Patient cohorts
Patient with initial surgery for MB at Beijing Tian Tan hospital, Capital Medical University between 2002 to 2018 were included in the study. Tumor specimens were stored in liquid nitrogen or in formalin-fixed paraffin-embedded (FFPE) blocks at Beijing Neurosurgical Institute. Patients with complete clinical data (such as age, EOR, and survival), preoperative magnetic-resonance image (MRI) scans, and surgical tissues obtained during initial surgery (before radiation or any other adjuvant treatment) were included. The EOR was established based on surgeons’ reports and confirmed with postoperative enhanced T1-weighted MRI scans. All normal cerebellums were provided by the Human Brain Bank (Chinese Academy of Medical Sciences & Peking Union Medical College).
Molecular subgroup analysis
Molecular classification (wingless [WNT], sonic hedgehog [SHH], Group 3, or Group 4) was established using the Agilent Whole Human Genome Oligo Microarray Kit, 4×44K (Agilent Technologies, Santa Clara, CA, USA; GSE116028; n = 38), the QuantiGene Plex Gene Expression Assay (QGP, version 2.0, Affymetrix, Santa Clara, CA, USA, n = 97), and immunohistochemical (IHC) staining with antibodies (β-catenin, SFRP1, NPR3, and KCNA1, n = 117), as we described previously [27, 28].
UHPLC-MS/MS analysis
Absolute quantities of 5hmC were measured, as we previously described [29]. Briefly, DNA isolation was performed using the Wizard® Genomic DNA Purification Kit (A1620, Promega, Madison, WI, USA) according to the manufacturer’s protocol. Nucleosides were separated by UHPLC on a T3 column (Waters, 186003538) and detected using a triple-4 quadrupole tandem MS instrument (Waters, ACQUITY UPLC XEVO TQ-S). The mass/change (m/e) transitions of 228.4 to 112.2 (cytosine), 242.3 to 126.1 (mC), and 258.2 to 124.2 (hmC) were monitored and recorded. Quantification was performed in comparison with standard curves generated using pure nucleoside standards, which were run with the same batch of samples. 5hmC percentages were calculated using the formula: 5hmC% = M (5hmC)/(M [cytosine] + M [5mC]) × 100.
Immunohistochemistry analysis
Immunohistochemistry (IHC) staining for 5hmC, Ki-67 and TET1/2 was performed on FFPE sections, as previously described [28, 30, 31]. Briefly, FFPE tissues were cut into 4-μm sections, followed by deparaffinization and rehydration using xylene and ethanol. Next, the slides were incubated in 3% hydrogen peroxide for 10 min in phosphate-buffered salineand then in blocking solution (CSA II Kit; Dako, Glostrup, Denmark) for 60 minutes at room temperature. The slides were incubated overnight with primary antibodies against 5hmC (1:800, ab214728, Abcam, US), Ki-67 (1:1000, ab15580, Abcam), TET1 (1:1000, HPA019032, Sigma, US), and TET2 (1:100, ab94580, Abcam). The number of pixels representing positively stained nuclei was detected using Image-Pro Plus image-analysis software (Media Cybernetics, Inc, MD, US). Positive staining was defined as a dark-brown staining pattern, confined to the nuclear region. For each sample, the mean value of ten snapshots was calculated to represent the percentage of positive cells. For 5hmC staining, normal cerebellum and non-tumor cells (endothelial cells) in the microenvironment were used as a positive-control tissue. All IHC slides were separately reviewed by two senior neuropathologists (L.L. and T.W.). A final score for 5hmC staining was then calculated by multiplying the score of proportion of positively stained tumor cells (0-100%) and the score of staining intensity (0,1,2,3).
Gene expression analysis
To evaluate the expression levels of 5hmC-related genes in MB, we downloaded normalized gene-expression data generated with the Removal of Unwanted Variation method from the Gene Expression Omnibus database (accession number GSE124814; https://www.ncbi.nlm.nih.gov/geo/), which included cerebellar data for 1350 patients with MB and 291 normal brain samples [32]. Expression levels of the TET1/2 genes and the isocitrate dehydrogenase 1 and 2 (IDH1 and IDH2) genes were analyzed in MB and normal cerebellum samples.
We detected the RNA-expression levels of TET1/2 in our MB samples using the QGP Assay. All samples were analyzed using a Luminex® instrument, and gene-expression levels was normalized to the geometric mean of the expression for 2 housekeeping genes (ACTB and GAPDH).
TP53 mutation analysis
Mutations of TP53 gene (exons 2 through 11) in SHH-MB was detected by Sanger sequencing, as previously described [33]. Briefly, genomic DNA derived from FFPE samples was prepared with Wizard® Genomic DNA Purification Kit (A1120, Promega, US) according to the manufacturer's protocols. The PCR profile was performed as follows: 95°C for 2 min, 56°C for 1 min for 35 cycles. The final extension was added at 72°C for 10 min before storage at 4°C. The PCR products were loaded onto a 1% of agarose gel for electrophoresis. The reactions were analyzed by an automated Genetic Analyzer ABI 310 system (ABI, CA) according to the manufacturer's instructions.
Statistical analysis
Significant differences between two groups were analyzed using Student’s t-test (two-tailed). Data are reported as mean ± standard deviation (SD). Relationships were evaluated by the Pearson correlation coefficient. For the survival analyses, OS was defined as the time from diagnosis until death, and PFS was defined as the time from the date of surgical resection until the date of tumor progression. Estimated 5- year OS and PFS were calculated using Kaplan–Meier analysis and data are reported as the mean ± standard error (SE). Patient cohorts were divided into two groups according to 5hmC levels. The optimal cut-off value (0.061 for 5hmC) was defined as the point with the most significant split using Cutoff Finder (http:// molpath.charite.de/cutoff/index.jsp). Significant differences between survival curves were determined using the log-rank test. The discriminatory capacity of 5hmC-based classification was evaluated by Harrell’s C index and time-dependent receiver operating characteristic (ROC) curve analysis, using the “survival ROC” package in R software, as previously described [34].
Univariate and multivariable Cox proportional hazard regression to estimate hazard ratios (HRs) for PFS and OS, including 95% confidence intervals (CIs). The multivariate model was performed with backward stepwise selection (inclusion criterion: score test, P < 0.10; exclusion criterion: likelihood-ratio test, P > 0.10). Variables included the 5hmC levels (low vs. high), the molecular subgroup (WNT, SHH, Group 3, or Group 4), the pathological subtype (classic MB [CMB], desmoplastic/nodular MB [DNMB], or large cell/anaplastic MB [LC/AMB]), age (<3, 3–17, or ≥18 years old), metastatic status (yes vs. no), and receipt of craniospinal irradiation (CSI; yes vs. no). Covariates (EOR and receipt of chemotherapy), which were deemed as important prognostic factors in previous studies[7, 15], were also included in the Cox regression models.
The nomograms were established based on the Cox model for predicting 3-, 5-, and 10-year PFS and OS rates of the validation cohort using the “rms” package (version 4-4.2) of R software. Calibration curves were generated to compare associations between the observed and predicted outcomes, as previously described [35]. Time-dependent ROC curve analysis was used to evaluate the discriminative ability of the nomogram. All statistical analyses were performed using the SPSS Statistical Package software (version 23.0, IBM Inc., Chicago, US) or R software (version 3.4.3; http://www.r-project.org). P < 0.05 was considered to reflect a statistically significant difference.