Patient-derived mouse xenografts from pediatric liver cancer predict tumor recurrence and advise clinical management

PDX ESTABLISHMENT

This study was approved by the Assistance Publique-Hôpitaux de Paris ethical committee, and informed written consent was obtained from each patient. Tumor samples were grafted in the interscapular region of 6- to 8-week-old female athymic nude mice. Growing tumors were serially transplanted onto recipient mice and underwent comparative examination to confirm preservation of their histological features and live-frozen to immortalize the model. No impact of tumor growth on liver function of recipient mice was observed (Supporting Table S1). More information is available as Supporting Information.

PDX HISTOLOGICAL AND IMMUNOHISTOCHEMICAL CHARACTERIZATION

Histological and immunohistochemistry studies were performed in the pathology laboratories at Bicêtre Hospital, Gustave Roussy Institute, and Necker Hospital using the Benchmark Autostainer (Roche) and antibodies currently in use in the clinical practice as described in Cairo et al.29 All detailed immunostaining protocols are available upon request.

AFP DETECTION

AFP detection in blood of pediatric liver cancer patient-derived xenograft (PLC-PDX)-carrying mice was performed at Bicêtre Hospital using the AFP diagnostic kit BRAHMS AFP KRYPTOR (Thermo Scientific) and 80-100 μL of blood per animal, according to the manufacturer's instruction.

CYTOGENETIC ARRAY

Genomic DNA (gDNA) from eight primary tumours and two intrahepatic recurrences from 9 patients and their corresponding PDXs were profiled by using the cytogenetic array CytoScan HD (Affymetrix). Details are provided as Supporting Information. The statistical environment R (v 3.1.1)30 with the package OmicCircos was used to represent genomic alterations in a Circos plot.31 Data are available at the Gene Expression Omnibus database under accession number GSE76100.

CTNBB1 AND NFE2L2 MUTATION

CTNNB1 mutation analysis was performed on tumor complementary DNA by polymerase chain reaction (PCR) followed by Sanger sequencing as described in Cairo et al.29 PCR conditions for detecting mutations in exon 2 of the NFE2L2 gene are described in Eichenmüller et al.18

FIFTY-EIGHT-GENE PROFILING FOR HB214, HB217, AND RT001 HUMAN AND PDX SAMPLES

Mutation profiling of 58 genes, among the most frequently mutated in cancer according to the COSMIC database, was performed at BGI in Beijing (China) on gDNA by exon trapping with NimbleGen microarray followed by deep sequencing by using Illumina's HiSeq technology, with at least 50× effective mean depth for each sample.

IN VIVO STUDIES

For efficacy studies, mice with subcutaneously growing tumor were allocated to each treatment arm according to their tumor volume A compound inducing a tumor volume reduction of treated versus control group superior to 42% was considered as active according to previously published rules.32 However, for this study we adopted the more stringent PPTP tumor response criteria.33 More details are provided as Supporting Information.

STATISTICAL ANALYSES

All statistical analyses were performed by using the tests described in the article and the GraphPad Prism (version 5) software.

ESTABLISHMENT OF A PRECLINICAL PANEL OF CHILDHOOD LIVER CANCER PDXs

From May 2010 to August 2015, 58 tumor specimens from 51 patients were grafted. These tumors consisted of 48 HBs, one TLCT, one lung metastasis with HCC features, two RTs, two HSs, and four FLCs (Supporting Table S2). The large majority of samples were from primary tumors that received preoperative chemotherapy or at relapse. From these samples, 24 PLC-PDXs were generated from 20 patients, including 20 HB-PDXs, one TLCT-PDX, one HCC-PDX, and two RT-PDXs, whereas no PDXs could be established from HS and FLC samples. Concerning HB samples, successful grafting was observed for the majority of specimens from recurrent or metastatic tumors (9 of 12 [75%]), whereas for primary tumors the take rate was lower (13 of 39 [33.3%]). The characteristics of the PLC-PDXs established recapitulated the heterogeneity observed in the clinic, particularly for the most aggressive forms of pediatric liver cancer. Among the HB-PDXs, one model was derived from an AFP-negative HB, a very aggressive form of the disease observed in less than 1% of cases.14 For 6 patients, we could graft primary tumors and recurrent disease. For patients 13, 19, and 35, PDXs could be established from the recurrent diseases, but not from the primary tumors, whereas for patients 26, 37, and 41, we could generate models from both the primary tumor and at relapse.

HISTOLOGICAL AND MOLECULAR CHARACTERIZATION OF PLC-PDXs

We evaluated the histological, genomic, and genetic relevance of the tumor grafts as compared to the parental patient's tumors. PLC-PDX retained the histological features of the tumors of origin, particularly concerning the tumor cell morphology and tissue organization of the more undifferentiated components (Fig. 1A, left panel). To verify the stability of the histological phenotype over passages, HB-214 PDX histology was checked at passages 2 and 4. Both passages closely resembled the tumor of origin (Fig. 1A, right panel). Immunohistochemical analysis of diagnostic markers currently used to classify liver malignancies showed similar staining of the histological components in the parental tumors and in corresponding PLC-PDXs (Fig. 1B; Supporting Table S3). However, for many PLC-PDXs originating from primary tumors, we noticed over-representation of the more undifferentiated components, namely, embryonal and small cell undifferentiated (SCUD; Table 1). On the other hand, PLC-PDXs generated from intrahepatic recurrences or distant metastases closely reproduced the parental tumor histology, as shown, in particular, for patients 26 and 37, suggesting that the PLC-PDX models could be more representative of the relapsing disease than of the primary pathology.

PLC-PDXs REFLECT THE GENOMIC LANDSCAPE OF THE PATIENT POPULATION

Given that the most frequent genetic alterations in HB are mutation of CTNNB1 and NFE2L2, we investigated mutation of these genes in tumors and PDXs by reverse-transcriptase PCR and Sanger sequencing. Identical CTNNB1 mutations were found in patients' tumors and matched PDXs, at a frequency very similar to that found in the patients' population, arguing against a possible growth advantage in mice for the tumors harboring these mutations. CTNNB1 mutation was found in 19 of 22 PLC-PDXs (RT-PDXs are excluded from the analysis), corresponding to 16 of 18 (89%) patients with mutation, with predominant exon 3 deletion, observed in 13 PDXs from 11 patients (69%), and point mutation in six PDXs from 5 patients (31%). Among the three PLC-PDXs with wild-type CTNNB1 are the HCC-PDX and two HB-PDXs established from the same patient. As expected, NFE2L2 showed a much lower mutation rate, we could identify G81S mutation only in two PDXs from the same patient.18 We also checked INI1/SMARCB1 mutation for the two RT-PDXs developed, and found deletion of this gene in the paired donor/PDX samples (Fig. 2A). HB and RT share a very low incidence of mutation as observed by NGS, next-generation sequencing (NGS) analysis.12, 17, 20 To check whether PDX models maintained this distinctive feature, we analyzed the mutational status of 58 genes among the most frequently mutated in solid tumors by exon sequencing in the HB213 parental tumor and PDX model, HB-214 PDX and nontumor sample, HB-217 PDX, and RT001 parental tumor and PDX. Besides CTNNB1, NFE2L2, and INI1/SMARCB1 mutations, no other mutation could be found, and rare single-nucleotide polymorphisms (SNPs), with allele frequency of less than 5% in the Caucasian population, were identified in the paired tumor/PDX or nontumor/PDX samples (Supporting Table S4).

We next evaluated further the relevance of the models by CytoScan HD chip (Affymetrix) for 10 paired donor samples and established PLC-PDXs (nine HB and the one TLCT). First, we evaluated the maintenance of 530,026 SNPs present in the array and found that a mean of 97.7 ± 2.4% of the SNPs were maintained between paired parental tumor and PDX. Second, we studied the copy number variations (CNVs) of the same samples. The CIRCOS plot shown in Fig. 2B illustrates that chromosomal alterations are unique to each tumor and are, overall, well preserved in the corresponding PDXs. In that regard, 78% of the CNVs in the primary tumors are maintained in the PDXs. Only the HB-238-RED-231 tumor/PDX pair showed several discrepancies in the chromosomal CNV, which could be partly attributed to reduced tumor cellularity in the parental tumor. Among frequent alterations, we observed gain of chromosome 1q in 60% of paired cases associated to 1p loss in 2 cases, chromosome 4/4q-ter loss (50%), gain of chromosome 8/8q (50%), gain of chromosome 17 (60%), and gain of chromosome 20 in 90% of paired samples. We also noticed that in HB-239 and in HB-244-RED-239 pairs, which correspond to the primary tumor and its intrahepatic recurrence, some genomic alterations were unique to the primary tumor (i.e., loss of chromosome 1p, 4, 7, 11, and 18) and were well reflected in the corresponding PDX models.

PDXs RECAPITULATE HB BIOLOGY AND PHYSIOPATHOLOGY

Besides investigating the genetic and histological features of PLC-PDXs, we measured tumor physiopathological properties, such as tumor growth and AFP secretion. These two parameters were well preserved in the PDXs established (Fig. 3A; Supporting Tables S5 and S6). Following primo-implantation, appearance of the tumor burden in animals was observed with a delay varying between 2 weeks and 4 months. Tumorgrafts from the same patient that developed in different mice and with different latency period showed overlapping growth curve and histological traits, suggesting different percentages of living tumor cells in the fragments implanted rather than selection of different tumor clones. Each PLC-PDX showed unique growth curves (Fig. 3A; Supporting Table S5), and AFP secretion in the blood of HB-carrying mice was concordant with the level observed in patients (Supporting Table S6). Mice carrying HB-229 PDX, derived from a nonsecreting tumor, showed very low AFP in the bloodstream. The HCC-PDX showed weaker AFP expression compared to the AFP-secreting HB models, and the PDX with highest relative secretion levels was the TLCT-PDX. Conversely, no AFP was detected in the blood from RT-001 PDX.

HB GROWTH IN MOUSE PREDICTS POOR PROGNOSIS

To evaluate whether the ability to establish a PDX is dependent on specific clinicopathological features, we compared the clinical and histological parameters of all HBs implanted into recipient mice (Table 2). Because HCC-PDX was generated from a recurrence in a patient previously operated of HB and TLCT-PDX shares HB and HCC features, they were included in the analysis, whereas patients with rhabdoid tumors were excluded. Statistical analysis showed that the two parameters most strongly associated with successful tumor growth in mice are the percentage of necrotic/fibrotic tissue in the resected tumor and circulating AFP levels at the end of preoperative chemotherapy (P < 0.0001 and P = 0.0002, respectively; Fisher's exact test). All models were derived from tumors with necrosis/fibrosis not exceeding 50% and high levels of circulating AFP posttreatment measured a few days before surgery. High AFP level at diagnosis (P = 0.0565; Fisher's exact test) and decreased AFP level posttreatment measured in log₁₀ scale less than 3-fold (P = 0.0378; Fisher's exact test) are also associated with tumor growth in mouse. Other annotations significantly associated with xenograft growth are the implantation of tumor fragments from relapsing tumors (P = 0.0066; Fisher's exact test), prevalence of an immature epithelial component (P = 0.0011; Fisher's exact test), and, curiously, a preoperatory tumor extent (PRETEXT) stage II or less (P = 0.017; Fisher's exact test). We also observed that primary tumors that allowed PDX establishment relapsed more frequently than those that failed to grow in mice (P = 0.0413; Fisher's exact test). This led us to also investigate whether PDX development can predict patient outcome. PDX growth was associated with shortened disease-free survival (hazard ratio [HR] = 8.494; P = 0.0049; log-rank [Mantel-Cox] test; P = 0.0038; Gehan-Breslow-Wilcoxon test), and with decreased overall survival (HR = 7.224; P = 0.0152; log-rank (Mantel-Cox) test; P = 0.0037; Gehan-Breslow-Wilcoxon test; Fig. 3B). These results indicate that the ability of pediatric liver tumors to grow in immunodeficient mice is predictive of patient outcome, as already shown for other types of cancer.34

PLC-PDXs SHOW DIFFERENT SENSITIVITY TO CISPLATIN

To evaluate PLC-PDX response to standard-of-care treatments, the impact of doxorubicin and cisplatin administration was tested in eight PLC-PDXs. For the in vivo experiments, we used PPTP guidelines33 as read out to interpret tumor response. HB-217 PDX showed impaired tumor growth, classified as stable disease (SD), when treated with cisplatin alone or in combination with doxorubicin, whereas HB-213 and HB-214 were resistant to both treatments, and tumor response was annotated as progressive disease (PD; Fig. 4). Doxorubicin alone had no effect on tumor growth and did not show cooperative effect when administered in combination with cisplatin, except for a slight effect in HB-214 between day 9 (D9) and D14. Response to cisplatin alone was tested in two additional HB-PDXs, the TLCT and the HCC-PDX, with weak or no response classified as PD. Because the RT donor for RT-001 PDX was misdiagnosed as HB, it was initially treated according to the HR protocol, to which it showed no response. When we treated RT-001 PDX with cisplatin and doxorubicin as single drugs or together, the derived xenograft proved completely insensitive, in agreement with what was observed for the patient's tumor.

IRINOTECAN/TEMOZOLOMIDE COMBINATION INDUCES STRONG TUMOR REGRESSION IN PDXs FROM RECURRENT HBs WITH AGGRESSIVE CLINICAL FEATURES

To evaluate whether these models could help to identify new therapeutic options for children with aggressive, relapsing HB, we tested several compounds (irinotecan, temozolomide, temsirolimus, sirolimus, nefopam, sorafenib, crizotinib, and paclitaxel) that are currently being tested in children with recurrent HB, alone or in combination. These agents were administered to three PDXs that recapitulate the most aggressive forms of HB: the TLCT-derived TT-001 model, the AFP-negative, SCUD-enriched HB-229 PDX, and the HC-001 model (Table 3). In a first set of experiments using mice with TT-001 PDX, we clearly observed that irinotecan alone or in combination with temozolomide were the only treatments capable of inducing tumor regression (Table 3; Fig. 5). Because the irinotecan/temozolomide combination induced toxicity (Supporting Figure S1A), irinotecan administration was suspended after the first administration, which explains the sudden regrowth observed in the related curve (Fig. 5). Conversely, all other treatments induced only partial or inconsistent growth inhibition in this PDX model (Table 3). In light of the toxicity problems observed, we introduced a reduced dosage of 10 mg/kg of irinotecan every 5 days instead of 40 mg/kg twice a week for the HB-229 PDX model. The reduced irinotecan/temozolomide combination was well tolerated (Supporting Fig. S1B) and very effective on this model, given that all tumors underwent maintained complete response (MCR; Table 3; Fig. 5) with no evidence of disease at the end of the study, 76 days after beginning of treatment (not shown). Irinotecan alone induced transient tumor stabilization, whereas the other treatment groups were classified as progressive disease. In the different models, tumor sensitivity resulted in decreased level of circulating AFP (Supporting Table S7). Histological analysis of residual tumors after chemotherapy was carried out in 2 of the 6 mice bearing HB-229 at the end of the experiment (D76, 7 weeks after the end of treatment). In both animals, treatment efficacy was witnessed by the presence of few residual dysmorphic tumor cells with nuclear vacuoles. In mouse 64, these cells were all Ki67 negative, whereas in mouse 112 two very small tumor cell foci had gained back their original structure and were positive to Ki67 staining (Supporting Fig. S2). This observation strongly correlated with circulating AFP rising above the detectability threshold at this late time point (Supporting Table S7), suggesting tumor relapse. Conversely, nonresponder HC-001 only showed slightly decreased Ki67 staining (not shown). The HCC-PDX study clearly showed that no treatment could induce tumor regression or stabilization, and the best result was attenuation of tumor growth by irinotecan alone or in combination with either temozolomide or sirolimus (Table 3). Sensitivity to the irinotecan/sirolimus combination was witnessed by the decrease of circulating AFP in treated mice (Supporting Table S7). Of note, treatment with irinotecan in combination with temozolomide also showed efficacy on HB-239, but not on its intrahepatic relapse HB-244-RED-239 PDXs (Table 3; Fig. 5). This result perfectly matched the clinical situation, given that no response was observed when the patient was treated with irinotecan/temozolomide at tumor recurrence after liver transplantation. Altogether, these results reinforce the use of irinotecan as a possible new important compound for second-line treatment of patients with HB and the combination with temozolomide as a potentially effective therapeutic option for children with aggressive, recurrent HB.