Genetic alterations of TP53 and OTX2 indicate increased risk of relapse in WNT medulloblastomas

This study aimed to re-evaluate the prognostic impact of TP53 mutations and to identify specific chromosomal aberrations as possible prognostic markers in WNT-activated medulloblastoma (WNT-MB). In a cohort of 191 patients with WNT-MBs, mutations in CTNNB1, APC, and TP53 were analyzed by DNA sequencing. Chromosomal copy-number aberrations were assessed by molecular inversion probe technology (MIP), SNP6, or 850k methylation array hybridization. Prognostic impact was evaluated in 120 patients with follow-up data from the HIT2000 medulloblastoma trial or HIT registries. CTNNB1 mutations were present in 92.2%, and APC mutations in 6.8% of samples. One CTNNB1 wild-type tumor gained WNT activation due to homozygous FBXW7 deletion. Monosomy 6 was present in 78.6%, and more frequent in children than adults. 16.1% of tumor samples showed TP53 mutations, of those 60% with nuclear positivity for the p53 protein. Loss of heterozygosity at the TP53 locus (chromosome 17p13.1) was found in 40.7% (11/27) of TP53 mutant tumor samples and in 12.6% of TP53 wild-type cases (13/103). Patients with tumors harboring TP53 mutations showed significant worse progression-free survival (PFS; 5-year-PFS 68% versus 93%, p = 0.001), and were enriched for chromosomes 17p (p = 0.001), 10, and 13 losses. Gains of OTX2 (14q22.3) occurred in 38.9% of samples and were associated with poor PFS and OS (5-year-PFS 72% versus 93%, p = 0.017 resp. 5-year-OS 83% versus 97%, p = 0.006). Multivariable Cox regression analysis for PFS/OS identified both genetic alterations as independent prognostic markers. Our data suggest that patients with WNT-MB carrying TP53 mutations or OTX2 gains (58.1%) are at higher risk of relapse. Eligibility of these patients for therapy de-escalation trials needs to be debated. Supplementary Information The online version contains supplementary material available at 10.1007/s00401-022-02505-5.


Introduction
Activation of the WNT signaling pathway is present in about 10% of all medulloblastomas (MB). These are now considered an independent MB entity with favorable outcome [11,30]. WNT-pathway activation was first identified by immunohistochemical demonstration of nuclear β-catenin accumulation [11]. However, nuclear β-catenin accumulation alone is believed to be insufficient to detect all WNT-MBs as some cases lack detectable nuclear accumulation [12,16,26], and current neuropathological guidelines recommend the combination of different methods to identify WNT-MB.
In most WNT-MBs, pathway activation is a result of activating mutations in exon 3 of CTNNB1, leading to stabilization and subsequent nuclear accumulation of its product β-catenin [10,13,23,47]. However, other members of the WNT signaling pathway like the tumor suppressor genes APC, Axin1, or Axin2/Conductin were shown to harbor mutations as alternative genetic alterations [9,19,22,23], while only CTNNB1 and APC mutations were found in a substantial fraction of WNT-MBs [33,35,38,43,44]. The favorable prognosis of WNT-MBs seems independent of the activating mutation [40,43]. A substantial proportion of APC-mutant WNT-MB occur in the context of an APC germline mutation (either familial or de novo) [43]. In these cases, the MB is part of the tumor spectrum of familial adenomatous polyposis (FAP), and usually occurs as first manifestation of the disease before multiple colorectal adenomas are present.
Given the very favorable prognosis with current standard therapy, patients with WNT-MB are enrolled in clinical studies aiming at a reduction of treatment intensity and hereby spare treatment-related side effects. Common to most of these trials is the reduction of radiotherapy dose to the whole brain and whole spine, either to 15 Gy (NCT01878617, St. Jude), or 18 Gy (NCT02724579, Children's Oncology Group, NCI, and NCT02066220, SIOP-Europe), which is very likely to reduce cognitive sequelae for the patients [31]. However, most patients with WNT-MB cannot be rescued when the disease recurs after standard therapy [32]. Therefore, it is of major importance to find biomarkers for identification of patients in which intensity of treatment should not be reduced, and of those patients qualifying for reduction of treatment intensity. In this large retrospective cohort, we aimed to re-evaluate the TP53 status as prognostic factor in WNT-MBs, and to screen for chromosomal aberrations and other biomarkers as possible prognostic factors.

Patients
Over a period of 20 years (2000-2020), 191 MB samples with confirmed WNT activation were identified. WNT-pathway activation was defined by the presence of at least two of the following features: nuclear positivity of β-catenin, mutation of a WNT-pathway component, e.g., CTNNB1 (exon 3) or APC, and/or monosomy 6.
One-hundred and twenty patients participated in either the HIT-2000 trial (NCT00303810), the HIT-2000interim registry (NCT02238899), or the I-HIT-MED registry (NCT02417324), so detailed information on the treatment and follow-up was available (clinical cohort). The HIT2000 trial and both registries have been approved by the responsible ethics committees. Informed consent for exploratory research on tumor tissue had been given by the patients or their legal representatives.

Molecular inversion probe array (MIP)
A MIP array including 330,000 inversion probes (Version v2.0, Affymetrix, Santa Clara, CA, USA) was used as previously described to identify copy-number gains and losses in 108 samples [42]. The raw MIP data file was analyzed using the Nexus Copy Number 8.0 Discovery Edition software (BioDiscovery, El Segundo, CA, USA). BioDiscovery's SNP-FASST2-Segmentation algorithm was used to make copy number and loss of heterozygosity calls. Further details are given in Online Resource Methods.
Forty WNT-MBs, including 13 CTNNB1-wild-type cases, were screened for APC mutations by one of two DNA-NGS panels as further described in Online Resource Methods and Online Resource Table 2.

450k/850k methylation bead arrays
Genome-wide methylation profiles were generated by hybridization to Illumina 450 k/850 k methylation bead arrays in 94 cases. The Heidelberg Brain Tumor classifier v11b4 was used to annotate the profiles to a methylation class [4,18]; scores > 0.9 were considered as match. In 14 cases, copy-number information for numerical aberrations was retrieved from methylation bead arrays.

Statistical analysis
Progression-free survival (PFS) and overall survival (OS) were defined as time from first tumor surgery to (a) first tumor progression or death of any cause (PFS) or (b) to death of any cause (OS). PFS and OS rates were estimated using the Kaplan-Meier method. Univariable survival rates were compared using the log-rank test. The most relevant risk factors were identified by a multivariable Cox regression analysis with variable selection independently for PFS and OS based on minimization of Akaike's information criterion. For variable selection, each combination of the 12 covariates (therapy protocol, clinical risk, histology, OTX2 status, TP53 copy number, TP53 mutation, chr6, chr10, chr13, chrX, WNT signaling activating mutation, and age at diagnosis) were considered (4096 different covariate combinations, including an intercept only model). Robustness of the results was investigated by repeatedly applying a subsampling approach independently for PFS and OS. Each of the 100 subsampling data sets consists of 80% of the original cohort whereby maintaining the ratio of events and non-events as in the complete cohort for PFS and OS, respectively. On each subsampling data set, Kaplan-Meier analyses were repeated and p value ranges for the log-rank test were obtained. Multivariable Cox regression analysis and variable selection procedure for the complete cohort were performed on each subsampling data set to investigate robustness of the derived model. The analyses have been performed in R version 4.1.1 with packages survival version 3.2-11 and survminer 0.4.9. All analyses were hypothesisgenerating. P values were adjusted figurewise by the Bonferroni-Holm (BH) method to account for multiple testing. A (corrected) p value less than 0.05 was considered to be statistically significant.

Cohort parameters
In our cohort of 191 WNT-MBs, 125 patients were children (age 3-15 years), 36 were adolescents (16-20 years), and 30 were adults (≥ 21 years). The median age was 13 years. A female predominance was found (ratio 1.4:1 (Table 1, Online  Resource Table 3). The majority of tumor samples had classic histology (n = 181; 94.8%); only nine cases presented with large cell/anaplastic (LC/A) histology (4.7%; Fig. 1, Fig. 2 a Distribution of the 176 CTNNB1 mutations in exon 3 at or adjacent to the 4 phosphorylation sites (purple color), including 4 short in-frame deletions. b Immunohistochemical staining of nuclear β-catenin accumulation in a WNT-MB (ID149) harboring two heterozygous APC mutations; normal cells with blue nuclei serve as internal control; scalebar = 20 µM. c, d Copy-number (upper) and allele ratio (lower) plots from molecular inversion probe array of a sibling pair (ID183 (c), ID154 (d)); the copy-neutral loss of heterozygosity within chromosome arm 5q in ID183 (c) is enlarged and the APC locus is highlighted by the gray vertical line; of note, ID154 (d) has additional mosaic losses of chromosomes 10, 12 and Y. e Distribution of APC mutations in 13 CTNNB1 wild-type WNT-MBs; the homozygous R414C is the only missense mutation and likely benign according to the ACMG/AMP classification rules, but this case (ID151) has a second homozygous stop mutation (R283*) and a copy-neutral loss of heterozygosity, so that the R414C mutation will not be translated to protein and has no tumorigenic impact anyway. One CTNNB1-mutant WNT-MB (ID122) has a heterozygous R2226* APC mutation (not shown), which is likely pathogenic. f, g Kaplan-Meier progression-free (PFS) and overall survival (OS) plots of patients with CTNNB1-mutant versus APC-mutant WNT-MB; the green line shows ID191 without CTNNB1 or APC mutation, but with a WNT-activating homozygous loss of FBXW7. The color-filled regions of the Kaplan-Meier plots indicate pointwise 95%-confidence intervals for the Kaplan-Meier estimate. The numbers at risk refer to the complete cohort. p(adj), adjusted p value after correction for multiple testing (Bonferroni-Holm)  Table 4). Among the 120 patients with clinical data, 32 (26.7%) were clinically high-risk (HR); of these, 16 patients because of metastatic disease and the remaining because of large postoperative residual tumor or LC/A histology (eight patients each). Eighty-four patients received "standard-risk" medulloblastoma therapy with 23.4 Gy CSI with a boost to the posterior fossa or tumor bed of 54 Gy followed by maintenance chemotherapy or hyperfractionated radiotherapy with a CSI-dose of 36 Gy and boosts to posterior fossa and tumor bed according to the HIT-SIOP-PNET4 trial [27]. The remaining patients received "high-risk" medulloblastoma therapies, usually with 35.2 Gy CSI followed by a boost to the posterior fossa and maintenance chemotherapy according to the HIT'91 trial in the maintenance chemotherapy arm [17] or according to the MET-HIT2000-AB4 regimen ( Table 1, Online Resource Table 5) [41].
Of the 15 CTNNB1-wild-type tumors, 13 carried APC mutations. Of the remaining two tumor samples, one (ID191) had a homozygous deletion of the FBXW7 gene located on chromosome 4q31.3 (Online Resource Fig. 2), whereas from the other sufficient DNA for APC sequencing was not available. However, this case qualified for the diagnosis of WNT-MB because of the presence of both nuclear β-catenin accumulation and monosomy 6. 450k/850k methylation array data was available in 94 cases, confirming the assignment to the WNT-MB group in 87 cases; in 7 cases, no matching score was reached (Fig. 1).
TP53 mutations were found in 16.1% (30/186 cases; Table 1). Twenty six were missense mutations located in exons 5-8; only one was a missense variant in exon 4. Additionally, we found one splice site mutation and one frameshift deletion (both exon 6), and one case had a focal homozygous deletion of the TP53 locus (chromosome 17p13.1; Fig. 4a, Online Resource Fig. 3). An accumulation of p53 protein in ≥ 5% of tumor cells was present in 20.4% of cases (38 of 186; Figs. 1, 4b). TP53 mutations were found in 18 of 38 p53-positive tumor samples. The TP53 locus showed loss of heterozygosity in 18.5% (24/130 cases), in some only present in a fraction of tumor cells (mosaicism). Of the 30 TP53-mutated cases, allele status was available from 27 samples, and 11 of these had a copy loss on chromosome 17p13.1. A significant predominance of male patients was found in cases with TP53 mutations (18 of 30 (60%), p = 0.020, Chi-square test). TP53 mutations occurred in all age groups (6-23 years) without a significant association with one of the age groups. TP53 mutant and TP53 wild-type cases did not cluster separately in UMAP analysis (Online Resource Fig. 4). LC/A histology was significantly enriched in TP53-mutated WNT-MBs (4 of 30 versus 5 of 156, p = 0.020, Chi-square test).
The E17K hotspot regions of AKT1-3 were Sanger sequenced in 165 cases, and the E17K mutation was found 6 times each in AKT1 and AKT3 (12 of 165; 7.27%).

APC mutations
Of 13 APC-mutant cases, 8 had larger regions with copyneutral loss of heterozygosity (cnLOH) spanning the APC locus up the telomeric region of chromosome 5q; one case had a focal loss at the APC locus (5q22.2). Eight of these nine cases with deletion carried a single mutation, and one carried a double mutation on the same allele (R283* and R414C) with allele variant fractions between 75 and 100%.

Occurrence of APC-mutant WNT-MB in siblings
Our WNT-MB cohort contained one sibling pair with a female patient, who developed a classic MB at the age of 11 with metastatic spread at diagnosis (ID183). Her brother was 19 when the tumor was detected, and histology was also classic, but without metastatic disease (ID154). Both were CTNNB1 wild-type, but APC sequencing revealed the same R213* mutation in both tumors, suggesting that it represented a germline mutation. The sister had an allele variant fraction of almost 100% and a large cnLOH on chromosome arm 5q, whereas the brother had a balanced APC locus, but an additional two base-pair frameshift deletion at amino acid L1129. Both APC mutations in this tumor had allele variant fractions of 40-50%.
Additionally, both siblings had no OTX2 gains, but were affected by TP53 alterations. The sister presented a monoallelic TP53 mutation (R282P) without loss of the TP53 locus, whereas the brother had a chromosome arm 17p loss, but wild-type TP53 (Fig. 2c, d). Both siblings are alive without relapse or secondary tumor.
This effect was retained in a multivariable Cox regression analysis for PFS and OS, where OTX2 gain (HR 5.98  (Fig. 5d-f). The model selection procedure applied to the 100 subsampling data sets revealed the model containing the three mentioned covariates to have the minimal average AIC as well as the minimal median rank when ranking models according to the AIC (PFS: average AIC = 65.88 OS: average AIC = 49.92; Fig. 5g and Online Resource Fig. 17). Fig. 4 a Distribution of TP53 mutations; colored mutations represent splice site mutation (blue), frameshift deletion (green), and nonsense mutation (red). b Immunohistochemical staining for p53 protein in a WNT-MB (ID184) with a heterozygous R248Q TP53 mutation and balanced TP53 locus; a strong nuclear staining pattern is visible focally, whereas some regions lack p53 staining (left side of the image); scalebar = 50 µM. c, d Kaplan-Meier progression-free (PFS) and overall survival (OS) plots of patients with TP53 mutant versus TP53 wild-type WNT-MB. e Summary plot of copy-number aberrations (n = 115 WNT-MBs) from molecular inversion probe array (n = 108) and SNP6 array (n = 7). The arrow indicates the OTX2 locus on chromosome arm 14q. Blue bars, gains; red bars, losses. Thickness of bars indicates frequency of alterations. f Copy-number plots from molecular inversion probe arrays of a WNT-MB case with focal OTX2 gain (upper part) and balanced OTX2 (lower part) on chromosome arm 14q. g, h Kaplan-Meier PFS and OS plots of patients with OTX2 gain versus no OTX2 gain ◂

Discussion
Several ongoing clinical trials aim to improve quality of life for patients with WNT-MB by de-escalation of therapy, especially reduced radiation dosage. However, despite the excellent overall prognosis (of WNT-MB patients aged 15 years or younger), patients who relapse might not survive their disease. To prevent these patients from relapse, their upfront identification by suitable prognostic tumor markers is required. Currently, older age is considered the main risk factor for non-metastatic WNT-MB, but data of adolescent and adult WNT-MBs are scarce and results are not fully consistent [5,7,14,15,24,25,32,36,39,44,45].

Limitation of clinical parameters as risk predictors
Data on survival of adult patients (≥ 21 years) in our cohort are sparse, but our data confirmed the worse outcome for adolescents (16-20 years). Conventional clinical parameters used for stratification into standard or high risk were not suitable as predictive markers in our WNT-MB cohort (Fig. 3c, d). Although data for WNT-MB with clinical highrisk features are rare, this observation is in line with published data [32,41].

APC-mutant MB shows similar outcomes as CTNNB1-mutated WNT-MBs
WNT-MBs with pathway-activating loss-of-function mutations in the APC gene had similar PFS/OS compared to CTNNB1-mutated cases, thus validating previous publications [40,43]. This indicates that patients with APC-mutant WNT-MB might be included into therapy de-escalating clinical trials equally to CTNNB1-mutant WNT-MB in the absence of other risk factors.
In contrast to activating CTNNB1 mutations, which almost always occur as somatic events in the tumor, pathogenic APC mutations frequently are germline events. These cause the cancer prediposition syndrome FAP, which is associated with a very high lifetime risk for colorectal cancer. As a consequence, patients with an APC-related MB should be offered genetic counseling and screened for APC germline mutations with the aim to offer surveillance for colorectal cancer [1].

TP53 mutation correlates to increased risk of recurrent disease
We identified TP53 mutations as a robust marker for increased risk of relapse in WNT-MBs. This finding corresponds to the recent description of an over-representation of TP53 mutant tumors in relapsed WNT-MB leading to the hypothesis of a prognostic impact of TP53 mutations [37]. Zhukova et al. did not find an impact of TP53 mutations on OS, but might have missed their prognostic relevance for PFS, as this was not analyzed [46]. In line with this publication, we also could not find a significant correlation of TP53 mutations to worse OS. However, multivariable analysis also showed a prognostic impact of TP53 mutations on OS in our cohort, although this finding was less significant than for PFS and has to be validated in future trial cohorts.
TP53 loss also indicated significantly worse PFS, which was expected as loss at the TP53 locus was highly associated with TP53 mutations.

Focal gain of the master regulator OTX2 indicates poor prognosis
The master regulator gene OTX2 encodes a transcription factor essential for the normal development of CNS tissues like brain, cerebellum, pineal gland, and eye [3]. Both in the developing cerebellum and cerebellar medulloblastoma, OTX2 is highly expressed, and has been shown to possess oncogenic activity in non-WNT/non-SHH medulloblastomas [2]. Its obvious biological role in WNT-activated medulloblastoma has not been studied in detail so far. In this study, gains of OTX2 were identified as negative and independent prognosticator. Indeed, all patients with tumors lacking a TP53 mutation who died or had tumor relapse harbored an OTX2 gain. Whereas IHC positivity of OTX2 protein is visible in almost all WNT-and non-WNT/non-SHH-MB, these mostly focal gains can only be found in a subset of WNTactivated tumors and similarly in non-WNT/non-SHH-MBs, but not in SHH-MBs. OTX2 has been shown to be highly repressed in SHH-MB [28] and copy-number gains of OTX2 have not been described in large series of SHH-MB [14, Fig. 5 a, b Kaplan-Meier progression-free (PFS) and overall survival (OS) plots of patients with WNT-MBs without TP53 mutation and OTX2 gain (blue line), one of these two alterations (red line), or both alterations (green line  15,39]. In standard-risk non-WNT/non-SHH-MBs, OTX2 gain was not associated with poor outcome [14]. As these gains are mostly focal, high-resolution technologies, such as MIP or gene-specific probes, should be used to detect this alteration. Of note, the copy-number information deduced from widely used DNA methylation might not be suitable to detect these OTX2 gains.

Multivariable analysis and proposal of a novel risk prediction model
In multivariable analysis, three genomic alterations were identified as independent indicators of outcome both PFS and OS, namely TP53 mutation, OTX2 gain, and monosomy 6. All relapsed patients could be identified by the first two alterations, so we suggest a risk prediction model based on these two markers. In this risk model including TP53 mutation and/or OTX2 gain, 58.1% of patients were identified having an increased risk for relapse of their disease after therapy according to the standard before the era of dosereducing trials in WNT-MB. In this model, the remaining 41.9% of patients had WNT-MB without these genetic alterations and represented low-risk patients; no relapse occurred in these 31 patients (Fig. 5a-c). As our cohort of WNT-MB patients included patients with high-risk clinicopathological features, we also performed explorative subgroup analyses on "pure standard-risk" cohorts with patients who would had qualified for inclusion into the SIOP-PNET-5 MB (inclusion up to 16 years of age at diagnosis; n = 50) or into the SJMB12 and ACNS1422 studies (up to 22 years of age; n = 64). Of note, OTX2 gain remained prognostic in the latter standard-risk cohort, although patient numbers are substantially reduced (Online Resource Fig. 18). As the nature of the study is retrospective and the patient cohort described here was not homogeneously treated in the framework of a single controlled trial, our proposed risk model needs validation in current or future therapeutic trials of WNT-MB patients to investigate the predictive power of the model. Upon confirmation, these genetic markers might allow to more precisely identify low-risk patients eligible for dose-sparing therapeutic strategies, while for the remaining patients, dose de-escalation should be avoided, but approaches aiming to increase survival might become of relevance.