knockdown impacted Huh7 proliferation, we reasoned that signaling pathways specifically affected by perturbations in purine synthesis that contribute to tumor proliferation should be down‐regulated in both Huh7 shIMPDH1 and shIMPDH2 cells. Gene ontology analysis revealed that mitogen‐activated protein kinase (MAPK)/RAS signaling was the most significantly down‐regulated signaling pathway (Fig. 6B). Reassuringly, genes associated with cell‐cycle progression and regulation of cell proliferation were also significantly down‐regulated. mRNA transcripts of MAPK/RAS pathway components were reduced by more than 2‐fold in response to IMPDH1 or IMPDH2 knockdown (Fig. 6C,D). We next interrogated the clinical relevance of the association of MAPK/RAS signal activation with IMPDH expression levels in the TCGA‐LIHC and SGH‐HCC cohorts by performing a pairwise correlation between IMPDH1 and genes of MAPK/RAS pathway components. We found that tumors with high IMPDH1 levels were marked by elevated expression of MAPK/RAS pathway genes in both the TCGA‐LIHC (Fig. 6E) and SGH‐HCC data sets (Supporting Fig. S8C). Treating Huh7 cells with MPA and measuring downstream readouts of MAPK pathway activation revealed that IMPDH inhibition significantly decreased phosphorylation of MAPK substrates on both threonine and serine residues in consensus MAPK phosphorylation motifs as well as phosphorylation of p44 MAPK (ERK1) (Fig. 6F). MPA also resulted in decreased MAPK target gene expression (Fig. 6G). The minimal effect on wingless‐related integration site (WNT) and MYC target genes supports MAPK signaling as a selective signaling pathway dependent on active purine biosynthesis. Such gradation of pathway dependencies rules out a global deregulation of oncogenic pathways in response to antagonism of purine metabolism.

Combinatorial Targeting of Purine Biosynthesis Inhibits In Vivo HCC Tumorigenesis

Our data suggest a critical role for purine biosynthesis in the proliferation of HCC cells, mediated in part by a PI3K–E2F1 axis that precisely regulates purine biosynthetic enzyme expression. Given that PI3K inhibition only partially blocks the expression of these enzymes but combining PI‐103 with MPA synergistically inhibited HCC cell viability, we hypothesized that efficient blockage of purine biosynthetic enzyme output through the cooperative action of MPA and PI‐103 is critical for abrogating HCC tumor growth. To this end, we investigated the in vivo efficacy of combinatorial impediment of purine biosynthesis in a PDX mouse model. Mice transplanted with PDX #11 tumor cells were subjected to one of four arms of treatment for 3 weeks: vehicle control, MMF monotherapy, PI‐103 monotherapy, or a combination of MMF (120 mg/kg twice a day) and PI‐103 (10 mg/kg daily) (Fig. 7A). Compared to vehicle control, the tumor burden was decreased in animals receiving either monotherapy regimen. Importantly, we observed strongest reduction in tumor growth in the cohort that received the drug combination (Fig. 7B,C). Histologic analysis revealed a loss of normal cord architecture in vehicle‐treated tumors that was partially negated in the group administered the combinatorial therapy (Fig. 7D), demonstrating the efficacy of sustained and targeted inhibition of purine biosynthesis in HCC.

Discussion

Elevated energy consumption of tumor cells to support rapid proliferation necessitates the integration of various metabolic processes to sustain high levels of DNA replication and RNA production.⁽⁶⁾ In this study, we found that HCCs up‐regulate enzymes involved in purine metabolism. Such coordinated increased expression of all enzymes in the pathway, as opposed to regulation of purine synthesis through modulation of a single enzyme, is significant in the context of cancer metabolism because alterations in metabolic flux are often achieved by the actions of an upstream master transcriptional regulator to effect a synchronized and efficient mode of control over the entire synthetic pathway.⁽³⁸⁾ MYC has been shown to transcriptionally regulate purine biosynthesis in various cancers.^(9‐12) However, MYC does not perform a similar role in HCC. Instead, we report a PI3K–E2F1 axis, suppression of which antagonized the activation of the entire suite of purine synthetic genes to repress HCC tumorigenesis. Identification of PI3K signaling as a master regulator of purine metabolism is significant because aberrant activation of this network occurs frequently in cancers to decouple deregulated tumor cell growth from the external nutritional milieu.⁽³⁹⁾ PI3K–mTOR signaling is an essential driver of cancer metabolic reprogramming, in part through integration of glycolytic intermediates into purine synthesis.⁽⁴⁰⁾ It was therefore surprising that mTOR inhibition did not significantly alter HCC viability in the setting of purine synthesis inhibition, consistent with a previous report showing a mode of PI3K regulation of purine synthesis independent of mTOR/S6 kinase in mouse embryonic fibroblasts and C2C12 myoblasts.⁽⁴¹⁾ In the context of HCC, a recent study comparing the responses of 81 HCC cell lines to a panel of anticancer drugs demonstrated that, despite rapamycin and PI‐103 eliciting similar sensitivities in many lines, a significant fraction exhibited opposing vulnerabilities to PI3K or mTOR inhibition,⁽³¹⁾ attesting to a subset of HCC wherein mTOR is dispensable for PI3K regulation. Here, we demonstrate that PI3K signaling modulates E2F1 occupancy on purine synthetic gene promoters through RB phosphorylation, thus relieving sequestration of E2F1 to RB, consistent with a role for E2F1 in regulating the cell cycle and metabolic adaptations of cancer cells.⁽⁴²⁾ The effect of PI3K signaling on RB phosphorylation is likely mediated by the cyclin D–cyclin‐dependent kinase 4/6 complex⁽⁴³⁾ because we find no evidence of AKT phosphorylation sites on RB. Elevated expression levels and copy number gains in E2F1 in high‐grade HCCs^(44,45) are consistent with our analysis of the TCGA‐LIHC data set where we found up‐regulation of purine synthetic enzymes in high‐grade HCC, further highlighting a functional link between E2F1 and purine synthesis in promoting HCC malignancy. Taken together, we establish a central role for E2F1 at the interface of oncogenic signal transduction and tumor metabolism. In light of our findings that PI3K inhibition is sufficient to repress the purine synthetic gene expression stimulated by excess insulin, it is tempting to speculate a link between insulin resistance and HCC pathogenesis that is mediated by PI3K‐regulated purine synthesis.

Preferential exploitation of purine synthesis to propagate HCC tumorigenesis suggests that purine metabolism offers a point of vulnerability to therapeutically target HCC. Conventional therapies have not yielded dramatic improvements in survival cohort studies.^(3,4) This is probably due to extensive tumor heterogeneity as well as a complex interplay among oncogenic signaling pathways and metabolic reprogramming. Indeed, our findings that attenuation of purine synthesis resulted in preferential down‐regulation of MAPK/RAS signaling rather than a global deregulation of cell signaling hints at a convergence of the effects of purine metabolism on nucleic acid synthesis and the generation of secondary messenger cyclic (c)AMP and cGMP. In this context, treatment with cAMP target protein kinase A agonist forskolin affected MAPK pathway activity and subsequent mitogenesis in HCC cell lines,⁽⁴⁶⁾ and frequent activation and up‐regulation of MAPK pathway components are detected in HCCs.^(47,48) Taken together, attenuated MAPK signal transduction in response to the inhibition of purine synthesis likely plays a significant role toward decreased HCC tumorigenesis, although the contributions of other signaling pathways should not be discounted.

Given the limited success of current therapies, it is noteworthy that we observe trifurcation of PDX tumor response toward MPA treatment, suggesting that targeting of purine synthesis is most effective in a subset of patients. This distinction in tumor response can be accounted for by the graded expression of purine synthetic enzymes and metabolites in the PDX tumors, indicative of patient‐specific requirements for purine synthesis in driving HCC tumorigenesis. A comparison of proteomic and metabolomic expression in MPA‐resistant and susceptible PDX tumors suggests that tumors that respond to IMPDH inhibition are defined by two features: elevated expression of IMPDH1 or IMPDH2, and high guanosine levels. Conversely, tumors resistant to MPA express either low guanosine levels despite high IMPDH1 and IMPDH2 expression (PDX #12) or exhibit elevated guanosine levels despite no detectable IMPDH1 or IMPDH2 (PDX #19). This discrepancy might be accounted for by the enhanced contribution of the purine salvage pathway, allowing for the IMPDH‐independent generation of guanosine nucleosides from byproducts of DNA and RNA degradation. Collectively, this suggests that enhanced levels of both IMPDH1/IMPDH2 and guanosine could serve as potential biomarkers to inform the design of personalized therapy for patients with HCC to select those most likely to benefit from MMF treatment. It is also noteworthy that we observed increased AICAR levels in MPA‐sensitive tumors following MPA treatment, suggesting a potential utility of AICAR measurement as a proxy to measure the effectiveness of IMPDH inhibition.

An improvement over the use of a single therapeutic agent is the combinatorial approach of simultaneously targeting a key rate‐limiting step in purine synthesis driven by IMPDH and an upstream oncogenic regulator to enhance the antitumorigenic efficacy. In this regard, a recent study demonstrated the benefit of blocking pyrimidine synthesis using teriflunomide in conjunction with PI3K inhibitor BKM120 to increase the inhibition of glioblastoma tumorigenesis in a xenograft model.⁽⁴⁹⁾ For our study, we observed a synergistic effect of combining MPA with PI‐103 or HLM006474 in attenuating the proliferation of Huh7 cells and a PDX organoid culture (PDX #11). PDX #11 exhibited highest inherent purine synthetic activity among all PDX lines and therefore had the most severely impacted viability by MPA treatment. We subsequently showed that the tumor burden in a PDX #11 mouse model was reduced to a greater extent by cooperatively targeting purine biosynthesis using MMF and PI‐103 compared to either monotherapy. Our results, as summarized in Fig. 7E, demonstrate a preferential requirement for purine synthesis mediated by the activities of PI3K and E2F1 in the proliferation and survival of a subset of patients with HCC. This offers a tantalizing approach to delivering precision therapy to patients by targeted inhibition of purine metabolism at two distinct points of susceptibility.

Acknowledgment

We thank A. Tan, W.K. Tham, and S. Adav for technical and experimental support. We also acknowledge the Singapore Bioimaging Consortium‐Nikon Imaging Center for help with microscopy and image analysis.