Abstract
International consensus and the 2021 WHO classification recognize eight molecular subgroups among non-WNT/non-SHH (Group 3/4) medulloblastoma, representing approximately 60% of tumors. However, very few clinical centers worldwide possess the technical capabilities to determine DNA methylation profiles or other molecular parameters of high risk for group 3/4 tumors. As a result, biomarker-driven risk stratification and therapy assignment constitutes a major challenge in medulloblastoma research. Here, we identify an IHC marker as a clinically tractable method for improved medulloblastoma risk stratification.
We bioinformatically analyzed published medulloblastoma transcriptomes and proteomes identifying as a potential biomarker TPD52, whose IHC prognostic value was validated across three group 3/4 medulloblastoma clinical cohorts (n = 387) treated with conventional therapies.
TPD52 IHC positivity represented a significant independent predictor of early relapse and death for group 3/4 medulloblastoma [HRs between 3.67 and 26.7; 95% confidence interval (CI) between 1.00 and 706.23; P = 0.05, 0.017, and 0.0058]. Cross-validated survival models incorporating TPD52 IHC with clinical features outperformed existing state-of-the-art risk stratification schemes, and reclassified approximately 50% of patients into more appropriate risk categories. Finally, TPD52 immunopositivity was a predictive indicator of poor response to chemotherapy [HR, 12.66; 95% CI, 3.53–45.40; P < 0.0001], suggesting important implication for therapeutic choices.
This study redefines the approach to risk stratification in group 3/4 medulloblastoma in global practice. Because integration of TPD52 IHC in classification algorithms significantly improved outcome prediction, this test could be rapidly adopted for risk stratification on a global scale, independently of advanced but technically challenging molecular profiling techniques.
International consensus and the 2021 WHO classification recognize eight molecular subgroups among non-WNT/non-SHH medulloblastoma (encompassing group 3 and 4 tumors and representing approximately 60% of medulloblastoma cases). However, only very few clinical centers worldwide possess the technical capabilities to determine DNA methylation patterns or other molecular parameters of high risk for group 3/4 tumors. To our knowledge, no IHC-based assay is available for risk stratification of group 3/4 medulloblastoma. This study refines the approach to risk stratification in group 3/4 medulloblastoma. Reclassification of approximately 50% of group 3/4 tumors using TPD52 IHC will favor the choice of more appropriate therapeutic strategies. For example, identification of low-risk tumors allows reclassification of approximately 4% of children, who could benefit from less aggressive therapeutic approaches. Identification of a very high-risk group allows reclassification of approximately 15% of patients, who might instead benefit from therapy intensification. Finally, the observation that TPD52-based classification possesses predictive value for response to therapy could allow improved biomarker-driven therapy selection.
Introduction
Medulloblastoma, the most common malignant pediatric brain tumor and leading cause of cancer-related deaths, is recognized to comprise multiple molecular pathology disease entities, characterized by different demographics, molecular profiles, and clinical behavior (1–4). The current clinical staging system for medulloblastoma discriminates between low, standard, and high-risk tumors based on a combination of clinical, histologic, IHC, and genetic markers (5). Risk stratification for WNT and SHH tumors is relatively straightforward, as they can be routinely identified based on immunoreactivity for β-catenin (6) and Gab1 (7), respectively. On the other hand, while non-WNT/non-SHH (Group 3 and Group 4) tumors can be identified collectively based on lack of immunoreactivity for Gab1 and Yap1 (5, 7), achieving precise risk stratification for this large group of patients represents an important ongoing clinical challenge.
A large body of emerging literature indicates the presence of significant heterogeneity among Group 3/4 tumors, which include eight subgroups differing both from molecular and clinical standpoints (1, 8–10). While a subset of these tumors (mostly Group 4) is characterized by a favorable clinical course, other tumors are associated with poor patient outcome (such as Group 3 tumors harboring 8q gain with MYC genetic amplification). Historically, the identification of high-risk medulloblastoma has relied on young age (<3 years; ref. 11), the presence of metastatic disease (9), large cell anaplastic histology (LCA; refs. 5, 12, 13), and subtotal surgical resection (STR; ref. 14). However, recent studies questioned the role of LCA and STR (15) in Group 3/4 tumors, and their prognostic role remains unclear (16). Although early studies explored potential biomarkers for predictive utility (17), there is currently no IHC-based assay routinely available to clinicians for risk stratification. Additional powerful very high-risk parameters, such as MYC genetic amplification (18), chromosomal abnormalities (ref. 19; mainly 6q and 17q) or subgroup affiliation (10), are often difficult to delineate. This is largely due to complexity, lack of availability, and high cost of required diagnostic modalities such DNA methylation profiling or other advanced techniques, limiting their utility in pathology laboratories worldwide. As a result, although molecular subgroups are included in the latest WHO classification of brain tumors (5), their determination is frequently not feasible in North American and European practice, and is currently a major obstacle on a worldwide scale.
In addition to genetic divergence, medulloblastoma subgroups differ by underlying biological pathways, which could potentially lead to the emergence of effective subgroup specific therapies (20). The development of biomarker-driven treatment strategies for Group 3/4 patients, using tools available in a routine pathology setting, represents a significant clinical unmet need in the approach to medulloblastoma, as well as in clinical trial design. Rapid and reliable identification of very high-risk tumors would allow oncologists to assign patients to appropriate aggressive therapeutic protocols, sparing adverse effects of high-dose radio- and chemotherapy to patients with standard or low-risk medulloblastoma. The objective of this study is to identify a pathologic IHC marker for improved risk stratification of Group 3/4 medulloblastoma, for ready application in clinical practice. Herein, we describe the development of clinically tractable risk stratification models for Group 3/4 medulloblastoma based on IHC for TPD52, a cytoplasmic protein whose enhanced expression and prognostic significance were previously reported in breast (21), prostate (22), and hematologic malignancies (23, 24). While no literature has explored the role of TPD52 in medulloblastoma, some studies suggest that TPD52 is involved in regulating PI3K–AKT signaling (25) and cellular secretion by binding to Annexin VI (24), thereby functioning as a prosurvival factor. The stratification models based on TPD52 IHC proposed in our study allow assignment of patients into clinically informative biomarker-defined risk groups.
Materials and Methods
Analysis of proteomics and transcriptomics databases
Normalized proteomics data were accessed from previous studies (3, 4) and differential expression analysis between Group 3 and 4 tumors was performed (R, “limma” package, no transformation needed and default settings). We obtained mouse brain protein half-life values from previous studies (26). We obtained gene expression and survival data from the R2 platform (http://r2.amc.nl, GSE85217). For survival analysis of mRNA expression data, Group 3 tumors were split based on the median gene expression value.
BCCH patient cohort
The BCCH cohort consists of 72 medulloblastoma samples (including 39 Group 3/4 primary tumors) prospectively obtained, after written informed consent and Institutional Review Board (IRB) approval, between 1986 and 2012 from the BC Children's Hospital (BCCH, Vancouver, British Columbia, Canada) as previously described (27). Tumor blocks were used to build a tissue microarray (TMA), with each tumor represented in triplicate cores. Detailed information about this cohort, including methods for subgroup assignment, was previously published (27). The only inclusion criteria were age < 21 years old, histological diagnosis of medulloblastoma, and Group 3 or 4 affiliation (based on concomitant Yap1 and Gab1 negativity by IHC and confirmed by NanoString technology, as described previously (27)). Standard and high-risk stratification was achieved according to clinical parameters (age, M status and surgical resection; refs. 28, 29). No additional molecular analysis (i.e., MYC or MYCN amplification) was performed on this cohort at the time of diagnosis for risk stratification. All patients included in the survival analysis received standard-of-care treatment: after maximal safe surgical resection, CSI to the neuroaxis and the primary tumor site (in standard CS 23.4 Gy + PF 56 Gy or high doses CS 36–39.6 Gy + PF 54–56Gy based on clinical risk staging) followed by maintenance chemotherapy including vincristine, lomustine, cyclophosphamide, and cisplatin. A subset of children < 6 years old underwent radiation sparing high-dose chemotherapy protocols (chemotherapy-only group). Four patients who received nonstandard therapy (only radiotherapy) were excluded from the survival analysis, leaving 35 patients included in the cohort.
CHOP patient cohort
The CHOP cohort consists of 54 medulloblastoma samples (including 35 Group 3/4 primary tumors) prospectively collected, after written informed consent and IRB approval, between 2005 and 2014 from the Children's Hospital of Philadelphia (CHOP, Philadelphia, PA). Tumor blocks were used to build a TMA, with each tumor represented in duplicate cores. The only inclusion criteria were age <21 years old, histologic diagnosis of medulloblastoma and Group 3/4 affiliation, based on concomitant Yap1 and Gab1 negativity by IHC (according to clinical triage protocols and as previously described; ref. 7). Standard and high-risk stratification was achieved according to clinical parameters (age, M status, and surgical resection; refs. 28, 29). No additional molecular analysis (i.e., MYC or MYCN amplification) was performed on this cohort at the time of diagnosis for risk stratification. All patients received standard-of-care treatment: after maximal safe surgical resection, CSI to the neuroaxis and the primary tumor site (in standard CS 23.4 Gy + PF 56 Gy or high doses CS 36–39.6 Gy + PF 54–56 Gy based on clinical risk staging) followed by maintenance chemotherapy including vincristine, lomustine cyclophosphamide, and cisplatin. A subset of children < 3 years old underwent radiation sparing chemotherapy regimens (chemotherapy-only group).
Burdenko–Heidelberg patient cohort
The Burdenko cohort consists of tumor blocks from 353 primary medulloblastoma samples, obtained after written informed consent, between 2000 and 2018 at the Burdenko Neurosurgical Institute (Moscow, Russian Federation), which were molecularly subgrouped at the German Cancer Research Center (Heidelberg, Germany), where whole tissue sections were evaluated. Approval to link molecular to clinical and pathological data was obtained from the local IRB. The only inclusion criteria were age < 21 and molecular diagnosis of medulloblastoma Group 3 or Group 4, which, together with additional molecular features, was determined by 450k/850k methylation profiling (Classifier v11b4 algorithm), as described previously (30). Group 3/4 affiliation for the present cohort was also determined using NanoString technology, as previously described (31). The BH cohort served as an independent external validation cohort: IHC was performed at a different institution and evaluation of staining was performed by an independent pathologist.
In addition, for this cohort, after maximal safe surgical resection, therapy was standardized according to protocols of the German HIT study group (32, 33), providing a unique opportunity to validate initial outcome prediction results in our study. Briefly:
(i) <4 years old received radiation sparing chemotherapy-only protocols: children received HIT-SKK chemotherapy consisting of intravenous cyclophosphamide, vincristine, methotrexate (followed by leucovorin rescue after 42 hours), carboplatin, and etoposide and concomitant intraventricular methotrexate (36 single doses; 2 mg/dose), for a total duration of 6 months, as described else-where (32)
(ii) ≥ 4 years old and M0/M1 (radiotherapy + chemotherapy) received HIT-based radio-chemotherapy protocol in standard doses: CSI with 23.4 Gy to the neuroaxis and 30.6 Gy to the primary tumor site followed with 6–8 cycles of maintenance chemotherapy including vincristine (1.5 mg/m2 days 1, 8, and 15), lomustine (75 mg/m2 day 1), and cisplatin (70 mg/m2 day 1), as described elsewhere (NCT02066220)
(iii) ≥ 4 years old and M2/M3 received therapy intensification protocols: two cycles of neoadiuvant induction HIT-SKK chemotherapy (including intraventricular methotrexate injection) + hyperfractioned radiotherapy + four cycles of maintenance combination chemotherapy including vincristine (1.5 mg/m2 days 1, 8, and 15), lomustine (75 mg/m2 day 1), and cisplatin (70 mg/m2 day 1), as described elsewhere (33)
Detailed information about this cohort, including RNA-sequencing data, DNA copy-number analysis from 450/850k methylation array and molecular analysis methods, has been previously published (30), and is presented in Supplementary Table S1. Thirty-six patients for whom the therapeutic scheme varied from the one described above were excluded from the survival analysis (i.e., a total of 317 patients were included in this cohort).
IHC
IHC for TPD52, Yap1 and Gab1 was performed on formalin-fixed paraffin-embedded (FFPE) sections using standard protocols. Briefly, sections were deparaffinized and rehydrated before being incubated in Tris EDTA buffer (CC1 standard) at 95°C for 1 hour to retrieve antigenicity. Tissue sections were then incubated with primary antibody (TPD52 Abcam ab182578 1:200, Yap1 Santa Cruz Biotechnology, sc-101199 1:50, 1:250 or Gab1 Abcam ab27439 1:100) for 1 hour (Ventana Discovery platform). Tissue sections with bound primary antibody were then incubated with the appropriate secondary antibody (Jackson antibodies at 1:500 dilution), followed by Ultramap HRP and Chromomap DAB detection. Yap1 IHC was evaluated according to a binary classification scheme (i.e., positive vs. negative staining, as described; ref. 7). As TPD52 staining intensity was equivalent across different tumors, the staining was evaluated in a semiquantitative manner (i.e., estimation of percentage of positive cells per TMA core or slide). To address interobserver variability, evaluation of the immunostaining was performed by two investigators blinded to clinical variables (A. Delaidelli and P.H. Sorensen). For the BH cohort (whereby whole tissue sections were analyzed to allow for external validation), evaluation of the staining was performed by a senior neuropathologist (A. Korshunov) who was blinded to clinical variables. To address potential interobserver variability, a subset of slides was evaluated also by a second investigator. Only slight discrepancies in the estimation of percentages of positive cells were observed (i.e., 65% vs. 70%), and were readily were addressed by reaching a consensus. All the studies described adhered to the principles of the Declaration of Helsinki.
Statistical analysis
Statistical analyses were performed as follows: P values resulting from differential expression analysis between Group 3 and 4 tumors were adjusted for multiple testing according to Benjamini–Hochberg false discovery rate (FDR) method. Correlation analysis was performed according to Spearman Rank correlation coefficients. Differences in survival between groups were calculated with log-rank tests for univariate survival analysis. Associations and differences between groups for categorical variables were calculated with the Fisher exact test and χ2 test, while for noncategorical variables Mann–Whitney U tests were used. The extent of surgical resection was defined on the basis of postoperative MRI (≤1.5 cm2 or > 1.5 cm2 for GTR or STR, respectively). Progression-free survival (PFS) was defined as the time from the first tumor surgery to progression, relapse, or death. Overall survival (OS) was defined as the time from the first tumor surgery to death from any cause. Both were censored at last follow-up for patients without an event. Cox proportional hazard regression models were used to test the independent association of variables with patient outcome. The proportionality assumption for Cox modeling was validated using scaled Schoenfeld residuals. The global performance of the models with and without TPD52 IHC was compared by means of the concordance statistic (C-statistic). To evaluate the prediction accuracy of multivariate regression models with or without TPD52 staining, prediction errors over time were calculated using the loss function approach (34). To overcome the pitfalls of overfitting of the training data, we evaluated model accuracies also in an independent external dataset. To achieve this, the prediction error was calculated in the BCCH and CHOP cohorts with multivariate models fitted on the BH dataset, as described in the Results section. Comparison of prediction error and C-index for different Cox models was performed by comparing the underlying regression models with robust likelihood ratio tests. The ability of Cox models to classify risk was also assessed by computing the area under the time-dependent receiver operating characteristic (ROC) curves, calculated according to the Nearest Neighbor Estimation (NNE) method. Span for NNE was set at 0.25 * (cohort size) (-0.20). ROC curves were computed every 18 months of follow-up time up to 10.5 years, and the resulting areas under the curve were compared by paired t test. For risk stratification, a multivariate Cox regression model based on clinical features (sex, age, M stage, resection, therapy) and TPD52 percentage positivity was fitted to the BH cohort (training cohort) and used as a classifier to predict survival probabilities for patients in all cohorts at 10 years from surgery. Risk categories were defined as follows: low risk, 10-year survival probability ≥ 0.9; standard-risk, 10-year survival probability ≥ 0.75 and < 0.9; high-risk, 10-year survival probability ≥ 0.5 and 0.75; very high-risk, 10-year survival probability < 0.5. To determine whether TPD52 positivity is predictive of response to specific therapeutic approaches, we performed Cox proportional hazard regressions after stratifying the BH cohort into either TPD52-positive or -negative tumors. Results were confirmed by post hoc Cox proportional hazard regression with interaction between covariates therapy and TPD52 positivity for the entire BH cohort, which was used to calculate the HR for therapy at different values of TPD52 positivity. Given the similarity in terms of clinical characteristics and therapeutic approach, for some analyses we combined the BCCH and CHOP cohorts into a unique cohort. All statistical tests were two-sided and the local significance level was set at 0.05. Statistical tests for IHC experiments were carried out to confirm previously generated hypotheses, so resulting P values are not to be considered exploratory (i.e., no adjustment for multiple testing needed). All statistical analysis was performed with R 3.5.1, with packages “survival,' “survminer,” and “maxstat” for univariate and multivariate survival analyses, “pec” and “survivalROC” for prediction error and ROC curves, and “regplot' for nomograms.
Results
Analysis of proteomics and transcriptomics data identifies TPD52 as an IHC marker for aggressive medulloblastoma
Recent studies have highlighted prominent differences in the proteomes of medulloblastoma groups (3, 4). We set out to take advantage of this knowledge to initially identify a subset of proteins upregulated in the most aggressive subset of Group 3 tumors. Among these, we identified TPD52 as being differentially expressed in Group 3 as opposed to Group 4 medulloblastoma (Padj < 0.0001 and Padj = 0.0020 in the Forget and Archer datasets, respectively; Fig. 1A; Supplementary Table S2). While the expression of other potential candidates, including FKBP5, MAK16, RRP15, and PALMD, failed to display links to patient outcome (Supplementary Fig. S1A–S1D), higher TPD52 mRNA expression was strongly predictive of poor overall survival within Group 3 medulloblastoma (P = 0.011, Fig. 1B), highlighting its link to the most aggressive subset of tumors. TPD52 was also particularly stable at the genome-wide level in mouse brain (69.0 percentile, Supplementary Fig. S1E), making it a promising candidate for IHC. Finally, the commercial availability of mAbs makes TPD52 an excellent candidate for IHC in routine clinical practice. We therefore focused our subsequent work on TPD52 and performed preliminary studies to optimize IHC protocols on FFPE medulloblastoma sections (Supplementary Fig. S2, online).
Demographics of the group 3/4 medulloblastoma cohorts
We next tested the hypothesis that TDP52 can serve as an IHC marker of adverse clinical outcome in Group 3/4 medulloblastoma within three independent cohorts. The demographics of the Group 3/4 patients in the three cohorts were similar to the published literature (10) and are summarized in Table 1. Median age was 5.9 (IQR 4.1–9.4), 9.3 (IQR 4.7–11.8) and 7.0 (IQR 5.0–10.0) years in the BCCH, CHOP and BH cohorts, with a male proportion of 71.4%, 71.4% and 71.3% respectively. Median follow-up time was 6.26 (IQR 3.3–7.9), 3.25 (IQR 1.2–4.9) and 3.67 (IQR 2.0–7.2) years in the BCCH, CHOP and BH cohorts, with 5-year OS 76.1% (95% CI, 62.9–92.1), 70.4% (95% CI 55.7–89.0) and 73.3% (95% CI, 68.1–78.9) respectively. Additional details about the cohorts are presented in Table 1.
. | . | BCCH . | CHOP . | BH . |
---|---|---|---|---|
N | 35 | 35 | 317 | |
Age at diagnosis (median (IQR)) | 5.9 (4.1–9.4) | 9.3 (4.7–11.8) | 7.0 (5.0–10.0) | |
Sex (%) | F | 10 (28.6) | 10 (28.6) | 91 (28.7) |
M | 25 (71.4) | 25 (71.4) | 226 (71.3) | |
Follow up time (years, median (IQR)) | 6.26 (3.3–7.9) | 3.25 (1.2–4.9) | 3.67 (2.0–7.2) | |
5-year overall survival (%, 95% CI) | 76.1 (62.8–92.1) | 70.4 (55.7–89.0) | 73.3 (68.1–78.9) | |
Group (%) | Group3 | 12 (34.3) | 117 (36.9) | |
Group4 | 23 (65.7) | 160 (50.5) | ||
Histology (%) | Classic | 30 (85.7) | 258 (81.4) | |
LCA | 5 (14.3) | 59 (18.6) | ||
Therapy (%) | only chemotherapy | 6 (17.1) | 5 (14.3) | 24 (7.6) |
RT + chemotherapy | 29 (82.9) | 30 (85.7) | 180 (56.8) | |
Intensification | 0 | 0 | 113 (35.6) | |
Metastatic status (%) | M0 | 15 (42.9) | 20 (57.1) | 160 (50.5) |
M1 | 4 (11.4) | 2 (5.7) | 30 (9.5) | |
M2 | 4 (11.4) | 4 (11.4) | 17 (5.4) | |
M3 | 12 (34.3) | 9 (25.7) | 110 (34.7) | |
Resection (%) | GTR | 25 (71.4) | 30 (85.7) | 177 (55.8) |
STR | 10 (28.6) | 5 (14.3) | 140 (44.2) | |
MYC amplification (%) | No | 295 (93.1) | ||
Yes | 22 (6.9) | |||
MYCN amplification (%) | No | 292 (92.1) | ||
Yes | 25 (7.9) | |||
i17q (%) | No | 135 (42.6) | ||
Yes | 182 (57.4) | |||
450/850k subgroup (%) | I | 7 (2.2) | ||
II | 49 (15.5) | |||
III | 28 (8.8) | |||
IV | 23 (7.3) | |||
V | 31 (9.8) | |||
VI | 41 (12.9) | |||
VII | 60 (18.9) | |||
VIII | 78 (24.6) |
. | . | BCCH . | CHOP . | BH . |
---|---|---|---|---|
N | 35 | 35 | 317 | |
Age at diagnosis (median (IQR)) | 5.9 (4.1–9.4) | 9.3 (4.7–11.8) | 7.0 (5.0–10.0) | |
Sex (%) | F | 10 (28.6) | 10 (28.6) | 91 (28.7) |
M | 25 (71.4) | 25 (71.4) | 226 (71.3) | |
Follow up time (years, median (IQR)) | 6.26 (3.3–7.9) | 3.25 (1.2–4.9) | 3.67 (2.0–7.2) | |
5-year overall survival (%, 95% CI) | 76.1 (62.8–92.1) | 70.4 (55.7–89.0) | 73.3 (68.1–78.9) | |
Group (%) | Group3 | 12 (34.3) | 117 (36.9) | |
Group4 | 23 (65.7) | 160 (50.5) | ||
Histology (%) | Classic | 30 (85.7) | 258 (81.4) | |
LCA | 5 (14.3) | 59 (18.6) | ||
Therapy (%) | only chemotherapy | 6 (17.1) | 5 (14.3) | 24 (7.6) |
RT + chemotherapy | 29 (82.9) | 30 (85.7) | 180 (56.8) | |
Intensification | 0 | 0 | 113 (35.6) | |
Metastatic status (%) | M0 | 15 (42.9) | 20 (57.1) | 160 (50.5) |
M1 | 4 (11.4) | 2 (5.7) | 30 (9.5) | |
M2 | 4 (11.4) | 4 (11.4) | 17 (5.4) | |
M3 | 12 (34.3) | 9 (25.7) | 110 (34.7) | |
Resection (%) | GTR | 25 (71.4) | 30 (85.7) | 177 (55.8) |
STR | 10 (28.6) | 5 (14.3) | 140 (44.2) | |
MYC amplification (%) | No | 295 (93.1) | ||
Yes | 22 (6.9) | |||
MYCN amplification (%) | No | 292 (92.1) | ||
Yes | 25 (7.9) | |||
i17q (%) | No | 135 (42.6) | ||
Yes | 182 (57.4) | |||
450/850k subgroup (%) | I | 7 (2.2) | ||
II | 49 (15.5) | |||
III | 28 (8.8) | |||
IV | 23 (7.3) | |||
V | 31 (9.8) | |||
VI | 41 (12.9) | |||
VII | 60 (18.9) | |||
VIII | 78 (24.6) |
Note: Text in italics indicates the different classes for each variable presented in the table.
TPD52 immunopositivity is an independent predictor of adverse outcome for group 3/4 medulloblastoma
TPD52 IHC positivity was not significantly associated with clinical variables in the BCCH and CHOP cohorts (age, sex and metastatic spread at diagnosis, Supplementary Table S3, online). In both cohorts we found significant association of TPD52 positivity with chemotherapy only postsurgical therapy (P = 0.026 and P = 0.043 respectively), so we accounted for this variable in subsequent multivariate analyses. Thanks to the large size of the BH cohort, we were able to analyze the distribution of the percentage of positive cells across tumors. This analysis revealed the existence of three broad categories of tumors: TPD52 negative (no positive cells, 54.9% of tumors), TPD52 low (⇒ 20% and < 50% positive cells, 34.1% of tumors), or TPD52 high (⇒ 50% positive cells, 11.0% of tumors, see Supplementary Fig. S3A and Supplementary Table S3, online). We observed perfect interobserver agreement for this categorization (κ = 1), and we did not observe differences in terms of staining intensity across tumors. No tumors displayed very low positivity (1%–19% positive cells), and only 2 out of 353 tumors were categorized by both pathologists as being 50% positive (Supplementary Fig. S3A). The categorization of tumors as described above was more powerful than a binary (positive vs. negative) classification and corresponded to the optimal cut-off for univariate survival analysis, as identified by maximally selected rank statistics analysis (Supplementary Fig. S3D). RNA-seq data analysis revealed a significant link between TPD52 IHC positivity on tumor sections and high mRNA expression (TPM > 250, P < 0.0001, Supplementary Fig. S3B and S3CA, online), confirming the high specificity of TPD52 IHC. However, RNA-seq was unable to discriminate between TPD52 low and TPD52 high tumors (Supplementary Fig. S3B, online), highlighting the fact that RNA expression represents a limited predictor of protein level for TPD52. In the BH cohort, TPD52 high tumors (⇒ 50%) were enriched in younger children (median 5.0 years [IQR 4.0–8.5]) as opposed to TPD52 low (<50%) and negative tumors (median 8.0 [IQR 5.0–11] and 7.0 [IQR 5.0–10.0] years respectively, P = 0.029, Supplementary Table S3, online). While we found no association with sex, therapy and metastatic status, TPD52 positivity overlapped significantly with LCA histology (P = 0.023) and, from the molecular standpoint, the high-risk parameters Group 3 / Subgroup II affiliation (P < 0.0001), MYC amplification (P < 0.0001) and i17q (P = 0.019, Fig. 1C and D and Supplementary Table S3, online). Despite the significant association of TPD52 IHC with Group 3 tumors, the sensitivity and specificity values for this test were only 47% and 56% respectively.
Survival analysis by log-rank test and Kaplan-Meier estimator revealed that TPD52 immunopositivity is significantly associated with worse OS in all of the cohorts: BCCH (Fig. 2A, P < 0.0001), CHOP (Fig. 2B, P = 0.035), combined BCCH-CHOP (Fig. 2C, P < 0.0001) and BH (Fig. 2D, P < 0.0001). In addition, TPD52 immunopositivity was associated with worse PFS in the two cohorts for which these data were collected comprehensively: BCCH (Fig. 2E, P = 0.0035) and BH (Fig. 2F, P = 0.0019). Proposed risk factors for high-risk medulloblastoma include LCA histology (5), metastatic spread at diagnosis (16), STR (14) and Group 3 affiliation (2). We therefore set out to determine the prognostic value of TPD52 immunoreactivity relative to these variables, computing Cox proportional hazard regression models for the three cohorts. After accounting for clinical variables, surgical resection, metastatic spread, therapy and histology, TPD52 immunopositivity retained its significant prognostic value for adverse overall outcome in the BCCH (HR 21.43 [95% CI, 1.00–461.60], P = 0.050 for 100% positive cells, Table 2) and CHOP cohorts (HR 26.70 [95% CI, 1.01–706.23], P = 0.049 Table 2). In order to further test significance, we combined these two cohorts into a unique Cox proportional hazard regression model, confirming that TPD52 immunopositivity predicted adverse outcome within Group 3/4 tumors, irrespective of clinical variables, surgical resection, metastatic spread, and therapy (HR 45.22 [95% CI 6.41–319.00], P = 0.0001, Table 2). Multivariate analysis for the BH cohort confirmed that TPD52 immunopositivity was a predictor of poor OS (HR 3.67 [95% CI, 1.46–9.24], P = 0.0058 for 100% positive cells, so i.e., ∼1.84 for 50% positive) independently of clinical and molecular variables (group, MYC and MYCN amplification, i17q, Table 2), and within Group 3 tumors (Fig. 3A, P < 0.0001). Similarly, TPD52 immunopositivity was an independent predictor of poor PFS in the same cohort (HR 2.18 [95% CI, 1.02–4.65], P = 0.044, Table 2), and within Group 3 tumors (P = 0.0002, Fig. 3B). In direct comparison with established medulloblastoma prognostic markers for Group 3/4 tumors, in the BH cohort TPD52 appeared to have a predictive power stronger than MYCN status, Group 3 affiliation and i17q, and inferior to MYC status (see HRs in Table 2).
. | . | HR (95% CI) . | P . | . | |
---|---|---|---|---|---|
BCCH cohort (OS) | |||||
Age at diagnosis | 1.09 (0.88–1.34) | 0.44 | |||
Sex (male) | 0.39 (0.06–2.60) | 0.33 | |||
TPD52 immunopositivity (100%) | 21.43 (1.00–461.60) | 0.050 | |||
LCA histology | 2.71 (0.33–22.23) | 0.35 | |||
Metastasis status | |||||
M1 | 0.56 (0.01–25.14) | 0.76 | |||
M2/3 | 2.97 (0.45–19.45) | 0.25 | |||
STR resection | 2.52 (0.48–13.35) | 0.27 | |||
Therapy | |||||
Only chemotherapy | 4.75 (0.32–71.03) | 0.25 | |||
CHOP cohort (OS) | |||||
Age at diagnosis | 1.37 (1.04–1.81) | 0.024 | P < 0.05 | ||
Sex (male) | 0.12 (0.01–1.32) | 0.08 | |||
TPD52 immunopositivity (100%) | 26.70 (1.01–706.23) | 0.049 | P < 0.05 | ||
Metastasis status | |||||
M1 | 2.55 (0.14–45.0)1 | 0.52 | |||
M2/3 | 8.87 (0.88–89.23) | 0.06 | |||
STR resection | 0.00 (0.00–Inf) | 0.99 | |||
Therapy | |||||
Only chemotherapy | 33.95 (1.83–628.21) | 0.017 | P < 0.05 | ||
Combined BCCH-CHOP (OS) | |||||
Age at diagnosis | 1.14 (0.99–1.31) | 0.07 | |||
Sex (male) | 0.57 (0.19–1.75) | 0.32 | |||
TPD52 immunopositivity (100%) | 45.22 (6.41–319.00) | 0.0001 | P < 0.001 | ||
Metastasis status | |||||
M1 | 1.03 (0.14–7.38) | 0.97 | |||
M2/3 | 3.45 (1.13–10.55) | 0.029 | P < 0.05 | ||
STR resection | 0.68 (0.20–2.35) | 0.54 | |||
Therapy | |||||
Only chemotherapy | 5.43 (1.08–27.28) | 0.04 | P < 0.05 | ||
Cohort | |||||
BCCH | 1.03 (0.39–2.75) | 0.95 | |||
BH cohort (OS) | |||||
Age at diagnosis | 0.92 (0.85–1.00) | 0.044 | P < 0.05 | ||
Sex (male) | 1.31 (0.79–2.15) | 0.29 | |||
TPD52 immunopositivity (100%) | 3.67 (1.46–9.24) | 0.0058 | P < 0.01 | ||
Group | |||||
Group 3 | 0.68 (0.34–1.35) | 0.26 | |||
LCA histology | 0.77 (0.36–1.64) | 0.49 | |||
Metastasis status | |||||
M1 | 0.79 (0.27–2.30) | 0.66 | |||
M2 | 1.36 (0.27–6.70) | 0.70 | |||
M3 | 1.40 (0.40–4.94) | 0.59 | |||
STR resection | 0.84 (0.53–1.34) | 0.47 | |||
Therapy | |||||
Only chemotherapy | 5.34 (1.57–18.16) | 0.0072 | P < 0.01 | ||
Intensification | 2.13 (0.55–8.24) | 0.27 | |||
MYC amplification | Yes | 16.75 (6.52–43.01) | <0.0001 | P < 0.001 | |
MYCN amplification | Yes | 1.38 (0.63–3.00) | 0.41 | ||
i17q | Yes | 1.59 (0.96–2.64) | 0.07 | ||
BH cohort (PFS) | |||||
Age at diagnosis | 0.95 (0.89–1.01) | 0.10 | |||
Sex (male) | 1.17 (0.78–1.74) | 0.44 | |||
TPD52 immunopositivity (100%) | 2.18 (1.02–4.65) | 0.044 | P < 0.05 | ||
Group | |||||
Group 3 | 0.58 (0.33–1.01) | 0.06 | |||
LCA histology | 0.95 (0.53–1.70) | 0.86 | |||
Metastasis status | |||||
M1 | 0.93 (0.45–1.92) | 0.84 | |||
M2 | 2.26 (0.54–9.47) | 0.26 | P < 0.05 | ||
M3 | 1.97 (0.59–6.53) | 0.26 | P < 0.001 | ||
STR resection | 0.95 (0.66–1.36) | 0.76 | |||
Therapy | |||||
Only chemotherapy | 2.73 (0.87–8.50) | 0.08 | |||
Intensification | 0.99 (0.28–3.47) | 0.98 | |||
MYC amplification | yes | 14.71 (6.78–31.92) | <0.0001 | P < 0.001 | |
MYCN amplification | yes | 1.34 (0.72–2.49) | 0.36 | ||
i17q | yes | 1.83 (1.20–2.77) | 0.0045 | P < 0.01 | |
BH cohort (OS, with molecular subgroups) | |||||
Age at diagnosis | 0.93 (0.86–1.01) | 0.08 | |||
Sex (male) | 1.17 (0.71–1.93) | 0.54 | |||
TPD52 immunopositivity (100%) | 4.62 (1.99–10.76) | 0.0004 | P < 0.001 | ||
LCA histology | 1.91 (1.09–3.33) | 0.023 | P < 0.05 | ||
Metastasis status | |||||
M1 | 0.72 (0.25–2.11) | 0.55 | |||
M2 | 1.99 (0.36–11.12) | 0.43 | |||
M3 | 2.17 (0.54–8.68) | 0.27 | |||
STR resection | 0.82 (0.52–1.30) | 0.40 | |||
Therapy | |||||
only chemotherapy | 4.30 (1.31–14.08) | 0.016 | P < 0.05 | ||
intensification | 1.19 (0.27–5.22) | 0.81 | |||
Molecular subgroup | |||||
II | 3.35 (0.43–26.10) | 0.24 | |||
III | 2.75 (0.34–22.16) | 0.34 | |||
IV | 0.65 (0.06–6.57) | 0.71 | |||
V | 2.62 (0.32–21.13) | 0.36 | |||
VI | 3.04 (0.38–24.22) | 0.29 | |||
VII | 1.66 (0.20–13.40) | 0.63 | |||
VIII | 2.44 (0.31–19.02) | 0.39 | |||
BH cohort (PFS, with molecular subgroups) | |||||
Age at diagnosis | 0.95 (0.89–1.01) | 0.11 | |||
Sex (male) | 1.07 (0.72–1.59) | 0.75 | |||
TPD52 immunopositivity (100%) | 2.72 (1.34–5.54) | 0.0058 | P < 0.01 | ||
LCA histology | 1.79 (1.12–2.88) | 0.015 | P < 0.05 | ||
Metastasis status | |||||
M1 | 0.90 (0.44–1.86) | 0.78 | |||
M2 | 3.57 (0.77–16.54) | 0.10 | |||
M3 | 3.59 (0.97–13.29) | 0.06 | |||
STR resection | 0.92 (0.64–1.33) | 0.65 | |||
Therapy | |||||
Only chemotherapy | 2.65 (0.87–8.09) | 0.09 | |||
Intensification | 0.52 (0.13–2.05) | 0.35 | |||
Molecular subgroup | |||||
II | 2.26 (0.52–9.90) | 0.28 | |||
III | 1.52 (0.33–7.05) | 0.59 | |||
IV | 0.34 (0.06–2.00) | 0.23 | |||
V | 2.39 (0.54–10.61) | 0.25 | |||
VI | 1.87 (0.42–8.40) | 0.41 | |||
VII | 1.40 (0.32–6.21) | 0.65 | |||
VIII | 2.23 (0.52–9.62) | 0.28 |
. | . | HR (95% CI) . | P . | . | |
---|---|---|---|---|---|
BCCH cohort (OS) | |||||
Age at diagnosis | 1.09 (0.88–1.34) | 0.44 | |||
Sex (male) | 0.39 (0.06–2.60) | 0.33 | |||
TPD52 immunopositivity (100%) | 21.43 (1.00–461.60) | 0.050 | |||
LCA histology | 2.71 (0.33–22.23) | 0.35 | |||
Metastasis status | |||||
M1 | 0.56 (0.01–25.14) | 0.76 | |||
M2/3 | 2.97 (0.45–19.45) | 0.25 | |||
STR resection | 2.52 (0.48–13.35) | 0.27 | |||
Therapy | |||||
Only chemotherapy | 4.75 (0.32–71.03) | 0.25 | |||
CHOP cohort (OS) | |||||
Age at diagnosis | 1.37 (1.04–1.81) | 0.024 | P < 0.05 | ||
Sex (male) | 0.12 (0.01–1.32) | 0.08 | |||
TPD52 immunopositivity (100%) | 26.70 (1.01–706.23) | 0.049 | P < 0.05 | ||
Metastasis status | |||||
M1 | 2.55 (0.14–45.0)1 | 0.52 | |||
M2/3 | 8.87 (0.88–89.23) | 0.06 | |||
STR resection | 0.00 (0.00–Inf) | 0.99 | |||
Therapy | |||||
Only chemotherapy | 33.95 (1.83–628.21) | 0.017 | P < 0.05 | ||
Combined BCCH-CHOP (OS) | |||||
Age at diagnosis | 1.14 (0.99–1.31) | 0.07 | |||
Sex (male) | 0.57 (0.19–1.75) | 0.32 | |||
TPD52 immunopositivity (100%) | 45.22 (6.41–319.00) | 0.0001 | P < 0.001 | ||
Metastasis status | |||||
M1 | 1.03 (0.14–7.38) | 0.97 | |||
M2/3 | 3.45 (1.13–10.55) | 0.029 | P < 0.05 | ||
STR resection | 0.68 (0.20–2.35) | 0.54 | |||
Therapy | |||||
Only chemotherapy | 5.43 (1.08–27.28) | 0.04 | P < 0.05 | ||
Cohort | |||||
BCCH | 1.03 (0.39–2.75) | 0.95 | |||
BH cohort (OS) | |||||
Age at diagnosis | 0.92 (0.85–1.00) | 0.044 | P < 0.05 | ||
Sex (male) | 1.31 (0.79–2.15) | 0.29 | |||
TPD52 immunopositivity (100%) | 3.67 (1.46–9.24) | 0.0058 | P < 0.01 | ||
Group | |||||
Group 3 | 0.68 (0.34–1.35) | 0.26 | |||
LCA histology | 0.77 (0.36–1.64) | 0.49 | |||
Metastasis status | |||||
M1 | 0.79 (0.27–2.30) | 0.66 | |||
M2 | 1.36 (0.27–6.70) | 0.70 | |||
M3 | 1.40 (0.40–4.94) | 0.59 | |||
STR resection | 0.84 (0.53–1.34) | 0.47 | |||
Therapy | |||||
Only chemotherapy | 5.34 (1.57–18.16) | 0.0072 | P < 0.01 | ||
Intensification | 2.13 (0.55–8.24) | 0.27 | |||
MYC amplification | Yes | 16.75 (6.52–43.01) | <0.0001 | P < 0.001 | |
MYCN amplification | Yes | 1.38 (0.63–3.00) | 0.41 | ||
i17q | Yes | 1.59 (0.96–2.64) | 0.07 | ||
BH cohort (PFS) | |||||
Age at diagnosis | 0.95 (0.89–1.01) | 0.10 | |||
Sex (male) | 1.17 (0.78–1.74) | 0.44 | |||
TPD52 immunopositivity (100%) | 2.18 (1.02–4.65) | 0.044 | P < 0.05 | ||
Group | |||||
Group 3 | 0.58 (0.33–1.01) | 0.06 | |||
LCA histology | 0.95 (0.53–1.70) | 0.86 | |||
Metastasis status | |||||
M1 | 0.93 (0.45–1.92) | 0.84 | |||
M2 | 2.26 (0.54–9.47) | 0.26 | P < 0.05 | ||
M3 | 1.97 (0.59–6.53) | 0.26 | P < 0.001 | ||
STR resection | 0.95 (0.66–1.36) | 0.76 | |||
Therapy | |||||
Only chemotherapy | 2.73 (0.87–8.50) | 0.08 | |||
Intensification | 0.99 (0.28–3.47) | 0.98 | |||
MYC amplification | yes | 14.71 (6.78–31.92) | <0.0001 | P < 0.001 | |
MYCN amplification | yes | 1.34 (0.72–2.49) | 0.36 | ||
i17q | yes | 1.83 (1.20–2.77) | 0.0045 | P < 0.01 | |
BH cohort (OS, with molecular subgroups) | |||||
Age at diagnosis | 0.93 (0.86–1.01) | 0.08 | |||
Sex (male) | 1.17 (0.71–1.93) | 0.54 | |||
TPD52 immunopositivity (100%) | 4.62 (1.99–10.76) | 0.0004 | P < 0.001 | ||
LCA histology | 1.91 (1.09–3.33) | 0.023 | P < 0.05 | ||
Metastasis status | |||||
M1 | 0.72 (0.25–2.11) | 0.55 | |||
M2 | 1.99 (0.36–11.12) | 0.43 | |||
M3 | 2.17 (0.54–8.68) | 0.27 | |||
STR resection | 0.82 (0.52–1.30) | 0.40 | |||
Therapy | |||||
only chemotherapy | 4.30 (1.31–14.08) | 0.016 | P < 0.05 | ||
intensification | 1.19 (0.27–5.22) | 0.81 | |||
Molecular subgroup | |||||
II | 3.35 (0.43–26.10) | 0.24 | |||
III | 2.75 (0.34–22.16) | 0.34 | |||
IV | 0.65 (0.06–6.57) | 0.71 | |||
V | 2.62 (0.32–21.13) | 0.36 | |||
VI | 3.04 (0.38–24.22) | 0.29 | |||
VII | 1.66 (0.20–13.40) | 0.63 | |||
VIII | 2.44 (0.31–19.02) | 0.39 | |||
BH cohort (PFS, with molecular subgroups) | |||||
Age at diagnosis | 0.95 (0.89–1.01) | 0.11 | |||
Sex (male) | 1.07 (0.72–1.59) | 0.75 | |||
TPD52 immunopositivity (100%) | 2.72 (1.34–5.54) | 0.0058 | P < 0.01 | ||
LCA histology | 1.79 (1.12–2.88) | 0.015 | P < 0.05 | ||
Metastasis status | |||||
M1 | 0.90 (0.44–1.86) | 0.78 | |||
M2 | 3.57 (0.77–16.54) | 0.10 | |||
M3 | 3.59 (0.97–13.29) | 0.06 | |||
STR resection | 0.92 (0.64–1.33) | 0.65 | |||
Therapy | |||||
Only chemotherapy | 2.65 (0.87–8.09) | 0.09 | |||
Intensification | 0.52 (0.13–2.05) | 0.35 | |||
Molecular subgroup | |||||
II | 2.26 (0.52–9.90) | 0.28 | |||
III | 1.52 (0.33–7.05) | 0.59 | |||
IV | 0.34 (0.06–2.00) | 0.23 | |||
V | 2.39 (0.54–10.61) | 0.25 | |||
VI | 1.87 (0.42–8.40) | 0.41 | |||
VII | 1.40 (0.32–6.21) | 0.65 | |||
VIII | 2.23 (0.52–9.62) | 0.28 |
Note: Molecular variables and subgroups were included when available.
Even when multivariate analyses were performed accounting for DNA methylation-based molecular subgroups (I-VIII), TPD52 retained its prognostic value (P = 0.0004, Table 2), suggesting that TPD52 IHC identified the most aggressive tumors irrespective of their subgroup. Finally, since the therapeutic approach utilized is among the strongest determinant of prognosis, we tested the predictive power of TPD52 within groups of identically treated patients and with identical clinical risk (Supplementary Figs. S4 and S5, online). TPD52 IHC held its significant predictive potential independently of therapeutic approach and clinical risk, further highlighting its clinical utility. For unknown reasons, TPD52 IHC did not hold prognostic value in SHH tumors (data not shown). Taken together, these data indicate that TPD52 is a novel and powerful independent predictor of adverse outcome in Group 3/4 medulloblastoma.
TPD52 immunohistochemistry improves medulloblastoma outcome prediction
To evaluate if the inclusion of TPD52 IHC improves outcome prediction for Group 3/4 tumors, compared to current stratification models, we compared Cox proportional hazard regression models with TPD52 staining information to the same models without this information. As shown in Fig. 3C–J, inclusion of TPD52 immunoreactivity in current stratification models, whether based on clinical variables, clinical plus molecular variables (Group 3/4, MYC, MYCN amplification and i17q) or methylation-based molecular subgroups, significantly improves outcome prediction and reduces prediction error. Similar results were obtained when comparing fitness of the Cox models (Supplementary Table S4, online) and receiver operating characteristic curves for the models at different time points (Supplementary Fig. S6, online). Even when Cox proportional hazard regression models with and without TPD52 IHC were fitted to one cohort (BH) and applied to predict survival in a different dataset (the combined BCCH-CHOP cohort), models with TPD52 performed significantly better, highlighting that the predictive power of TPD52 is completely independent from a particular dataset (Supplementary Fig. S7, online). Collectively, these findings confirm that inclusion of TPD52 immunopositivity significantly improves current stratification models for Group 3/4 medulloblastoma.
Identification of novel risk categories
While medulloblastoma subgroup affiliation and risk stratification are achievable with the help of advanced molecular analysis techniques (i.e., methylation profiling), these are not available in the vast majority of pathology laboratories worldwide. Group/subgroup affiliation of Group 3/4 medulloblastoma is therefore extremely difficult to delineate, and risk status is routinely determined on the basis of clinical variables in global oncology practice. To further underscore the clinical utility of TPD52 IHC in current practice, we compared the standard risk classification based on clinical variables (28, 29) to an improved stratification algorithm with inclusion of TPD52 IHC (Fig. 4A–C). Classification after inclusion of TPD52 resulted in highly significant risk stratification in the BH cohort (P < 0.0001), as well as identification of two novel risk categories (low and very high, Fig. 4B). Importantly, ∼ 80% of newly identified low-risk tumors are classified as high-risk according to current clinical standards, therefore exposing children to the possibly unnecessarily toxic effects of therapy intensification or high-dose RT. In addition, 160/317 (50.5%) tumors would have been reclassified to a different risk category with inclusion of TPD52 in diagnostic algorithms (Fig. 4C; Supplementary Table S5, online). To further demonstrate the applicability of these results to multiple cohorts, we utilized the Cox proportional hazard regression model trained on the BH cohort as a classifier to predict risk categories in the combined BCCH-CHOP cohort (as an external validation dataset). This approach (Supplementary Fig. S7, online) produced results virtually identical to the ones presented in Fig. 4A–C, highlighting that the prediction algorithm with inclusion of TPD52 is applicable across multiple patient populations. The Cox multivariate regression models utilized for prediction are also presented as nomograms, diagrams that can be used in clinical practice for overall (Fig. 4D) or progression free survival prediction (Fig. 4E). An additional version of the nomograms, including molecular variables, is presented in Supplementary Fig. S7, online.
The development of biomarker-driven treatment strategies for Group 3/4 patients represents a major challenge in medulloblastoma research. We therefore set out to determine if TPD52 positivity could be predictive of response to specific therapeutic approaches. Multivariate regression analysis of the BH cohort indicated that, after accounting for clinical variables and risk stratification parameters, a radiation sparing chemotherapy approach was not an independent predictor of adverse outcome in TPD52 negative tumors (HR 1.8 [95% CI, 0.40–8.14], P = 0.44, Fig. 4F and Supplementary Table S6, online). In contrast, radiation sparing chemotherapy was highly predictive of adverse outcome among TPD52 positive tumors, independently of other clinical variables (HR 12.66 [95% CI, 3.53–45.40], P < 0.0001, Fig. 4G and Supplementary Table S6, online). In addition, the HR associated with chemotherapy only protocols (vs. RT + chemotherapy) was directly proportional to the percentage of TPD52 positive cells (Fig. 4H; Supplementary Table S7, online), further highlighting the predictive power of TPD52 IHC. While these results will need to be validated within the context of a randomized clinical trial, they are indicative of the notion that TPD52 negative Group 3/4 tumors could represent preferential candidates for radiation-sparing chemotherapy protocols.
Together, these results indicate that inclusion of TPD52 IHC in the clinical staging of Group 3/4 medulloblastoma could provide a significant advantage for the identification of low-risk and very high-risk patients, as well as for tumors likely to respond to chemotherapy-only protocols.
Discussion
Medulloblastoma is an heterogenous disease comprised of at least four different groups (2). Recent evidence indicates that several additional subgroups can be identified within Group 3/4 tumors, each of which distinguished by a different molecular background and clinical behavior (8–10). However, while DNA methylation-based subgroup affiliation has been included in the latest WHO classification of medulloblastoma, this or other advanced molecular techniques for risk stratification of Group 3/4 tumors are for the most part not available to clinicians. Here, bridging the gap between next generation molecular classification and clinical practice, we describe IHC for the TPD52 protein as an inexpensive and efficient method for the identification low and very high-risk Group 3/4 medulloblastoma. Unfortunately, no medulloblastoma clinical trial cohort was available for IHC studies. To offset this limitation, we validated our results across three independent non-overlapping cohorts, originating from both North America and Europe, and treated according to respective local therapeutic protocols.
Our analyses demonstrate that TPD52 IHC was a strong predictor of aggressive behaviour in Group 3/4 medulloblastoma in uni- and multivariate survival analyses. TPD52 immunopositivity was also predictive of poor response to radiation sparing chemotherapy-only protocols, underscoring the potential usefulness of TPD52 not only in risk stratification but also for therapy assignment, although this observation needs to be confirmed within the context of a randomized clinical trial. Since TPD52 IHC is highly prognostic within Group 3 tumors, but not Group 4, when molecular profiling techniques are available the most useful application of this test is likely to be within Group 3 medulloblastoma. The most accurate outcome prediction in the absence of molecular profiling techniques was achieved when integrating TPD52 IHC with clinical variables in multivariate analyses (Table 2, where TPD52 staining % positivity was included but not classified according to specific cut-offs into categories such as negative, low, or high). This suggests that in future cohorts, the optimal outcome prediction in the absence of molecular profiling techniques would be achieved by utilizing the nomograms presented in Fig. 4D and E, where TPD52 percentage positivity is estimated and reported as a continuous variable (i.e., 20%, 30% etc…) from FFPE whole tissue sections.
While beyond the scope of the current article, limitations of this study include an absence of biological characterization of potential TPD52 functions in brain tumors. Indeed, no literature so far has examined the role of TPD52 in medulloblastoma biology and progression, and very few reports have investigated the role of TPD52 in other tumor types. Several studies suggest that TPD52, a member of a highly conserved family of proteins comprising TPD52, TPD52L1, and TPD52LS, is involved in regulating the cellular secretion process by binding to Annexin VI in a Ca++-dependent manner (24), thereby functioning as a prosurvival factor in breast (21) and prostate (22) cancer, as well as in hematologic malignancies (23, 24). Additional research is needed to elucidate the potential role of TPD52 in the progression of medulloblastoma or other malignant diseases.
Our results indicate that integration of TPD52 IHC in risk stratification of medulloblastoma improves outcome prediction. This has important clinical implications, as IHC for TPD52 could be adopted in neuropathology laboratories worldwide for risk stratification of medulloblastoma, after exclusion of WNT or SHH group affiliation using available methods. Rapid risk stratification of Group 3/4 tumors will assist in assigning patients in the high-risk group to more aggressive therapeutic approaches, while sparing the adverse effects of high dose radio- and chemotherapy to children in low-risk groups. Future work should aim at validating the prognostic role of TPD52 IHC in prospective clinical trials.
Authors' Disclosures
A. von Deimling reports a patent for methylation bases brain tumor diagnostics pending. V. Ramaswamy reports personal fees from Astra Zeneca outside the submitted work. No disclosures were reported by the other authors.
Authors' Contributions
A. Delaidelli: Conceptualization, data curation, formal analysis, supervision, validation, investigation, methodology, writing–original draft, writing–review and editing. C. Dunham: Data curation. M. Santi: Resources. G.L. Negri: Methodology. J. Triscott: Data curation. O. Zheludkova: Data curation. A. Golanov: Data curation. M. Ryzhova: Data curation. K. Okonechnikov: Data curation. D. Schrimpf: Supervision. D. Stichel: Data curation. D.W. Ellison: Supervision. A. von Deimling: Data curation. M. Kool: Supervision. S.M. Pfister: Supervision. V. Ramaswamy: Supervision. A. Korshunov: Supervision, validation, investigation. M.D. Taylor: Supervision. P.H. Sorensen: Conceptualization, supervision, writing and editing, and funding acquisition.
Acknowledgments
The authors would like to thank R. Rassekh, K. O'Halloran and C. Foster for data collection (BCCH cohort), as well as S. Lee for assistance with IHC. This work was supported by St. Baldrick's Foundation - Stand Up To Cancer Pediatric Dream Team Translational Research Grant (SU2C-AACR-DT-27–17) to P. Sorensen. Stand Up to Cancer is a division of the Entertainment Industry Foundation. Research grants are administered by the American Association for Cancer Research, the Scientific Partner of SU2C. A. Delaidelli was supported by a Canadian Institute of Health Research (CIHR) Fellowship, Killam Doctoral Scholarship and Harry and Florence Dennison Fellowship in Medical Research from the University of British Columbia (UBC). A. Korshunov is supported by the Helmholtz Association Research Grant (Germany). M. Ryzhova, A. Golanov, and O. Zheludkova are supported by a RSF Research Grant (# 18–45–06012). This project was also partially supported by the Hannah's Heroes Foundation and the Michael Cuccione Childhood Cancer Research (BCCH cohort).
The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.