CNS-PNETs with C19MC amplification and/or LIN28 expression comprise a distinct histogenetic diagnostic and therapeutic entity

Tumor specimens and clinical information were collected with consent as per protocols approved by Hospital Research Ethics Board at participating institutions. A total of 450 primary pediatric brain tumors with various histologic diagnoses—103 CNS-PNETs, 45 atypical rhabdoid teratoid tumors (ATRTs), 128 medulloblastomas (MBs), 105 ependymomas (EPNs), 50 high-grade gliomas (HGGs) and 20 choroid plexus carcinomas (CPCs) were examined in this study (Supplemental Table 1). All ATRTs diagnoses were confirmed for genetic alterations of SMARCB1/INI1 by Multiplex Ligation mediated PCR and/or targeted gene sequencing and for loss of SMARCB1/INI1/BAF47 protein expression by immunostaining. Of the 54 group 1 CNS-PNETs examined for detailed histopathologic correlates (Table 1 and Supplemental Table 2), 33 were previously reported [13, 17]; and 21 additional tumors with a diagnosis of CNS-PNET, ETANTR or embryonal tumor with multilayered rosettes (ETMR), MEP and EPB were collected for this study. Three tumors with an initial diagnosis of EPB were histologically reviewed and re-classified as ETANTR (PNET 5, 39 and 54). Tumor DNA and RNA were extracted by standard methods, quantified using NanoDrop Analyzer, and analyzed using the Illumina HT-12 v4 gene expression and 450 K Methylation arrays and OmniQuad 2.5 M SNP genotyping arrays (http://​www.​illumina.​com) to generate gene expression, methylation, and copy number profiles. Methylation profiling was performed at The Centre for Innovation at Genome Quebec; all other analyses were performed at The Centre of Applied Genomics (TCAG) at the Hospital for Sick Children.

Hematoxylin and eosin (H and E) staining was performed using standard protocols. Fluorescence in situ hybridization (FISH) and immunohistochemistry (IHC) were performed on 5-μm formalin-fixed paraffin-embedded (FFPE) sections for individual tumors or tissue microarrays (TMA). TMA construction was as previously detailed [18]. Tumor tissue blocks and corresponding slides were reviewed by an experienced neuropathologist (CH) for diagnostic accuracy and adequacy of tissues; samples that had extensive necrosis or <60 % tumor content were discarded. Representative tumor areas were identified, and three 1-mm tissue cores were selected with the goal of obtaining a sampling accuracy >95 % for each tumor represented on the TMA [6, 8]. A variety of tissues including liver, fetal cerebellum, placenta, breast carcinoma, and basal cell carcinoma were included on each array to serve as internal controls for various immuno-stains.

For FISH analyses, pre-labeled BAC probes (http://​www.​tcag.​ca/​cytogenomics) mapping to chr19q13.42 (RP11-381E3: 162, 225 bp) and an internal control chr19p13.11 locus (RP11-451E20: 165,783 bp) were used as described previously [13]. For IHC analyses, a heat-induced antigen retrieval process was used, followed by blocking endogenous peroxidase and biotin. Primary anti-LIN28 (#3978) and anti-phospho-S6 (Ser240/244) (#5364) antibodies were purchased from Cell Signaling Technology (Boston, MA, USA) and anti-DNMT3B (ab13604) was purchased from Abcam (Toronto, ON, CA). Antibody reactivity was visualized using the VectaStain ABC detection kit, Vector Laboratories (Burlingame, CA, USA).

For each protein, cytoplasmic and/or nuclear immunopositivity was visually scored based on both strength (0, 1, 2, 3–3+) and distribution (<25, 25–50, 51–75, and >75 % of tumor cells). Only tumors with strong staining (3–3+) in >25 % of tumor cells were considered to be positive. Tumors with FFPE slides were scored based on analyses of the whole tumor section, while tumors on TMAs were assessed on the basis of staining patterns in at least two of three tissue cores. Human testicular germ cell tumor tissue was used as a positive control for LIN28 staining, while sections processed in parallel without primary antibodies were used as negative controls. All immunohistochemical stains were scored independently by TS, DP, MB and NH, and reviewed by AH and CH.

Gene expression data from 59 primary CNS-PNETs arising in different locations were generated using the Illumina HT-12v4 arrays and normalized as previously described [17]. Methylation data generated using the Illumina human 450 k arrays for 45 primary CNS-PNETs were background-normalized in Genome Studio (v. 2011.1) to obtain beta values for downstream analyses. All X and Y chromosome probes (n = 11,649), single-nucleotide polymorphisms (dbSNP, n = 88,679), and unannotated probes (relative to hg19, n = 65) were excluded leaving a total of 385,184 probes for methylation analyses. Genes and probes were ranked by largest coefficient of variation and standard deviation, respectively. Unsupervised hierarchical clustering (HCL; Partek Genomics Suite, v6.6) of gene expression and methylation data were, respectively, established using Pearson’s Dissimilarity performed iteratively on 200–2,000 genes and 200–10,000 probes to identify a minimal gene set associated with the most stable gene expression and methylation tumor cluster patterns (Supplemental Fig. 2).

Median age at diagnosis was compared using Kruskal–Wallis test and Log Rank (Mantel–Cox) analysis was used to assess survival. All other clinical and biological characteristics were compared using Pearson Chi-Square (χ ²). A p value of <0.05 was regarded as significant for all analyses. All statistical analyses were done using SPSS version 22.0.

To generate stable tumor cell line, A664, a pre-treatment primary tumor from a non-C19MC amplified/LIN28+ CNS-PNET (PNET398) was dissociated by gentle manual trituration followed by passage through a 40-μm mesh filter. Cells were grown in low-adhesion tissue culture flasks (Sarstedt, Montreal, QC, CA) in defined serum-free media at 37 °C, 5 % CO₂. Culture media consisted of human neural stem cell proliferation media (Stem Cell Technologies, Vancouver, BC, CA) supplemented with heparin, epidermal growth factor (EGF, 10 ng/ml, Sigma Aldrich, St. Louis, MO, USA), and fibroblast growth factor (FGF, 10 ng/ml, Stem Cell Technologies, Vancouver, BC, CA). Cells were dissociated using Accumax (Millipore, MA, USA) by manual trituration and re-plated at ~40 % confluency (20,000 cells/ml) and maintained stably for >10 consecutive passages.

For drug treatment and growth assays, A664 cells were plated in 96-well dishes at 2,000 cells/well and incubated overnight. Cells were then treated with rapamycin (Sigma Aldrich, St. Louis, MO, USA), 5-azacytidine (Sigma Aldrich, St. Louis, MO, USA) or vorinostat (Selleckchem, Burlington, ON, CA) and evaluated for changes in cell proliferation assays at various days post-treatment using CellTiter 96^® AQueous One Solution Cell Proliferation Assay kit (Promega, Madison WI, USA). Viable cell numbers were determined based on absorbance at 575 nm using a Versamax microplate reader (Molecular Devices, Sunnyvale, CA, USA).

For LIN28 knockdown, A664 cells were plated at a density of 1.5 × 10⁵ cells/ml in 6-well plates. Scrambled control or LIN28-specific siRNA (Fisher Scientific, Ottawa, ON, CA) was transiently transfected into cells using Lipofectamine 2000 (Life Technologies, Burlington, ON, CA) as per the manufacturer’s protocol. Cells were harvested at 48-h post-transfection for protein extraction and analysis by Western blotting. Western blot analysis and chemiluminescence detection were performed using standard protocols. Secondary HRP-linked donkey anti-Rabbit or HRP-linked sheep anti-Mouse (#NAV934, 931) was from GE Healthcare (Baie-d’Urfe, QC, CA).

For miRNA-specific qRT-PCR, single-stranded cDNA was synthesized from 10 ng of RNA using a miR-specific stem-loop reverse-transcription RT-primer and the TaqMan^® MicroRNA Reverse Transcription Kit (Life Technologies, Burlington, ON, CA). qRT-PCR was performed using TaqMan^® Universal PCR Master Mix, no AmpErase^® UNG (Life Technologies, Burlington, ON, CA) according to the manufacturer’s instructions; miRNA expression levels were normalized relative to RNU6B using the ΔC _t method, and compared to that of 16-week-old whole human fetal brain control.

We [13, 17] and others [10, 11] previously reported frequent amplification of the C19MC oncogenic miRNA cluster and high LIN28 protein expression in an aggressive sub-group of CNS-PNETs. To comprehensively define the clinico-pathologic features of these newly recognized molecular entities, we first tested the diagnostic robustness of the C19MC and LIN28 markers in a spectrum of pediatric brain tumors for which materials were available for central re-review of histopathology and confirmation of histologic diagnosis according to the current WHO CNS tumor classification criteria. These comprised an institutional cohort of 128 MBs, 45 ATRTs, 105 EPNs, 50 HGGs, 20 CPCs, and 103 CNS-PNETs from various institutions; 450 and 416 samples had materials available for FISH and LIN28 IHC analyses. As reported previously, C19MC amplification or copy number gains were restricted to and detected in 24.3 % (25/103) of CNS-PNETs, while LIN28 immunopositivity was observed in 21.4 % (22/103) of all CNS-PNETs. However, in contrast to C19MC amplification, cytoplasmic LIN28 immunopositivity was not restricted to CNS-PNETs but was also observed in a spectrum of other tumors including 24.4 % (11/45) of ATRTs and 19.5 % (8/41) of HGGs (Supplemental Table 1). These results indicate that LIN28 immunopositivity alone may not be sufficient to establish a diagnosis of CNS-PNETs.

To evaluate the relationship of CNS-PNETs with C19MC amplification and LIN28 positivity to known histologic sub-types of CNS-PNETs [14], we next conducted more detailed evaluation of tumor histopathology and clinical features in an expanded cohort of 54 CNS-PNETs identified on the basis of C19MC amplification and/or LIN28 expression (Table 1, Supplemental Table 2). Interestingly, of 45 tumors with complete histopathologic review, 22/45 (49 %) exhibited ependymoblastic rosettes with neuropil consistent with a histologic diagnosis of ETANTR (including 3 tumors initially diagnosed as EPBs), 12/45 (26.7 %) displayed papillary features characteristic of MEP with variable amounts of neuropil and rosette formation, while a subset of 11/45 (24 %) tumors lacked characteristic histologic features of ETANTRs, EPBs or MEPs but exhibited variable differentiation or relatively bland histology compatible with classic or CNS-PNET-NOS (Fig. 1). In addition to spanning histologic sub-classes, we observed that C19MC amplified/LIN28+ CNS-PNETs also exhibited multiple CNS locations; cerebral origins were most common (41/54; 76 %), but 24 % (13/54) arose in other CNS locations including the cerebellum (6/54; 11 %), brain stem (5/54; 9 %), pre-sacral space (1/54; 2 %) and optic nerve (1/54; 2 %). Of note, 5/37 (14 %) tumors with overlapping data on C19MC and LIN28 status exhibited LIN28 positivity only, without changes in C19MC expression or copy number; these tumors were found in each histologic class (please see Supplementary Table 2 for details).

Analyses of clinical features showed that 24/37 (65 %) of tumors were non-metastatic at diagnosis and arose predominantly in young children with a median age of 29 months at diagnosis. The incidence of tumor metastases, patient demographics (gender and age) and overall survival did not differ significantly across histologic classes. Detailed survival data available for 36/54 patients indicated a similar highly aggressive course with rapid disease progression and death for a majority of patients regardless of tumor histologic class (Fig. 2a). The median survival for all patients was only 12 ± 1.9 months (range of 0–165 months). However, patients who received treatment had significantly longer survival as compared to untreated patients (median survival of 13 ± 1.9 versus 0.060 ± 1.84 months; Fig. 2b). Notably, six patients (5 with specific treatment information) had substantially longer survival of 38–204 months; three patients remained alive at 56, 165 and 204 months (Supplemental Table 2). These findings suggest that chemotherapy ± radiotherapy treatment may benefit a subset of patients.

We had previously demonstrated using global gene expression profiles that C19MC amplified/LIN28+ CNS-PNETs arising in the cerebral hemispheres comprised a molecular class with relative enrichment of primitive neural features that was distinct from two other molecular groups (designated group 2/3) of hemispheric CNS-PNETs [17]. The convergence of clinical phenotypes observed in C19MC amplified/LIN28+ tumors with different histologic features and anatomic locations, prompted us to investigate whether all C19MC amplified/LIN28+ tumors, regardless of histology or location, comprised a common molecular entity. Indeed, unsupervised hierarchical cluster analyses of gene expression data from 22 C19MC amplified/LIN28+ tumors, which included five ETANTRs, seven MEPs, and ten PNETs from different CNS locations, clustered together as a common molecular class distinct from Group 2/3 CNS-PNETs (Fig. 3a). Similarly, unsupervised cluster analyses of global methylation profiles generated from an overlapping cohort of 19 C19MC amplified/LIN28+ tumors, comprised of 5 ETANTRs, 5 MEPs, and 9 PNETs from various locations, showed they formed a common molecular cluster distinct from the group 2/3 hemispheric CNS-PNETs (Fig. 3b). In keeping with our prior studies of hemispheric CNS-PNETs [13, 17], C19MC amplified/LIN28+ tumors with different location and histology also exhibited a gene expression signature enriched for pluripotent and neural cell lineage genes. Notably, all tumors with or without C19MC amplification expressed high levels of pluripotent genes LIN28/LIN28B, and low levels of neural differentiation genes including the NF1 family of transcription factors (Fig. 3c, d). Collectively, our analyses indicate that all CNS-PNETs with C19MC amplification and/or LIN28 immunopositivity, regardless of histologic sub-class or location, share similar molecular and genetic makeup, thus indicating they comprise a common biological and diagnostic entity.

In recent studies, we showed that the LIN28/let7/P13K-mTOR pathway is up-regulated and can be targeted by Rapamycin in a C19MC amplified primary ETANTR cell line [19]. In addition, we demonstrated that the unique methylation landscape of C19MC amplified tumors is in part imposed via C19MC-mediated repression of RBL2, and consequent up-regulation of DNMT3B [9]. The unified molecular signature of LIN28+/C19MC amplified and non-amplified CNS-PNET indicated that they may share common therapeutic pathways. We observed that LIN28/LIN28B and P13K-mTOR components, IGFBP1-3, which have been linked to a common functional pathway [20, 22], as well as DNMT3B were enriched in gene expression signatures of CNS-PNETs with or without C19MC amplification (Fig. 3d). Consistent with these observations, IHC analyses revealed up-regulation of phospho-S6 (pS6), an mTOR pathway target, as well as DNMT3B in C19MC amplified and non-amplified primary tumors (Fig. 4a, b).

To investigate whether pharmacologic inhibitors of mTOR and DNA methylation may have therapeutic roles in all group 1 CNS-PNETs, we tested the effects of rapamycin as well pharmacologic inhibitors of DNA methyl transferase function (5-azacytidine and vorinostat), on growth of a tumor cell line (A664) derived from a non-C19MC amplified primary cerebral PNET398. Using quantitative RT-PCR and siRNA-mediated knockdown of LIN28, we confirmed that the LIN28/let7/P13K-mTOR axis was conserved in A664 cells. Indeed, rapamycin treatment significantly inhibited A664 cell growth with concomitant down-regulation of pS6 expression (Fig. 5b–d). Similarly, we observed that 5-azacytidine and vorinostat significantly inhibited A664 cell growth in a dose-dependent manner (Fig. 5e, f). Collectively, these data highlight regulators of P13K/mTOR signaling, as well epigenomic modifiers, as novel promising therapeutic targets for these recalcitrant infantile tumors.