Identification of IGF2 as Genomic Driver and Actionable Therapeutic Target in Hepatoblastoma

Hepatoblastoma (HB) is the most common pediatric liver cancer, generally occurring in children under 3 years of age (1). Although its incidence has substantially increased over the last 30 years, HB is a rare disease, with 1.8 cases per million children/year (2, 3). Unlike hepatocellular carcinoma (HCC), the most common primary liver tumor in adults, HB develops in the absence of an underlying liver disease or viral etiology. Most HB tumors occur sporadically, and only 5% of cases are associated with genetic diseases such as Beckwith–Wiedemann syndrome or Familial adenomatous polyposis (4).

Despite adjuvant and neoadjuvant cisplatin-based chemotherapy improves clinical outcomes after resection (5), recurrence-free survival at 3 years for patients with advanced stages is only of 34% (6). Furthermore, chemotherapy regimens are associated with severe and lifelong side effects such as ototoxicity and cardiomyopathy (7), which have especially impactful consequences for patients who are treated at early ages. Considering this, new treatment strategies are urgently awaited, and the identification of novel actionable drivers in HB is a current unmet medical need.

HB is a heterogeneous disease, so a better understanding of the molecular features of HB tumors is a prerequisite for developing targeted therapies. In this line, new HB classifications are emerging and include the so-called hepatocellular neoplasms not otherwise specified (HCN-NOS; refs. 8, 9). Globally, HBs present one of the lowest somatic mutation rates of all cancer types (10, 11), being mutations or deletions in the gene encoding β-catenin (CTNNB1) the most recurrent alteration (70%–80% of HB patients; refs. 12, 13), followed by mutations in NFE2L2 (10%) or in the TERT promoter (6%; ref. 10), the latter mainly representing HCN-NOS cases (9). On the other hand, insulin-like growth factor 2 (IGF2) overexpression in HB has been proposed as a potential oncogenic driver (14, 15). In HCC, IGF signaling has been identified as an epigenetic driver (16, 17) and has shown great promise as an actionable target in preclinical HCC experimental models (18). Sensitivity of HB cells to IGF1R/IGF2 pathway inhibition has been demonstrated in vitro (15). In the clinical context, IGF2 inhibition with xentuzumab (BI 836845; see ref 19 for full sequence and structure), a monoclonal antibody against IGF1 and IGF2, is under investigation in different tumor types (NCT02191891, NCT03659136, NCT03099174, NCT02123823, and NCT02204072). In this regard, preliminary results of phase I trials have revealed that xentuzumab has promising antitumor activity and is well tolerated (20). However, its clinical efficacy and safety in children with HB have not yet been explored.

Here, we analyzed 31 HB tumors, including HB (n = 29) and HCN-NOS (n = 2), and identified IGF2 as the main targetable deregulated pathway in HB. We then showed that IGF2^high HBs are associated with poor outcome, which was the basis for testing xentuzumab alone or in combination with cisplatin—the backbone of all neoadjuvant chemotherapies in HB patients and the standard of care for standard-risk HB—in HB cell lines, in a patient-derived xenograft (PDX) tumor organoid model and in a xenograft murine model mimicking IGF2^high HBs. In vitro, the combination exerted a synergistic reduction in viability and, in vivo, it improved survival and inhibited tumor angiogenesis compared with cisplatin. Overall, our findings suggest that HB patients with high IGF2 levels could benefit from xentuzumab combined with chemotherapy, and are a first step toward the implementation of precision medicine in HB.

The study cohort included primary HB tumor and adjacent nontumor samples obtained from 31 patients (21) who received resection after standard neoadjuvant therapy. Data from RNA-sequencing, methylation array, SNP array, and human transcriptomic array (GSE132219) have been previously reported elsewhere (ref. 21; Supplementary Fig. S1). The patient's clinicopathologic data are summarized in Table 1. As a validation cohort, we used an in-house cohort of 50 HB tumors and 2 nontumor formalin-fixed paraffin-embedded (FFPE) tissue samples from patients enrolled in the SIOPEL-3 clinical trial (22). A second validation cohort consisting of published RNA-sequencing data (GSE104766) from 25 HB tumors and their respective nontumor adjacent tissue was used (23). All samples were collected in accordance with European and Spanish law. Informed written consent was obtained from each patient in accordance with European guidelines for biomedical research. The study conformed to the ethical guidelines of the 1975 Declaration of Helsinki, and it was approved by the Human Ethics Committee of the Hospital Universitari Germans Trias i Pujol. All samples were histologically reviewed by two expert liver pathologists (R.A. and C.G.).

RNA-seq data from 31 HB tumor and adjacent nontumor pairs from the study cohort (21) were used. Raw expression counts were normalized with the weighted “trimmed mean method” implemented in the edgeR package (24) to correct library size differences between samples. The gene-expression profile in HB tumors was analyzed using the NTP, preranked GSEA and ssGSEA modules from GenePattern (www.genepattern.org) and gene signatures from MSigDB or previously reported (Supplementary Table S1).

Secondly, the in-house validation cohort of 50 HB tumors and 2 adjacent tissues was profiled using the NanoString nCounter Technology with a manually curated list of 65 genes, including HB markers and genes of key signaling pathways (Supplementary Table S2). RNA from tumor FFPE unstained sections was extracted using truXTRAC FFPE RNA microTUBE purification Kits (COVARIS). RNA quality control was assessed using Tape Station RNA chips (Agilent), and RNA quality was measured using DV200 metrics. RHOT2 and PNN genes were used as housekeeping genes. NanoString output files were analyzed by nSolver Analysis Software 4.0.

For the validation cohort, RNA-sequencing data (GSE104766) from tumor and nontumor HB samples from 25 HB patients were used (23). Raw expression counts were normalized following the same procedure as previously described (21).

To identify groups of correlated genes that are enriched in tumors compared with nontumor samples, we used the weighted gene coexpression network analysis (WGCNA) R package (25). Networks enriched in tumors were selected based on FDR < 0.001 and module differential connectivity ratio < 0.1. Subsequently, to identify the main signaling pathways in each network, we applied the pathway enrichment module using the Gene Ontology database and selected the top significantly enriched term (ref. 26; all fold change (FC) > 2 and BH-adjusted P < 0.05). Hub genes (i.e., the top 5 most interconnected genes of the network) were identified based on intramodule connectivity (25). The selected genes were filtered for cancer-related genes according to the OncoKB Cancer Gene List (www.oncokb.org) and for targetable alterations using the PANDRUGS database (www.pandrugs.org). Only drugs with high interaction evidence levels (i.e., direct targets) were considered.

IGF2 overexpression was defined as FC > 4 versus mean tumor-adjacent tissue based on qRT-polymerase chain reaction (PCR) data. For RNA-seq data, a cutoff of FC > 2 versus median tumor-adjacent tissue was established based on ROC curve calculations (Supplementary Fig. S2; Supplementary Table S3).

Methylation data from 27 HB tumors and 22 adjacent nontumor pairs obtained from an Infinium MethylationEPIC 850K array (Illumina) were previously reported (21) and are accessible through GSE132219. Data were normalized and converted to B values using the ChAMP pipeline (27). The differential methylation status of the IGF2 gene promoters in HB tumors and adjacent tissue was conducted using the ChAMP R package (v 2.22.0) by analyzing the array probes located within the maternal allele of adult promoter (P1), fetal promoters (P2, P3, and P4), and ICR1 regulatory region (CTCF-binding site) according to the Ensembl Genome Browser (Ensembl.org) and previous studies (ref. 18; Supplementary Table S4). Statistical comparisons for differential methylation analysis were performed using M values (28). IGF2 hypomethylation was defined as the mean FC of the B values in the fetal promoter region compared with adjacent nontumor samples minus standard deviation as previously defined (18).

LOH in HB samples was established from SNP-array data of 30 HB tumor samples. SNP-array and LOHs data were previously reported (21) and are accessible through GSE132037. Samples with LOH in the chromosomal region 11:2,129,112–2,149,603 from the GRCh38.p12 human genome version were considered positive for LOH in the IGF2 gene.

The IGF2 exon-level expression analysis was performed using the data of 36 probes of the Human Transcriptome Array data (GSE132219; ref. 21). Specifically, expression of isoforms containing the adult-specific exons 1 (hg19, chr 11: 2,170,833–2,170,356) and 2 (11: 2,169,037–2,168,796) and fetal-specific exon 6 (11: 2,160,619–2,159,459) was assessed (18). The supervised differential splicing between tumor and nontumor samples as well as IGF2^high versus IGF2^low tumors was conducted using Transcriptome Analysis Console software (Thermo Fisher Scientific).

Mutations were called from paired tumor/normal RNA-seq data. Mapped RNA-seq reads were subjected to splitting, trimming, local indel realignment, and base-score recalibration preprocessing with the IndelRealigner and TableRecalibration tools from Genome Analysis Tool Kit (GATK; ref. 29) under the GATK Best Practices for RNA-seq paradigm (https://gatkforums.broadinstitute.org/gatk/discussion/3892/the-gatk-best-practices-forvariant-calling-on-rnaseq-in-full-detail). MuTect (30) was then used with default settings to quantify somatic mutation burden. CTNNB1 and AXIN1 alterations were validated by PCR and Sanger sequencing, as previously described (21).

DNA was extracted from macrodissected fresh frozen (or FFPE tumor tissues using the MagMAX DNA Multi-Sample kit (Thermo Fisher Scientific) or truXTRAC FFPE DNA microTUBE Kit-Column Purification (Covaris). Qubit was used to quantify DNA concentration and integrity. Then, we performed PCR using specific primers (Forward: GGTGAAGGGGCAGGACGGGTGC; Reverse: GGCTTCCCACGTGCGCAGCAGGA) and 50 ng of purified genomic DNA. The PCR reaction was performed by PCR MasterMix (Thermo Fisher Scientific) including a negative and a positive control. The PCR reaction conditions were: initial denaturalization 95°C for 2 minutes, 40 cycles of denaturalization 95°C for 30 seconds and annealing 60°C for 30 seconds and extension 72°C for 2 minutes, and a 10-minute final extension at 72°C. The PCR product was sequenced by Sanger sequencing. Sequencing reactions of both DNA strands were performed with the BigDye Terminator v3.1 Cycle Sequencing Kit (Thermo Fisher Scientific), run on a Genetic Analyzer ABI 3130 and analyzed with SeqScape v5.3.1 (both from Applied Biosystems). Briefly, 5 μL of PCR product was incubated with 2 μL of illustra ExoProStar 1-Step (GE Healthcare Life Sciences) for 45 minutes at 37°C and 15 minutes at 80°C. After purification, 2 μL of purified PCR product was mixed with 2 μL of BigDye Buffer, 1 μL of BigDye, 1.6 μL of 1 μmol/L of primer and 3.4 μL of DNase RNase-free water. The sequencing reaction conditions were: 95°C for 5 minutes, 96°C for 10 seconds, 30 cycles of 50°C for 50 seconds, and a 4-minute extension at 60°C.

HepG2 (ATCC) and Huh6 (Japanese Collection of Research Bioresources) HB cell lines were cultured in Dulbecco's Modified Eagle Medium (Thermo Fisher) supplemented with 10% heat‐inactivated fetal bovine serum. Cells were maintained at 37°C in a 5% CO₂ atmosphere. The anti-IGF1/2 monoclonal antibody xentuzumab (BI 836845; see ref. 19 for full sequence and structure) was produced and provided by Boehringer Ingelheim, and cisplatin was purchased from Selleckchem (S1166). PBS was used as a control.

To generate the 2D PDX-derived cell line, PDX tissue was minced and digested in PBS + 1 mg/mL collagenase IV for 90 minutes at 37°C. Tumor dissociate was strained though a 70-μm strainer and washed with complete RPMI (20% FBS, 1% glutamine, 1% penicillin/streptomycin). 100,000 cells were plated on collagen-coated 35-mm plates in human-2D media (complete RPMI, 40 ng/mL recombinant human EGF, 10 μmol/L Y-27632, and 5 μmol/L A83-01). Cells were cultured and maintained in human-2D media 37°C in a 5% CO₂ atmosphere.

Cell proliferation and colony formation capacity were evaluated in HepG2 (ATCC), Huh6 (Japanese Collection of Research Bioresources), and 2D PDX-derived cells treated with xentuzumab (Boehringer Ingelheim), cisplatin (Selleckchem), its combination, or vehicle. For the matrix viability assay, 4,000 cells/well were seeded in 96-well plates for 24 hours in a humidified atmosphere at 37°C and 5% CO₂. Thereafter, cells were incubated for 72 hours with increasing concentrations of cisplatin and xentuzumab. Cell viability was measured with absorbance at 560 and 590 nm from resazurin. Combination index values were calculated for each column and averaged across each matrix. To assess colony formation capacity, 600 cells/well were seeded in 6-well plates. After 24 hours, Huh6 and HepG2 cells were treated with cisplatin (0.3 μmol/L), xentuzumab (1 μmol/L), or their combination and incubated for 12 days, with drug-containing media changed three times per week. Thereafter, cells were rinsed with cold PBS, fixed with methanol, and stained with 0.5% Crystal Violet (Sigma-Aldrich). The number of colonies was manually counted. All experiments were performed in triplicate.

A patient-derived HB tumor with high IGF2 expression provided by Xentech (31) was implanted and grown subcutaneously in 6- to 8-week-old female NOD/SCID immunosuppressed mice (n = 3; Charles River Laboratories). Animals were weighed, and tumor volume was assessed three times per week using bilateral caliper measurements according to the formula: volume = length × width² × 0.5. Once tumors reached 1,000 mm³, the animals were sacrificed and tumors were collected to generate three independent organoid lines (PDX-derived organoids). Histologic and molecular characterization of the PDX has been reported elsewhere (PDX ID: HB-235; ref. 31). To generate the organoids (32), PDX tumors were minced and digested in sterile digestion media (PBS, 0.125 mg/mL collagenase from clostridium histolyticum, 0.125 mg/mL dispase II, and 0.1 mg/mL DNaseI) for 2 hours at 37°C. Tumor dissociates were strained through a 70-μm strainer and washed with basal media (Advanced DMEM/F-12, 1% glutamine, 1% penicillin/streptomycin, 10 mmol/L HEPES). Cell dissociates were cultured at 50,000 cells per 50 μL Matrigel (Corning) in 24-well plates in human tumor organoid media (basal media, 1:50 B27 no vitamin A, 1:100 N2, 1 mmol/L N-acetylcysteine, 10% Rspo1-conditioned media, 10 mmol/L nicotinamide, 10 nmol/L recombinant human [Leu¹⁵]-gastrin I, 50 ng/mL recombinant human EGF, 100 ng/mL recombinant human FGF10, 25 ng/mL recombinant human HGF, 10 μmol/L forskolin, and 5 μmol/L A83-01). Studies were performed in compliance with the “Guide for the Care and Use of Laboratory Animals” and they were approved by the Institutional Animal Care and Use Committee (IACUC-2016-0296).

Organoids were seeded and treated with serial dilutions of cisplatin (Selleckchem) or xentuzumab (Boehringer Ingelheim) for cell viability analyses and IC₅₀ calculations. After 3 days of incubation, cell viability was measured with absorbance at 560 and 590 nm from resazurin. Combination index values were calculated for each column and averaged across each matrix.

Caspase-dependent apoptosis was assessed by measuring caspase-3 and -7 activity (Caspase-Glo 3/7 Assay, Promega). Organoids were treated with or without 100 nmol/L of IGF2 (PeproTech) and increasing concentrations of xentuzumab (0, 1, 10, 100 μmol/L) and cisplatin (0, 1, 5 μmol/L). All organoid experiments were performed using the 3 independent organoid lines and in technical triplicates.

To generate the xenograft murine model of IGF2^high HB, 5×10⁶ HepG2 cells were subcutaneously injected into the right flank of immunocompromised 6- to 8-week-old athymic (nu/nu; Crl:NU-Foxn1nu) female mice (n = 55 mice; Charles River Laboratories). Animals were weighed, and tumor volume was assessed 3 times per week using bilateral caliper measurements according to the following formula: volume = length × width² × 0.5. When tumors reached 150 mm³, animals were randomly assigned to receive xentuzumab (n = 14), cisplatin (n = 13), combination therapy (xentuzumab plus cisplatin, n = 14) or drug vehicle (n = 14). Xentuzumab (100 mg/kg) was administered twice per week; and cisplatin (2 mg/kg) was administered daily. Both drugs were administered intraperitoneally. Body weight, tumor growth, and survival (defined as time to reach 1,500 mm³) were monitored in all the mice. Once animals reached the survival endpoint, they were sacrificed following Institutional Animal Care and Use Committee guidelines. Studies were performed in compliance with the “Guide for the Care and Use of Laboratory Animals,” and they were approved by the Institutional Animal Care and Use Committee (IACUC-2016-0296).

Tumor samples from the mouse model were fixed in buffered 4% paraformaldehyde for 24 hours and embedded in paraffin to create FFPE blocks. An expert pathologist blinded to the treatment arms (CM) assessed the percentage of viable and necroapoptotic tissue in H&E-stained slides. This was used together with the ex vivo tumor volume to calculate the viable and necroapoptotic tumor volume. CD31 staining (Abcam ab28364) was assessed by IHC on 3-μm-thick FFPE tissue sections after heat-induced antigen retrieval and quantified by the digital pathology imaging software QuPath (version 0.2.0). To ensure representative sampling of the entire tumor, 5 regions of interest were analyzed.

Huh6 and HepG2 cells and organoids were stimulated with 100 nmol/L of IGF2 (PeproTech) to induce IGF1R pathway activation, along with cisplatin (0.3 μmol/L for cell lines and 1 μmol/L for organoids), xentuzumab (0.1 μmol/L), its combination or vehicle. For Huh6 and HepG2 cell lines, after 24 hours of stimulation, cells were washed with PBS, collected by scraping and lysed in lysis buffer (50 mmol/L Tris pH = 7.4, 150 mmol/L NaCl, 1% Triton X-100, 0.1% SDS, 0.25 mmol/L EDTA, 1% sodium deoxycholate) containing protease and phosphatase inhibitors. For organoids, after 24 hours of stimulation, organoids were collected in Cell Recovery Solution (Corning), and matrigel was dissolved by rotation for 1 hour at 4°C. Pelleted organoids were then lysed in 2× Laemmli sample buffer containing 4% of 2-mercaptoethanol, heated at 95°C for 10 minutes, sonicated, and heated again at 95°C for 5 minutes. Protein concentration was determined using Protein Assay Dye Reagent Concentrate (Bio-Rad), and bovine serum albumin (BSA) was used as a standard. Protein extracts were resolved by SDS-PAGE and transferred onto polyvinylidene difluoride membranes by western blot. Membranes were blocked with 3% BSA and incubated overnight at 4°C with primary antibodies. Horseradish peroxidase–conjugated secondary antibodies were applied at room temperature for 1 hour. The specifications and dilutions of antibodies used for western blotting are listed in Supplementary Table S5. Blots were developed using ECL plus solution (GE Healthcare) and imaged with the FUJI film Laser Image Analyzer.

The expression of IGF2 from promoter 1 (adult isoform), promoter 3 (fetal isoform), total IGF2, IGF1R, H19, and TGFβ mRNA levels were tested in the following sets of samples: HB tumor samples (n = 27), paired adjacent nontumor liver samples (n = 25), HB cell lines (HepG2 and Huh6 cells), organoids and, as controls, healthy liver samples (n = 5, from patients with focal nodular hyperplasia or cystadenoma, without underlying liver disease). Total RNA from tissues and cell lines was isolated using the RNeasy Mini Kit (Qiagen). The cDNA was reverse transcribed from 1 μg of total RNA using RNA to cDNA EcoDry Premix Double Primed (Clontech, Takara Biotechnology) in a total volume of 20 μL. Quantitative real-time PCR was performed using TaqMan probes for IGF2-P1 (Hs01005962-m1), IGF2-P3 (Hs00171254-m1), IGF2-P1/2/3 (Hs01005963-m1), IGF1R (Hs00609573-m1), H19 (Hs00262142_g1), and TGFβ (Hs00998133-m1). Each sample was analyzed in triplicate and normalized to the internal control 18S ribosomal RNA (Hs99999901-s1). The FC of mRNA expression in tumor tissue was calculated referring to the average mRNA expression levels in adjacent nontumor tissue.

To assess miR-483-5p expression levels, micro-RNA from tumor and paired adjacent tissue was obtained using the TaqMan Micro-RNA Reverse Transcription Kit (Thermo Fisher Scientific), and qRT-PCR was performed following the TaqMan Small RNA assay and TaqMan probes for miR-483-5p (Thermo Fisher Scientific, 002338). Each sample was analyzed in triplicate and normalized to the internal control miR-361-5p (Thermo Fisher Scientific, 000554). FC was calculated referring to the averaged mRNA expression levels in adjacent nontumor tissue.

Statistical analyses were conducted using R (version 3.6.2, R Foundation for Statistical Computing). Comparisons of continuous variables were performed using Kruskal–Wallis and Dunn tests for nonparametric distributions, or ANOVA and Tukey tests for parametric distributions. Correlations for categorical variables were analyzed using Fisher exact test. Survival and time to response were assessed with Kaplan–Meier estimates and log-rank test.

The data generated in this study, including RNA-sequencing, methylation, and SNP-array, are available within the article, its supplementary data files, and in Gene-Expression Omnibus (GSE104766, GSE132219, and GSE132037). Data from validation cohorts were obtained from Gene-Expression Omnibus (GSE132219 and GSE104766).

To identify molecular alterations driving oncogenesis in HB, we used RNA-seq data from a cohort of 31 HB tumors (ref. 21; 29 HBs and 2 HCN-NOS) and matched nontumor tissues (study cohort, Supplementary Fig. S1; Table 1). Through gene coexpression network analysis, we identified seven network modules that were significantly differentially expressed between tumor and nontumor samples from the study cohort. The top enriched gene ontology terms in these network modules included IGF2 signaling pathway (module 1), cell cycle and survival (modules 6–7), and immune response (modules 2–3; FDR<0.001, Supplementary Table S6). Notably, the Wnt signaling pathway, which is a commonly deregulated in HB (21), was also enriched in the network module 1 (GO:0016055, P = 0.009). We also explored the top interconnected genes within each network (hub genes) and identified several targetable cancer-related candidates including IGF2, TGFB1, and BIRC3 (Supplementary Fig. S3). Notably, IGF2 was the targetable gene with the highest expression levels in the study cohort (Supplementary Fig. S3; Supplementary Tables S7–S8), making it the most suitable candidate for our study.

We then assessed the rate of HB tumors overexpressing IGF2 in the study cohort. Of the 31 HBs, 22 (71%) were classified as IGF2^high based on RNA-seq data (FC > 2 vs. nontumor adjacent tissue), and this was aligned with the IGF2 levels measured by qRT-PCR (Fig. 1A and B) and with previous studies (14, 15). Additionally, IGF2 pathway deregulation was also observed at the level of IGF1R and H19 (Supplementary Fig. S4), in line with other reports (33).

These results were confirmed in two independent cohorts: in the first, our in-house validation cohort (n = 50), 78% of patients (39/50) exhibited high levels of IGF2; and in the second (23), 76% (19/25; Supplementary Fig. S4).

Overall, our results indicate that IGF2 overexpression is the main targetable alteration among the top deregulated pathways in our HB cohort and is present in 70% to 80% of cases.

We analyzed the transcriptomic and methylation profile of the study cohort to gain further insight into the molecular characteristics of IGF2^high HB tumors (Fig. 1B; Supplementary Fig. S5). IGF2^high tumors were enriched in high-risk molecular classes of HB, including the MRS-3, the C2-pure, the C2A, and the “Liver progenitor” classes of HB (15, 21, 23, 34), as well as the EpiCB epigenetic cluster signature (21), and the HCC proliferation class (35). IGF2^high tumors also showed enrichment in progenitor cell markers (EPCAM, AFP, DLK1, and PEG3) and gene sets associated with cell-cycle progression (Fig. 1B; Supplementary Fig. S5). Only 1 of 31 cases were found to have TERT promoter mutations, and this single case corresponded to an HCN-NOS tumor. In terms of clinical features, IGF2^high patients presented shorter recurrence-free survival after resection than IGF2^low patients (median 34 months vs. not reached for IGF2^low; P = 0.02; Fig. 1C), which align with their molecular features. No additional clinicopathologic differences were observed between IGF2^high and IGF2^low tumors (Table 1).

In contrast, tumors with low IGF2 levels (IGF2^low; 9/31, 29%) were enriched in low-risk and mesenchymal HB classes (MRS-1, C2B, and “Mesenchymal” HB classes; refs. 15, 21, 23), together with gene sets associated with inflammatory and interferon signaling, TGFβ pathway activation, and epithelial-to-mesenchymal transition (Fig. 1C). The high presence of inflammatory signaling was confirmed with gene signatures capturing specific immune cell populations including cytotoxic T cells (Supplementary Fig. S5). The expression levels of TGFB1 and TGFβ pathway-related genes were higher in IGF2^low tumors than in IGF2^high and nontumor samples (all P < 0.0001; Supplementary Fig. S5). These results indicate that IGF2^low tumors present a mesenchymal and highly infiltrated phenotype, which is in line with their association with a better outcome compared with IGF2^high tumors.

These molecular features of IGF2^high and IGF2^low HBs were confirmed in an independent HB cohort of 25 patients (23). In line with our data, gene-expression analysis in the external cohort revealed enrichment of proliferation and cell cycle–related gene sets in IGF2^high tumors, compared with inflammatory and interferon signaling in IGF2^low cases (FDR < 0.05; Supplementary Fig. S5).

Next, we investigated the mechanisms of IGF2 overexpression in HB. Previous reports have described aberrant IGF2 levels in HB associated with hypomethylation of the fetal P2, P3, and P4 promoters, as opposed to physiologic P1-mediated expression (36). Here, we used methylation 850K-array data to assess the methylation differences between IGF2^high and IGF2^low tumors. Tumors with high IGF2 expression levels presented significantly decreased methylation of the IGF2 fetal promoters compared with tumors with low IGF2 levels (Fig. 2A; Supplementary Fig. S6). In particular, IGF2 fetal promoter hypomethylation was present in 50% of IGF2^high samples (9/18 vs. 1/9 in IGF2^low samples, P < 0.05; Fig. 3). To assess whether the hypomethylation of the IGF2 fetal promoter correlated with higher expression of its fetal isoform, we expanded our analysis to the IGF2 splicing profile based on an HTA (human transcriptome array) expression array at exon level. Our data demonstrated that the IGF2 fetal isoform was overexpressed in IGF2^high tumors compared with IGF2^low tumors (Fig. 2B and C). Furthermore, overexpression of the fetal IGF2 isoform in IGF2^high was confirmed by qRT-PCR in both the study cohort and in-house validation cohort (Supplementary Fig. S6). Our results suggest that IGF2^high HB tumors present a fetal IGF2 expression pattern through hypomethylation of the fetal promoters. In addition, we also identified gain of methylation in the IGF2/H19 imprinting region ICR1 of IGF2^high tumors, in line with previous studies (15). Particularly, ICR1 gain of methylation was present in 61% of IGF2^high patients (11/18 vs. 1/9 in IGF2^low samples, P < 0.05; Fig. 3; Supplementary Fig. S6).

Additionally, the 11p15.5-imprinted locus, which contains IGF2, has also recently been reported as the second most frequently altered locus in hepatoblastoma (15), mostly through copy-neutral LOH. Because IGF2 presents monoallelic expression in normal conditions, LOH in this genomic region increases its expression (14, 37). We used our SNP-array data to determine the rates of this allelic imbalance in IGF2^high vs. IGF2^low HB tumors. IGF2^high tumors were significantly enriched in 11p15.5 LOH compared with IGF2^low tumors [12/21 (57%) vs. 1/9 (11%), P < 0.05; Fig. 2D and E; Fig. 3]. Interestingly, the four IGF2^high HB samples corresponding to infant patients (< 8 months of age) showed 11p15.5 LOH (Fig. 3), which suggests that their IGF2 overexpression was not due to physiologic regulation in fetal and early postnatal stages (38), but due to genetic aberrations.

Finally, the expression of the micro-RNA miR-483-5p has been reported to promote fetal IGF2 transcription by binding to its 5′ untranslated region (UTR; ref. 39) and as a mechanism of IGF2 overexpression in HCC (18). In our study cohort, miR-483-5p levels were significantly higher in IGF2^high HB tumors than in IGF2^low HB tumors, both according to the FC versus adjacent nontumor tissue in RNA-seq (2.4 vs. 0.7, P < 0.05) and qRT-PCR data (3.2 vs. 0.6, P < 0.001; Fig. 2F). Overall, miR-483-5p overexpression, defined as FC > 2 vs. adjacent nontumor samples, was detected in 55% (11/20) of IGF2^high HB tumors and only in 11% of IGF2^low tumors (P < 0.05; Fig. 3).

Altogether, our results indicate that IGF2 overexpression in HB is driven by hypomethylation of its fetal promoters, ICR1 deregulation, LOH in 11p15.5, and miR-483-5p expression. Most IGF2^high HB patients [91% (20/22)] presented at least one of these mechanisms of IGF2 overexpression (Fig. 3).

The presence of IGF2 overexpression in 71% of HB patients and their poor outcome prompted us to test IGF2 blockage in HB. In addition, the monoclonal antibody against IGF1 and IGF2, named xentuzumab, has shown promising antitumor effects in other cancer types (20). To this end, the antitumor effects of xentuzumab were evaluated in vitro (HB cell lines and HB PDX organoids) and in an in vivo murine model of HB. The effects were compared with those of a conventional therapeutic agent for HB patients (cisplatin; ref. 5). IGF2 expression levels in HB cell lines (HepG2 and Huh6) were assessed by qRT-PCR (Supplementary Fig. S7). HepG2 cells presented high IGF2 expression, defined as FC > 4 compared with nontumor adjacent tissue (median FC = 19.60), and were used as a model to recapitulate IGF2^high HB tumors. Conversely, Huh6 cells showed low IGF2 expression (median FC = 0.04; Supplementary Fig. S7) and thus were considered to phenocopy the IGF2^low HBs. As observed in IGF2^high human HB tumors, HepG2 cells also showed high expression of the receptor IGF1R, low H19 expression, and high expression of the fetal isoform of IGF2 (P2-4 derived; Supplementary Fig. S7). Previous studies (14) have reported that HepG2 cells present fetal promoter hypomethylation and LOH genomic imbalances as opposed to Huh6. For validation purposes, we generated a cell line derived from a human PDX model with high expression of IGF2 (hereinafter referred to as PDX-derived cell line).

We tested the effect of IGF2 blockade and chemotherapy in IGF2^high cells (HepG2 and PDX-derived cell line) and IGF2^low cells (Huh6). In IGF2^high cells, xentuzumab strongly decreased cell viability in combination with low doses of cisplatin (0.1–1 μmol/L), where cisplatin was not effective as monotherapy, eliciting a strong synergistic effect (combination indexes: HepG2, CI = 0.47; PDX-derived cells, CI = 0.44; Fig. 4A; Supplementary Fig. S7). In contrast, the synergism of IGF2 blockade and chemotherapy was not observed in Huh6 cells (IGF2^low), where the antiproliferative effects were mainly driven by cisplatin, with a mild added effect of xentuzumab (CI = 0.94; Fig. 4A).

Consistently, xentuzumab in combination with cisplatin reduced the colony formation capacity of IGF2^high HepG2 cells compared with monotherapies, including cisplatin (all P < 0.05; Fig. 4B). Conversely, the combination treatment did not show significant differences in the Huh6 cell line compared with monotherapies. When comparing the effect of the combination, this was 5.1-fold stronger in HepG2 cells (IGF2^high) than in Huh6 cells (IGF2^low), corresponding to a 46% vs. 9% reduction, respectively (P < 0.001; Supplementary Fig. S7). On the other hand, xentuzumab alone induced significant differences in colony formation compared with the control (all P < 0.01), but it did not induce superior antitumor effects compared with cisplatin monotherapy (Fig. 4B). The impact of xentuzumab alone in HepG2 cells was between 4.5-fold stronger than in Huh6 cells (22% vs. 5%, P < 0.01; Fig. 4B and Supplementary Fig. S7).

Overall, the IGF1/2 inhibitor xentuzumab in combination with cisplatin strongly reduced the proliferation and colony formation capacity of HB cells, and this effect was more pronounced in IGF2^high cells than in IGF2^low cells.

Considering the antitumor capacity of IGF2 inhibition with xentuzumab combined with cisplatin observed in IGF2^high HB cell lines, we proceeded to further evaluate the antitumor and molecular effects of this combination in more complex experimental models such as organoids derived from a PDX murine model generated from an IGF2^high HB tumor (31). First, we confirmed by RT-PCR that the high IGF2 levels in the original HB tumor were maintained in the organoids. Indeed, IGF2 levels in the organoid model were 19.5-fold higher when compared with mean IGF2 levels of the adjacent human liver tissue (Supplementary Table S9).

Next, organoids were treated with xentuzumab, cisplatin, xentuzumab + cisplatin, or control IgG, and organoid viability was assessed. A strong synergistic reduction in cell viability was observed with the combination of low doses of xentuzumab (1 nmol/L) and cisplatin (< 1 μmol/L), where monotherapies were not effective (combination index: 0.43; Fig. 4C). A synergistic peak was observed for 1 μmol/L cisplatin and 100 nmol/L xentuzumab. This was further confirmed by calculating the IC₅₀ for each condition. Indeed, the IC₅₀ obtained for cisplatin combined with 0.1 μmol/L xentuzumab was 25-fold lower than the one obtained for cisplatin as monotherapy (99.19 nmol/L vs. 2.57 μmol/L), whereas the IC₅₀ for xentuzumab alone was not reached with the assessed concentrations (>100 nmol/L; Fig. 4D; Supplementary Fig. S8).

Next, we investigated the effect of the combination of xentuzumab and cisplatin at low concentrations. The combination of xentuzumab and cisplatin activated apoptosis in a dose-dependent manner in organoids (Fig. 4E), whereas monotherapies did not. The combination also reduced the activation of the IGF1R pathway, as measured by western blotting of IGF1R and Akt phosphorylation, in both organoids and cell lines stimulated with IGF2 (Fig. 4F; Supplementary Fig. S9). Intriguingly, xentuzumab reduced IGF2 pathway activation, but this effect was markedly enhanced by the combination, partially explaining the synergism of the combination and stressing the importance of combining both treatments to obtain antitumoral responses.

Therefore, these data validate the results observed in cell lines, suggesting that the efficacy of cisplatin could be greatly improved by the combination with IGF2 inhibition in IGF2^high HB patients.

The antitumor potential of IGF2 inhibition in combination with chemotherapy was tested in a xenograft murine model of IGF2^high HB, generated through subcutaneous implantation of HepG2 cells. Mice bearing HB tumors were randomized to receive cisplatin, xentuzumab, cisplatin + xentuzumab, or control IgG. Mice treated with the combination of xentuzumab and cisplatin showed significantly increased survival (median survival: 16 days) compared with those treated with cisplatin alone (10 days, P = 0.038 vs. combination) or IgG control (13 days, P = 0.027 vs. combination; Fig. 5A). Conversely, the monotherapy treatment arms were unable to significantly reduce the viable tumor volume compared with the control. No significant differences in body weight or other toxicity signs were observed, indicating that all treatments were well tolerated (Supplementary Fig. S10).

Next, we analyzed the tumors and the antitumor effects of the treatments at the histologic level. The combination of cisplatin and xentuzumab significantly decreased the fraction of viable tumor volume compared with cisplatin alone (P < 0.01) and the control (P < 0.05; Fig. 5B and C). Furthermore, CD31 expression, a well-defined marker of angiogenesis, was significantly decreased in tumors from the combination treatment and cisplatin alone, likely because of the previously reported antiangiogenic effect of cisplatin treatment and IGF1/2 blockade (refs. 18, 40; Fig. 5D).

Overall, these results indicate that the combination of xentuzumab and cisplatin improved survival and showed an increased antitumor effect compared with cisplatin alone. In addition, our data provide the rationale for testing IGF2 inhibitors in combination with cisplatin in trials with HB patients who express high levels of IGF2.

HB is a rare pediatric liver cancer that has largely been understudied at the molecular level. Recently, the molecular profile of these tumors has been deciphered using multiomics approaches (9, 15, 21, 23, 41). Despite these efforts, there are still no available targeted therapies in HB. Hence, HB patients receive standard preoperative platin-based chemotherapy regimens (3), associated with serious adverse effects, which could be partially avoided with other therapeutic approaches (7). Herein, we provide evidence suggesting that IGF2 overexpression is a targetable alteration in HB (considering HBs and HCN-NOS) and describe the epigenetic mechanisms of IGF2 overexpression. Additionally, we showed remarkable antitumor effects of the combination of IGF2 inhibition and cisplatin in preclinical models.

Gene coexpression network analysis revealed that IGF2/IGF1R signaling is one of the most deregulated pathways in HB. In line with previous studies showing high IGF2 levels in HB (14, 42), IGF2 overexpression was detected in 71% of HB cases, being the most overexpressed targetable gene in tumors compared with nontumor tissue. IGF2 overexpression is associated with progenitor cell markers (EPCAM, AFP, DLK1, and PEG3) and HB molecular classes of high-risk tumors (C2, refs. 21, 34; MRS-3, ref. 21). In line with this, it has been reported that genes with epigenetic regulation such as IGF2 are abundantly expressed in the fetal liver (43) and that HB tumors with liver progenitor features are highly proliferative and have poor survival (15, 34). On the other hand, IGF2^low tumors were associated with a mesenchymal HB phenotype, with a higher presence of EMT and TGFβ signaling markers, together with higher immune infiltrate. Importantly, patients with IGF2^high tumors presented poorer recurrence-free survival compared with IGF2^low tumors, highlighting the need for new therapies for this subgroup of patients (44).

IGF2 is a well-known paternally imprinted gene that encodes a fetal peptide hormone that regulates cellular proliferation and differentiation during the early stages of development (45). It is highly expressed in the liver during fetal stages from the promoters P2, P3, and P4, and in adulthood, it is expressed from both alleles of promoter P1, albeit at much lower levels (36). Reactivation of IGF2 expression through fetal promoter demethylation has been described as a crucial mechanism for IGF2 overexpression in liver cancer (18, 41, 46). Other mechanisms of IGF2 overexpression in cancer involve ICR1 deregulation (15, 47), LOH of the 11p15.5 region (14, 15), and miR-483-5p expression (18, 48). Our integrative analysis revealed that these four mechanisms are pivotal in driving IGF2 overexpression in HB, explaining 91% of the cases of IGF2^high HB. Others have previously suggested a complex interaction between IGF and Wnt/β-catenin signaling and potential activation of IGF1R signaling due to CTNNB1 mutations in colon cancer (49, 50).

The high prevalence of IGF2 overexpression in HB and its association with poor outcomes make it an excellent candidate as a targetable alteration to develop personalized therapeutic approaches, because targeted therapeutic options for these patients are urgently needed. Targeted inhibition of IGF1/2 ligands with xentuzumab abrogates proliferative signaling through IGF1R, without affecting metabolic cell functions (18, 51). Earlier studies in HCC pointed to the receptor IGF1R as a potential target (17); however, subsequent clinical trials blocking this receptor using IGF1R mAb or IGF1R/insulin receptor tyrosine kinase inhibitors failed to demonstrate beneficial outcomes (44) and resulted in metabolic-related toxic effects due to the inhibition of insulin metabolic signaling. Therefore, xentuzumab has important advantages over previous attempts to inhibit the IGF2/IGF1R pathway.

In HCC and colon cancer in vivo models, xentuzumab has shown encouraging antitumor activity and a favorable safety profile, specific for IGF2^high tumors (18, 51). A potent antiproliferative effect of xentuzumab has also been reported in cell lines from several cancer types, including HCC and HB (15, 18). This promising background prompted us to further evaluate the potential of xentuzumab as a new targeted therapy for IGF2^high HB tumors by assessing its effects on tumor organoids and a murine xenograft model. In this regard, xentuzumab combined with cisplatin significantly improved the effects of the compounds administered as monotherapy. Specifically, xentuzumab in combination with cisplatin was able to improve by 25-fold the efficacy of cisplatin alone in HB organoids and demonstrated an enhanced antitumor effect in IGF2^high cells compared with IGF2^low, indicating a new treatment strategy that would increase survival in IGF2^high HB patients. In addition, the combination was effective at low cisplatin doses, suggesting that - in this setting - chemotherapy doses could be reduced and the chemotherapy-related adverse events diminished. This is especially important considering that chemotherapy causes serious adverse effects in HB patients such as ototoxicity (7, 52), which is detrimental to language and social development, particularly in young children (53). Of note, patient stratification based on IGF2 tumor levels at time of diagnosis would be required in clinical studies testing xentuzumab in HB.

On the other hand, in other tumor types, treatment with IGF2/IGF1R inhibitors has been demonstrated to reactivate the antitumor effect of DNA-damaging agents such as cisplatin (54). Our results support the idea that IGF1R pathway inhibition enhances sensitivity to DNA-damaging agents, as we see reactivation of the efficacy of cisplatin in the combination setting both in our HepG2 xenograft model and in our in vitro models.

The combination reduced aberrant angiogenesis, as observed by CD31 staining, similarly to monotherapies, and in line with the previously described antiangiogenic effects of xentuzumab and cisplatin (18, 40) in other tumor types. Specifically, both IGF2 blockage and cisplatin have been reported to inhibit aberrant vasculature formation by downregulating VEGF signaling (55) and promoting the release of matrix metalloproteinases (40), respectively.

One limitation of the current study lies in the lack of specific cellular models for each of the newly described HB subtypes. Thus, despite HepG2 cells lines have been recently suggested to be derived from an HCN-NOS hepatoblastoma tumors (10), in the present study were used as model overall representing HB cases with high levels of IGF2. In this sense, clinical studies are awaited to clarify whether patients with HCN-NOS tumors need to be managed different from patients with HB tumors.

A second limitation of the study was the use of post-chemotherapy specimens. Here, we acknowledge that the implementation of our findings into the clinical setting will require further validation in diagnostic biopsies and the evaluation of the safety and antitumoral effects of the combination in ad hoc preclinical models.

Overall, this study defines IGF2 as a promising molecular target in HB and provides evidence that IGF2 inhibition with xentuzumab enhances the efficacy of cisplatin. Therefore, combining xentuzumab with cisplatin could exert a greater antitumor effect than cisplatin alone in a subset of HB patients (∼70%) presenting IGF2 overexpression. This provides a preclinical rationale for exploring this targeted therapy in combination with cisplatin in patients with IGF2^high HB.

L. Torrens reports grants from Boehringer Ingelheim during the conduct of the study. M. Puigvehi reports personal fees from AbbVie and Gilead outside the submitted work. P. Czauderna reports grants from European Commission during the conduct of the study. J.M. Llovet reports grants from Alex's Lemonade Stand–Foundation for Childhood Cancer during the conduct of the study; grants and personal fees from Bayer HealthCare Pharmaceuticals, Eisai Inc, Ipsen, Bristol-Myers Squibb, personal fees from Eli Lilly, Glycotest, Roche, AstraZeneca, Merck, Genentech, Omega Therapeutics, Mina Alpha, Boston Scientific, Exelixis, BlueJay, and Captor Therapeutics outside the submitted work. No disclosures were reported by the other authors.

J. Abril-Fornaguera: Conceptualization, data curation, formal analysis, investigation, visualization, writing–original draft. L. Torrens: Conceptualization, data curation, formal analysis, investigation, writing–original draft. C. Andreu-Oller: Investigation, writing–review and editing. J. Carrillo-Reixach: Investigation, visualization, methodology, writing–original draft. A. Rialdi: Investigation, visualization. U. Balaseviciute: Investigation. R. Pinyol: Investigation, writing–review and editing. C. Montironi: Investigation. P.K. Haber: Investigation. Á. Del Río-Álvarez: Investigation. M. Domingo-Sàbat: Visualization. L. Royo: Investigation. N.K. Akers: Investigation. C.E. Willoughby: Investigation, writing–review and editing. J. Peix: Investigation. M. Torres-Martin: Investigation. M. Puigvehi: Investigation. S. Cairo: Resources. M. Childs: Resources. R. Maibach: Investigation. R. Alaggio: Resources. P. Czauderna: Resources. B. Morland: Resources. B. Losic: Investigation. V. Mazzaferro: Resources. E. Guccione: Resources. D. Sia: Conceptualization, supervision, investigation, methodology, writing–review and editing. C. Armengol: Conceptualization, resources, data curation, software, formal analysis, supervision, funding acquisition, validation, investigation, writing–review and editing. J.M. Llovet: Conceptualization, resources, data curation, software, formal analysis, supervision, funding acquisition, investigation, methodology, writing–original draft, project administration, writing–review and editing.

We thank Dr. Catherine Guettier, from the Université Paris-Saclay, for the pathologic review of the samples included in the study. This study was funded by a grant from Boehringer Ingelheim. J. Abril-Fornaguera was supported by the PREDOCS-UB grant from the University of Barcelona (UB) and by a mobility grant from Fundació Universitària Agustí Pedro i Pons. J. Carrillo-Reixach and A. Del Río-Álvarez were supported by the Catalan Agency for the Management of University and Research Grants (AGAUR, 2019 FI_B01024 and 2022_FI_B00528). U. Balaseviciute was supported by the EILF-EASL Juan Rodés PhD Studentship from the European Association for the Study of the Liver (EASL) and the EASL International Liver Foundation (EILF). R. Pinyol was supported by the University of Barcelona (UB) and IDIBAPS/Fundació Clínic Barcelona. C. Montironi was supported by a Rio Hortega grant from Instituto de Salud Carlos III (ISCIII), Fondo Social Europeo (CM19/00039). P.K. Haber was supported by the fellowship grant of the German Research Foundation (DFG/HA 8754/1-1). M. Puigvehi received a Juan Rodés scholarship grant from Asociación Española para el Estudio del Hígado (AEEH). C.E. Willoughby was supported by a Sara Borrell fellowship (CD19/00109) from the ISCIII and Fondo Social Europeo. J. Peix received support from the Generalitat de Catalunya (PERIS, SLT017/20/000206). E. Guccione was supported by Alex's Lemonade Stand Foundation. D. Sia was supported by the Gilead Sciences Research Scholar Program in Liver Disease. C. Armengol received funding from Instituto de Salud Carlos III (PI10/02082 and PI13/02340), the European Union's Horizon 2020 research and innovation program under grant agreement N° 668596 (ChiLTERN) and N° 826121 (iPC), CIBERehd (CB06/04/0033), AGAUR (2017-SGR-490), and the Ramón y Cajal (RYC-2010-07249) program of the Spanish Ministry of Science and Innovation. J. M. Llovet was supported by grants from the Samuel Waxman Cancer Research Foundation; the Spanish National Health Institute (MICINN, PID2019-105378RB-I00); NIH (R01 DK128289-01); through a partnership between Cancer Research UK (CRUK), Fondazione AIRC per la Ricerca sul Cancro and Fundación Científica de la Asociación Española Contra el Cáncer (FAECC; Accelerator Award, HUNTER, Ref. C9380/A26813); and by grants from the Acadèmia de Ciències Mèdiques i de la Salut de Catalunya i Balears; and the Fundación Científica de la Asociación Española Contra el Cáncer (FAECC; Ref. PRYGN223117LLOV).

The publication costs of this article were defrayed in part by the payment of publication fees. Therefore, and solely to indicate this fact, this article is hereby marked “advertisement” in accordance with 18 USC section 1734.