New insights into diagnosis and therapeutic options for proliferative hepatoblastoma

PATIENT SAMPLES

All patients were recruited in accordance with European and French law and institutional ethical guidelines.17 A set of 73 liver samples (44 HB and 29 nontumoral [NT] liver including 29 pairs of tumor and adjacent NT liver) was collected from 39 patients treated at French university hospitals or from the SIOPEL Liver Tumor and Tissue Bank (www.siopel.org).

RNA-SEQUENCING ANALYSIS

Total RNA from 50 frozen patient tissues constituting the tumoral and adjacent NT tissues and two cell lines was extracted using the mirVana kit (Thermo Fisher Scientific) according to the supplier's protocol. The RNA integrity number was evaluated by the Agilent 2100 Bioanalyzer. A total of 4 μg was used for generation of each poly-A library with an Illumina TruSeq Stranded mRNA Sample Preparation kit according to standard protocol. Paired-end 125-nucleotide sequencing was performed by the MGX-Montpellier GenomiX platform on an Illumina HiSeq 2500. The quality of the sequencing was evaluated by FastQC and mapped to the human genome (GRCh38) with CRAC (version 2.5.0 with options “-k 22,” “--no-ambiguity,” and “--stranded”),18 and the biological events, including single-nucleotide polymorphisms, indels, and splice junctions, were extracted using cractools (version 1.24 and command “extract”). Reads for each gene were summarized by featureCounts19 using the annotation from Ensembl (version 82). Differential expression was performed by DESeq220 between tumoral subgroups and/or NT samples. Genes with a log2 fold-change absolute value >2 and a false discovery rate <0.05 were considered differentially expressed. Principal component analysis was performed with FactoMineR21 on the regularized logarithm20 of the 1,000 most expressed genes. Sample clustering was performed on variance stabilizing transformed counts20 by heatmap.2 R function from the gplots package, using default parameters (Euclidian distance and “complete” clustering method). Gene set enrichment analysis (GSEA) of MSigDB v5.0 set H and C2:CGP was performed with GSEA 2.1.022 using the results of RNA sequencing (RNA-seq) collapsed to gene names using the Ensembl 82 annotations. Twenty-four NT samples were contrasted with either 18 C1, four C2B, or five C2A samples (three HBs and two cell lines). The sequencing data were deposited in Gene Expression Omnibus (accession number GSE104766).

STATISTICAL ANALYSIS

Statistical analyses were performed using GraphPad Prism 6.0 software as described.17

See Supporting Information for further methods.

RNA-SEQ DATA REVEAL THREE SUBGROUPS OF HB

We performed RNA-seq transcriptomic analysis of 25 paired tumor and 25 NT HB samples and HB cell lines HuH6 and HepG2. The CRAC pipeline was used for read processing and mapping, whereas DESeq2 was used for read normalization and differential gene analysis. We used the 1,000 most expressed genes to visualize sample distances by principal component analysis (Fig. 1A). All NT samples and class C1 HB formed well-defined clusters. Samples from group C2 were more dispersed and appeared to separate into two distinct clusters that we termed “C2A” and “C2B.” The transcriptome profiles of HB cell lines were very similar to that of the C2A group. An analysis of differential gene expression was conducted between groups C1 and C2, and all samples were reclustered hierarchically according to the 300 most differentially expressed genes. Clustering with this set of differentially expressed genes reproduced the apparent clustering observed for the principal component analysis and, importantly, confirmed the division of group C2 into two subgroups (Fig. 1B). However, the grouping of patients appeared to correspond only partially with clinical features such as high or standard risk, invasion, and age. Patients assigned as high risk, according to PRETEXT or the Children's Hepatic Tumors International Collaboration–Hepatoblastoma Stratification classification, were spread among the four clusters, whereas some of the C2 group HB known to have worse survival were assigned to the low-risk group in the Children's Hepatic Tumors International Collaboration–Hepatoblastoma Stratification classification. This result confirmed the relevance of associating clinical stratification and molecular analysis to increase the diagnostic efficiency of high-risk HB.

From RNA-seq results we extracted β-catenin mutation status in tumor and NT samples (Fig. 1B, bottom panel). We found that 88% of tumors were affected in β-catenin, with 32% carrying a point mutation and 56% having exon 3 deleted. These percentages are similar to those previously reported for HB in other cohorts.8, 9, 13, 14 Further comparisons with clinical data were limited due to the small number of samples. However, we observed some unusual characteristics in the C2 group: wild-type β-catenin (two of four samples) and other than fetal component (two of four samples) in C2B and high alpha-fetoprotein expression (three of three samples) and the atypical-fetal component in C2A (one of three samples).

Lastly, the expression values for the genes comprising the previously proposed 16-gene signature8 were extracted from the RNA-seq data (Fig. 1C; Supporting Fig. S1) and investigated to determine if this signature was sufficient to distinguish the various groups. This signature, which was designed initially to distinguish between HB groups C1 and C2, indeed identified the C1 and C2 samples but failed to completely separate C2A from C2B (Fig. 1C; Supporting Fig. S1). Moreover, despite being very effective, the 16-gene signature requires analytical expertise for correct interpretation. Therefore, patient stratification requires a more suitable signature, preferably with a unique biomarker for each tumoral subgroup.

A ROBUST FOUR-GENE SIGNATURE FOR MOLECULAR CLASSIFICATION OF HB

From the RNA-seq analysis, we searched for a cohort of genes that possessed the greatest potential to discriminate among the tumoral subgroups. Two important criteria we held in selecting potential candidates were a high level of RNA expression and the availability of suitable antibodies for routine detection of either altered mRNA or protein levels, respectively. This analysis yielded a set of four genes, hydroxysteroid 17-beta dehydrogenase 6 (HSD17B6), integrin alpha 6 (ITGA6), TOP2A, and vimentin (VIM) (Fig. 2A), which were determined to be sufficient to recapitulate the HB subgrouping shown in Fig. 1A. Then, to evaluate this four-gene signature on other cohorts, we reclustered samples from previously published transcriptomic data sets: E-MEXP-1851 from Cairo et al.8 and GSE75271 from Sumazin et al.9 Using expression of probes corresponding to HSD17B6, ITGA6, TOP2A, and VIM, we recapitulated the correct clustering for 78 out of 80 samples (97.5%) (Supporting Fig. S2A,B).

HSD17B6 is a liver-specific cytoplasmic oxidoreductase member of the retinol dehydrogenase family involved in androgen catabolism. ITGA6 is a cell-surface protein involved in cell adhesion and overexpressed in breast cancer stem cells.23 TOP2A is a well-described nuclear enzyme controlling the topologic states of DNA, and it is involved in fundamental processes such as chromosome condensation.24 VIM is a type 3 intermediate filament protein expressed in mesenchymal cells.25 Differential expression of these genes in the various HB subgroups was determined by quantitative RT-PCR on an expanded cohort of 44 HB and 34 NT samples. We saw strict correlation in differential gene expression between quantitative RT-PCR and RNA-seq results, with the HSD17B6 and TOP2A increases being particularly accurate in characterizing NT tissue and C2A HB, respectively (Fig. 2B). Lastly, we used immunohistochemistry to confirm expression of the protein products of the four genes in 13 of the samples (Fig. 2C). Because HSD17B6 is typically expressed in liver, as expected, it was detected exclusively in the cytoplasm of NT hepatocytes (Fig. 2C). ITGA6 was detected primarily at the membrane of tumoral hepatocytes and less frequently in the cytoplasm, with a stronger diffused staining in the C1 group (Fig. 2C). TOP2A was detected in all six C2A samples, where it was visible in the nucleus of 20%-30% of malignant hepatocytes. Strikingly, the protein was absent from samples from all other HB subgroups. These results, supported by the significant up-regulation of TOP2A in C2 cases analyzed in cohorts from Cairo et al.8 and Sumazin et al.9 (Supporting Fig. S2C), indicate that TOP2A is a suitable biomarker for the high-risk C2A HB subgroup (Fig. 2C). Finally, VIM protein expression was observed in all subgroups. However, only in C2B hepatocytes was it organized into visible spindle-shaped structures, whereas it was restricted to the stroma in C1 and C2A tumoral hepatocytes (Fig. 2C).

TRANSCRIPTOMIC STUDIES SHOW MOLECULAR TARGETS IN HB SUBGROUPS

In order to identify biological pathways specifically deregulated in high-risk HB, we performed a GSEA on the differential gene expression data (Fig. 3). All tumor samples displayed consistent up-regulation of notable HB genes, such as LIN28B,26 DLK1,27 the Wnt/beta-catenin pathway targets AXIN2 and LEF1, and their inhibitors NKD1 and DKK1. Genes involved in typical liver functions, including xenobiotic, lipid, and bile acid metabolism, and many long noncoding RNAs, such as antisense to KIRREL3, CALCRL, HAL, and the intergenic RP11-434D9.1 and LINC01093, were generally down-regulated in tumor samples, suggesting an important role in normal liver function. Samples from group C2B were distinguished by an up-regulated expression of TGFB1 and the transforming growth factor-β targets CTGF and JAG1. We also found higher transcript levels for SNAI2, S100A4, and VIM, which are markers of the epithelial–mesenchymal transition,28 and for the mesenchymal stem cell marker SPARC. Accordingly, high expression of VIM could indicate a mesenchymal phenotype for the C2B group.29 We also found higher transcript levels for the transcription factor RUNX230 and cadherin CDH11, which is characteristic for osteoblastic cell lines31 and suggests differentiation of mesenchymal stem cells into osteoblast-like cells. This is additionally supported by the histological data indicating the presence of osteoid foci in four out of six tested C2B samples (see Supporting Table S1).

Because our aim was to find targetable pathways in high-risk HB, we chose to focus on proliferative group C2A. The transcriptomic data showed that the HuH6 and HepG2 cell lines are analogous to the C2A group, and thus, the cell lines may represent a good model to characterize novel oncogenic pathways and test new drugs against HB. In this subgroup and the two cell lines, we found a persistent higher level of transcript for genes responsible for activation of cell-cycle checkpoints, chromatid assembly, mitosis, and such transcriptional pathways as E2F and MYC, which characterize highly proliferative tumors (Fig. 3B).32 Specific analysis of Ki67 expression from RNA-seq data confirmed the proliferative feature of C2A tumors. Strong immunolabeling of Ki67 was already observed for certain C2A tumors (data not shown), and transcriptomic results revealed a similar expression profile but 3 times higher for TOP2A (Supporting Fig. S3).

Genetic permutation analysis also revealed breast cancer 1 (BRCA1) network up-regulation and doxorubicin resistance in C2A (Fig. 3C). The FA–BRCA pathway is known to coordinate double-strand DNA homologous recombination repair. It employs a nuclear protein complex (FA core complex) that ubiquitinates Fanconi anemia D2 (FANCD2) and Fanconi anemia I (FANCI), leading to the recruitment of a nuclease complex, which initiates the nucleolytic incisions and DNA repair through translesion DNA synthesis polymerases.33 Specific inherited mutations in BRCA proteins are well known to increase the risk of various types of cancer.34 We observed a significant increase in transcript levels of three key components of the FA pathway, FANCD2, FANCI, and BRCA1, in group C2A within the quantitative PCR study on the HB cohort of 78 samples mentioned above (Fig. 3D), suggesting that the FA–BRCA pathway may constitute a target in proliferative HB.

TARGETING THE FA PATHWAY WITH BORTEZOMIB IMPAIRS HB CELL VIABILITY IN VITRO AND TUMORAL GROWTH IN VIVO

There are three well-known classes of FA–BRCA pathway inhibitors, the USP1/UAF1 inhibitors GW7647, ML323, and pimozide35; the NEDD8 inhibitor MLN492436; and proteasome inhibitors such as MG132 and bortezomib.37 The effect of proteasome inhibitors on apoptosis induction and growth arrest of hepatic cell lines (including HuH6 and HepG2) have been described in vitro38-41 but never investigated and detailed in this context of an up-regulation of the FA pathway in proliferative HB. Therefore, we tested all of the above-mentioned inhibitors on HuH6 and HepG2 cell lines to determine their median effect concentration (EC50) according to a 3-(4,5-dimethylthiazol-2-yl)-5-(3-carboxymethoxyphenyl)-2-(4-sulfophenyl)-2H-tetrazolium assay as a measure of cell viability (Supporting Fig. S4A). The EC50 values of GW7647, ML323, and MLN4924 on the HB cell lines were very high (15-979 μM), making them irrelevant for clinical use in HB. Pimozide had an effect on cell viability at a lower concentration (EC50 = 2.15-2.99 μM). Very interestingly, we found that the proteasome inhibitors MG132 and bortezomib were the most efficient with very low EC50 values (7 nM and 62 nM for bortezomib in HuH6 and HepG2 cells, respectively), rending them compatible for clinical applications in HB. Then we tested the effect of each drug at its EC50 alone or in combination with cisplatin (Fig. 4; Supporting Fig. S4B). GW7647 and ML323 had no additive effect with cisplatin on cell viability, whereas pimozide had only a slight additive effect. Surprisingly, we observed a significant synergic effect of the NEDD8 inhibitor (MLN4924) in combination with cisplatin, but its high EC50 excludes it from viable treatment options. Lastly, proteasome inhibitors had a significant additive effect on HB cell viability in combination with cisplatin (Fig. 4A), also confirmed by sulforhodamine B assays (Fig. 4B). Those tests confirmed that proteasome inhibitors could be promising treatments for HB.

In contrast to MG132, bortezomib is a reversible proteasome inhibitor already approved for treatment of pediatric acute lymphoblastic leukemia42 under the brand Velcade. Therefore, we preferentially used bortezomib for further analyses. As expected, after 3 days of treatment, bortezomib had a dramatic effect on HB cells, inducing early (annexin V) and late (annexin V–7-amino-actinomycin D) apoptosis in HuH6 and HepG2 cells; cisplatin alone had no significant apoptotic effect, and the combination of the two drugs did not further enhance bortezomib efficacy (Fig. 4C; Supporting Fig. S5). We also tested the effect of the two drugs on the clonogenic capacity of HuH6 cells (Fig. 4D). After 10 days, 44% and 7% of clones remained alive after treatment with 9 μM cisplatin and 7 nM bortezomib, respectively; and the combination of the two drugs yielded total elimination of clones (Fig. 4D). Finally, we tested the effect of bortezomib on tumor growth in the HuH6 xenograft mice model. The first sign of a significant decrease in tumor size was observed after 16 days, exclusively for mice treated with the combination of cisplatin and bortezomib. However, by the end of the experiment at day 22, the treatments with individual drugs were found to be as effective as the combined treatment (Fig. 4E).

BORTEZOMIB INHIBITS DNA REPAIR IN HB CELLS THROUGH FA–BRCA PATHWAY INACTIVATION

Homologous repair of double-strand DNA damage is partly coordinated by the FA–BRCA pathway, which can be inhibited by proteasome inhibitors such as bortezomib.37, 43 We therefore evaluated the effect of bortezomib on DNA repair in HB cells and more precisely on the FA pathway. Phosphorylation of S139 on H2AX is one of the earliest events following the induction of DNA damage. In order to evaluate whether bortezomib suppresses DNA repair in HB cells, DNA double-strand breaks were induced by γ-irradiation, and H2AX phosphorylation was examined. In untreated HuH6 and HepG2 cells, phospho-H2AX increased more than 6-fold and 4-fold, respectively, 2 hours after irradiation and then decreased to basal levels within 24 hours (Fig. 5A). With bortezomib pretreatment, the phospho-H2AX level steadily accumulated in both cell types in response to irradiation to levels similar to those observed for untreated cells; but this phosphorylation persisted for a further 30 hours after irradiation (Fig. 5A). FA pathway activity is regulated by the monoubiquitination of proteins FANCD2 and FANCI, which then promote recruitment of DNA damage effectors, for example, BRCA1, to nuclear foci where repair is required.37, 44 In HuH6 cells, FANCD2 and FANCI were already ubiquitinated, indicating that the FA pathway was constitutively activated in these cells (Fig. 5B). In contrast, irradiation slightly increased ubiquitination of FANCD2 and FANCI in HepG2 cells. After bortezomib treatment, the ubiquitination state of FANCD2 and FANCI was dramatically reduced in both cell lines, indicating inhibition of the FA–BRCA pathway (Fig. 5C). It was expected that inhibition of the pathway by bortezomib would attenuate the formation of repair foci in response to irradiation. HB cells treated with bortezomib prior to irradiation displayed a significant reduction in FANCD2 and BRCA1 foci after 24 hours as shown by reduced cellular immunofluorescence, whereas FANCD2 and BRCA1 protein levels remained unaffected (Fig. 5B-D; Supporting Fig. S6). Despite repeated attempts, we did not manage to detect FANCI by immunofluorescence. Altogether, these results indicate that global DNA repair capacity is severely compromised in cells treated with bortezomib due to an almost complete inactivation of the FA–BRCA signaling pathway.