Relevance of Molecular Groups in Children with Newly Diagnosed Atypical Teratoid Rhabdoid Tumor: Results from Prospective St. Jude Multi-institutional Trials

This article is featured in Highlights of This Issue, p. 2667

Atypical teratoid rhabdoid tumor (ATRT) is a rare, aggressive central nervous system (CNS) malignancy with an annual incidence of approximately 75 cases in the United States in children ≤19 years old (1). More than two thirds of affected children are <3 years old at diagnosis, with mortality rates approaching 70% (1–7). Although survival has improved with the use of multi-modality therapies, outcomes remain suboptimal, with younger age at diagnosis and presence of metastases associated with the worst outcomes (2, 3, 8, 9). Hallmark somatic inactivating alterations of SMARCB1 on chromosome 22q11.2, resulting in loss of its protein product INI-1 in the tumor, occur in more than 95% of patients with ATRT, with remaining patients having mutations in SMARCA4 located on chromosome 19p13.2. Presence of heterozygous germline alterations (GLA) in either gene results in the rhabdoid tumor predisposition syndrome, but its role in prognosis remains controversial (10–15).

Remarkably, despite very aggressive behavior, ATRT does not exhibit recurrent genetic alterations besides those in SMARCB1 or SMARCA4 (16). Two international collaborative studies reported three molecular groups of ATRT based on tumor DNA methylation and transcriptome findings, suggesting that distinct molecular mechanisms drive oncogenesis (17, 18). There is broad consensus on specific clinicopathologic characteristics of the three groups between the two studies (19), but limited information from prospective studies regarding the clinical relevance of these groups (20).

We report outcomes for participants with newly diagnosed ATRT treated in two prospective risk-adapted multi-institutional trials: those <3 years old at diagnosis, hereon called “infants” [St. Jude Young Children 07 (SJYC07); NCT00602667] and those ≥3 years old at diagnosis, hereon called “children” [St. Jude Medulloblastoma 03 (SJMB03); NCT00085202]. Our analyses incorporates molecular grouping (ATRT-MYC, ATRT-SHH, and ATRT-TYR; ref. 17) and aim to determine the clinicopathologic features and prognostic significance of the three molecular groups. In addition, we investigated the presence of germline SMARCB1/SMARCA4 alterations in the study cohort and the impact of these alterations on treatment outcomes.

SJYC07 and SJMB03 were nonrandomized, phase II, risk-adapted, multi-institutional clinical trials approved by the Institutional Review Boards of St. Jude Children's Research Hospital (Memphis, TN) and eight participating hospitals. Participants with histologic diagnosis of ATRT reported by a pathologist at the enrolling institution and centrally confirmed by study neuropathologists (D.W. Ellison and B.A. Orr), who did not previously receive anticancer therapy were enrolled in the SJYC07 (infants) and in SJMB03 (children) studies. Written informed consent was obtained from parents or legal guardians and from participants 14–17 years old. In addition, assent was obtained from participants 7–13 years old. Patients with other brain tumor diagnoses including medulloblastoma, supratentorial primitive neuroectodermal tumor (PNET), pineoblastoma, ependymoma, high-grade glioma, and choroid plexus carcinoma (SJYC07), and medulloblastoma, PNET, and PNET variants (SJMB03) were also eligible for enrolling in these trials. Both studies were conducted in accordance with the World Medical Association Declaration of Helsinki principles and ethical guidelines.

Study participants underwent maximal safe surgical resection at diagnosis. Extent of resection was determined by the operating surgeon and postoperative MRI, as described previously (21). Metastatic staging included brain and spine MRI and lumbar puncture for cerebrospinal fluid (CSF) analysis (unless medically contraindicated) before study enrollment.

Infants (SJYC07) with no evidence of CNS dissemination (M0 disease) were enrolled in the intermediate-risk arm (IR), whereas those with evidence of metastases (M+ disease) were enrolled in the high-risk arm (HR) of SJYC07 (Fig. 1A and B). Participants with M0 disease whose resection at diagnosis was less than gross total resection (GTR) were considered for second-look resection after two or four cycles of induction chemotherapy to achieve GTR before starting focal radiotherapy. Those with IR disease but <1 year old when completing induction chemotherapy received additional chemotherapy to delay radiotherapy until they were 1 year old, as described previously (21).

Children (SJMB03) were risk stratified as average risk (AR: M0 disease and <1.5 cm² residual tumor at primary site; prescribed 23.4 Gy CSI) or HR (M1–M3 disease or ≥1.5 cm² residual tumor at primary site; prescribed 36–39.6 Gy CSI), followed by consolidation chemotherapy (Fig. 1C). The SJMB03 trial was built on the predecessor St. Jude study SJMB96, the results of which have been published previously (9, 22).

Participants had to have normal organ function, performance score of at least 30, and begin treatment within 31 days of definitive surgery. Treatment was to be continued until completion of therapy, progressive or relapsed disease, unacceptable toxicity, or parental withdrawal of consent. Toxicities were graded per the Common Terminology Criteria for Adverse Events, version 3.0.

Tumor genomic DNA was extracted from formalin-fixed, paraffin-embedded (FFPE) tissue samples using the Maxwell RSC DNA FFPE Kit (Promega). Genome-wide DNA methylation data were generated using Illumina Infinium MethylationEPIC BeadChip arrays (Illumina) according to manufacturer's instructions. Data files were uploaded to the German Cancer Research Center (DKFZ) Molecular Neuropathology 2.0 classifier version 11b4 (https://www.molecularneuropathology.org/mnp), which provided copy-number variants and molecular groups (23). Tumors were classified as ATRT-MYC (MYC), ATRT-SHH (SHH), and ATRT-TYR (TYR), as described previously (17). Calibration scores of >0.9 were used as a cutoff for classifying these tumors as ATRT and for group classification on the classifier. For tumors with calibration scores <0.9 but still predicted to be ATRT by histology, additional t-distributed stochastic neighbor embedding (t-SNE) cluster analysis with a reference cohort was performed to classify them into different groups (23). To perform the clustering, a pairwise distance matrix was generated on the basis of the Pearson correlation from the samples with the most variably methylated probes (SD > 0.228). This matrix was then supplied to RTSNE package for Barnes-Hut t-SNE (24). Subset of Capper dataset containing only ATRT samples was then used to further refine the clustering.

Peripheral blood DNA was used to detect the presence of germline SMARCB1/SMARCA4 variants (Supplementary Information, germline methodology).

ATRT was a secondary cohort for both SJYC07 and SJMB03. Herein, we describe progression-free survival (PFS) and overall survival (OS) of the ATRT cohort, consistent with secondary objectives of the trials. In addition, the secondary biological objectives for molecular grouping of ATRT by tumor DNA methylation analysis and their association to outcomes are presented.

SJYC07 was activated in November 2007 and closed to accrual in May 2017. SJMB03 opened in June 2003 and closed to accrual in March 2013. All eligible participants for whom therapy was initiated were included in outcome analyses. All participants with adequate tumor for DNA methylation profiling were included in biological analyses. PFS was defined as time from diagnosis to date of relapse or progressive disease (PD) or death from any cause, or to date of last follow-up for patients without events. OS was defined as time from date of diagnosis to date of death from any cause or to date of last follow-up for survivors. Outcome distributions were estimated using Kaplan–Meier analysis and reported as ± one SE, where SEs were obtained by the Peto and Pike method. Fisher exact tests and exact χ² tests examined associations among categorical variables. Exact Wilcoxon rank-sum tests and Kruskal–Wallis tests examined associations between age at diagnosis and protocol-defined risk groups as well as with methylation groups. Differences in outcome distributions were examined by exact log-rank tests. Cox regression was used to estimate associations with multiple predictors of outcome. Cumulative incidence of local and distant failure was estimated by methods of Kalbfleisch and Prentice (25). Local failure and distant failure were defined as the time from diagnosis to local disease recurrence/progression or metastatic progression, respectively. Gray test was used to compare cumulative incidence curves.

The trials included 52 infants and 22 children with ATRT (Table 1). A diagnostic lumbar puncture could not be obtained because of medical contraindications in 13 infants who had no imaging evidence of metastases (disease status coded as MX). All SJMB03 participants received CSI. Eleven (all M0) were treated in the AR arm and 11 (3 M0 with residual tumor, 8 M+) in the HR arm. Supplementary Fig S1 shows treatment details for both studies.

Tumor DNA methylation data were available for 67 of 74 (91%) patients. Of them, 56 were classified as SHH, TYR, or MYC using the DKFZ classifier with a calibration score of >0.9. Eight other samples with classifier scores for groups between 0.23 and 0.89 were additionally grouped by cluster analysis to a reference cohort. Three samples could not be further classified molecularly and are not included in methylation group–based analyses presented here. These three samples, like all other samples, were centrally reviewed and confirmed histologically to be ATRT. The most common group was SHH (30/64; 47%), followed by TYR (21/64; 33%) and MYC (13/64; 20%). MYC patients were older than SHH and TYR patients (P = 0.019), a higher proportion of SHH patients were M+ (P = 0.016) and had GLA (P = 0.057), whereas most TYR patients were M0 and had infratentorial primary tumors. Copy-number alterations (CNA) of the SMARCB1 locus for groups were classified manually into different categories, and heatmaps demonstrating CNAs were generated (Fig. 2). Broad 22q chromosomal losses spanning across the SMARCB1 locus was observed predominantly in the TYR group whereas more focal losses were observed in the MYC group (Fig. 2B).

At the time of analysis, 24 of 74 (32%) patients were alive at a median follow-up of 8.4 years (range, 3.1–14.1 years) from diagnosis. First events included disease progression or death from disease in 53 patients, second malignancy in 1 patient (paraspinal desmoid tumor), and metachronous rhabdoid tumor in 2 patients (kidney and pelvis soft tissues) with a GLA in SMARCB1. Because treatments on the two trials differed markedly, we investigated outcomes by protocol.

Participants with IR disease (n = 34) had a 5-year PFS of 31.4 ± 9.2% and OS of 43.9 ± 9.5%, whereas the 5-year PFS and OS for HR participants (n = 18) were 0% (Fig. 3A and B). Participants with M0 disease (n = 23) had a 5-year PFS of 39.1 ± 11.5% and OS of 51.8 ± 12% (Fig. 3C and D). Univariate analysis showed the best survival for infants in the TYR group (5-year OS estimates 58.8 ± 11.9%, P = 0.023; Fig. 4A and B). Because patients in the TYR group were more likely to be nonmetastatic at presentation (Fig. 2A; Supplementary Table S1), we performed the same comparison by risk group and found no difference in outcome by methylation group in M0 patients (P = 0.74 for PFS and 0.73 for OS; Fig. 4C and D). In addition, outcomes did not differ by age at diagnosis, sex, primary location of tumor, and extent of best resection before radiotherapy for those with IR disease (Supplementary Table S2; Supplementary Fig. S2A–S2D).

Children with AR disease (n = 11) had a 5-year PFS of 72.7 ± 12.7% and OS of 81.8 ± 11%, whereas those with HR disease (n = 11) had a 5-year PFS and OS of 18.2 ± 9.5% (Fig. 3A and B). Children with M+ disease (n = 8) had a 5-year OS of 25.0 ± 12.5%. There was no difference in outcomes by sex or tumor location. In univariate analyses, outcomes of those with AR disease were better than those with HR disease (Supplementary Table S3).

Outcomes for MYC, SHH, and TYR groups by study and risk stratification are shown in Supplementary Fig S3A–S3C.

Thirty percent (16/53) of study participants who completed germline testing were positive for heterozygous GLA in either SMARCB1 (15/16) or SMARCA4 (1/16) [2/17 (12%) SJMB03 participants with positive results vs. 14/36 (39%) of SJYC07 participants; Fig. 5; Supplementary Table S4]. Median age at diagnosis for patients with a positive GLA was significantly lower at 0.7 years (range, 0.0–6.9) versus those without GLA having a median age of 2.3 years (range, 0.0–12.1; P = 0.003). There was a significant association between metastatic status and GLA. More M+ patients had GLA (8/17, 47%) compared with M0 patients (5/29;17%), MX excluded (P = 0.044). None of the MYC participants were positive for GLA compared with 11 of 24 (46%) SHH and 4 of 13 (31%) TYR participants (P = 0.057; Fig. 2). There was no evidence of GLA being associated with outcomes in SJYC07 participants (Supplementary Fig. S4A and S4B).

Fifty of 74 study participants had PD or relapse. Median time to treatment failure for infants (n = 39) was 5.7 months (range, 1.6–79.4 months). Three of 39 infants had PD more than 2 years from diagnosis, with one of them having tissue-confirmed PD after 7 years from diagnosis. Disease progression was observed during premaintenance phase in 18 of 23 infants who failed on the IR arm and all 16 infants who failed on the HR arm. None of the infants developed PD during the maintenance phase of chemotherapy. The median time from completion of maintenance therapy to PD for the remaining five of 23 infants who failed was 12.7 months (range, 0–67.2). Four of these five infants had ATRT-TYR and one ATRT-SHH (Supplementary Table S5). Only 5 patients are alive after PD at a median follow-up of 3.9 years (range, 0.2–8.1 years) since progression. Pattern of failure for those in the IR stratum (n = 23) was local (n = 9), distant (n = 8), or combined (n = 6). However, failure was predominantly distant (n = 7) versus local (n = 3) versus combined (n = 2) for 12 of 23 patients who had PD after completing focal radiotherapy.

Median time to PD for children (n = 11) was 9.2 months (range, 3.5–52.0 months). Two participants experienced treatment failures beyond 2 years, at 50.3 and 52 months from diagnosis. Tissue confirmation was not available for these 2 patients to determine whether this was true PD, metachronous tumor, or a second malignancy. Neither patient harbored germline SMARCB1/SMARCA4 alterations. All patients with PD died at a median of 2.0 months after PD (range, 0.4–54.1 months). Supplementary Fig. S5 shows the cumulative incidence of local and distant failures by protocol.

Febrile neutropenia was the most common adverse event (Supplementary Table S6). There were no cases of radiotherapy-induced necrosis in either trial. An infant with M+ disease developed febrile neutropenia and respiratory syncytial virus pneumonitis that required high-frequency ventilation and died after the family decided to withdraw support due to his medical condition and poor prognosis of ATRT.

This study reports survival data for infants and children with newly diagnosed ATRT treated in SJMB03 and SJYC07 trials. We observed very good outcomes in children ≥3 years old with M0 disease by using immediate postoperative CSI and adjuvant chemotherapy. However, patients with metastatic ATRT had a dismal prognosis. We also report clinical characteristics and outcomes of three epigenetic groups of ATRT, which is the largest prospectively treated cohort reported to date. Our study, very importantly, reports new findings while confirming some of the clinicopathologic characteristics of three epigenetic groups reported previously in retrospective studies (17). As reported previously, we found a predominance of SHH and TYR groups in younger children and the predominance of MYC group in older children. We also confirmed the pattern of somatic SMARCB1 CNA, with the TYR group showing a predominance of broad losses and the MYC group a predominance of focal losses. The majority of children with M+ disease were in the SHH group, whereas those in the TYR group primarily presented with localized posterior fossa tumors. The TYR group had a better OS, but for infants with M0 disease, group affiliation was not associated with outcome, in contrast to that reported in a recent study (26). We observed that infants with TYR group have an indolent progression and longer survival compared with those with SHH or MYC group (Fig. 4; Supplementary Table S5). Also, it is significant to note that TYR group overwhelmingly presents with localized disease. The biological basis, if any, of this relative indolent behavior remains to be elucidated.

Treating children with ATRT has historically been challenging, with dismal outcomes reported by several studies (3, 5, 27, 28). The contribution of radiotherapy in improving survival is controversial (2, 5, 9, 29). Although a multi-center study reported a 2-year PFS of 58% with multi-modality therapy, the majority of children <3 years old in that study had a follow-up period of <2 years from diagnosis, whereas all surviving children >3 years old received radiotherapy, including some receiving CSI (8). The European Rhabdoid registry study EU-RHAB reported a 6-year OS of 45% despite the use of multi-modality therapy, including intrathecal chemotherapy, whereas the Children's Oncology Group (COG) study ACNS0333 reported a 2-year OS of 48% when using high-dose chemotherapy (HDCT) and radiotherapy, which included four therapy-related deaths (2, 20, 26, 30, 31).

In our study, use of trimodality therapy in infants with M0 disease was associated with a 5-year PFS of 39.1 ± 11.5% and OS of 51.8 ± 12%. Interestingly, those with MX disease had inferior survival, suggesting that some of these participants may have had metastatic disease in the form of positive CSF, which was missed because of the absence of a lumbar puncture (LP). Hence, we recommend that every effort be made to obtain a diagnostic LP. If this is not possible, these children should be considered as having HR disease and treated accordingly. We have taken this approach in our ongoing ATRT trial SJATRT (NCT02114229). Outcomes were better for children ≥3 years old with M0 disease with nonbulky (<1.5 cm²) residual tumor treated with adjuvant CSI and consolidation chemotherapy, with a 5-year PFS of 72.7 ± 12.7% and OS of 81.8 ± 11%, thus confirming our previous report of a single-institution series showing excellent outcomes for this group of children (9). Use of CSI at a young age is associated with neurocognitive decline, with the additional risk of endocrine dysfunction and skeletal growth retardation (32). However, given the aggressiveness of ATRT and its resistance to multi-modality therapies, optimizing treatment at initial diagnosis is vital to the survival of these children. This warrants an informed discussion between the treating physician and family at the time of initial diagnosis to balance improved survival and possible long-term CSI-associated adverse outcomes. Children with metastases at diagnosis continue to fare poorly despite receiving CSI. Despite better outcomes reported compared with our results with the use of HDCT in conjunction with radiotherapy for those with M+ disease in COG ACNS0333, outcomes remain suboptimal, highlighting an urgent need to develop novel therapies for this group of patients (20).

The association between germline predisposition to rhabdoid tumors and prognosis is controversial. Although a study noted a higher risk of death in patients with a germline predisposition, other studies have not confirmed this observation (2, 14, 15). We found that germline predisposition was not associated with adverse prognosis in the SJYC07 cohort. This association could not be tested in the SJMB03 cohort, because only two participants harbored the alterations. We also found an association between metastatic status and GLA, with M+ patients more likely to harbor a GLA, and a marginal association between group affiliation and germline predisposition, with a higher incidence in the SHH group and none of the participants in the MYC group having a GLA in either SMARCB1 or SMARCA4.

Weaknesses of our study include the single-arm phase II design, relatively small sample sizes, and reliance on comparison with historical controls. As ATRT is an extremely rare condition, a large randomized study would be difficult to complete in a reasonable time frame. Also, given the rarity of this cancer, multi-variate analysis could not be conducted because of the small number of patients and events within each molecular group.

In conclusion, our results demonstrate that the TYR group is associated with nonmetastatic disease and superior survival in infants, whereas the SHH group is associated with metastatic disease and extremely poor outcomes in the presence of metastases. No association with outcomes was detected for molecular group affiliation in infants with M0 disease. Presence of germline SMARCB1 alterations was not associated with inferior survival. The use of maximal safe surgical resection, postoperative CSI, followed by adjuvant consolidation chemotherapy in children ≥3 years old with M0 disease, yields very high survival and should be considered as a treatment option by practicing oncologists for this highly aggressive, often fatal pediatric CNS malignancy. A subset of children <3 years old with M0 ATRT achieved long-term survival with the SJYC07 therapy and radiotherapy without using HDCT. Future clinical trials should use this information for risk stratification of patients and further refine it by combining biologically driven and possibly group-specific targeted therapies with conventional trimodality therapies to improve outcomes for children with this highly aggressive brain tumor.

S.A. Upadhyaya reports grants from NIH and American Lebanese Syrian Associated Charities during the conduct of the study. G.W. Robinson reports personal fees from Roche and Eli-Lilly outside the submitted work. A. Onar-Thomas reports grants from NCI during the conduct of the study. A. Onar-Thomas also reports grants and nonfinancial support from Novartis, Apexigen, Novocure, Pfizer, and Celgene, as well as nonfinancial support from SecuraBio, Senhwa, and Incyte outside the submitted work. I. Qaddoumi reports personal fees from Springworks and AstraZenca outside the submitted work. A. Vinitsky reports grants from NIH and ALSAC during the conduct of the study. E. Bouffet reports grants from Roche and Bristol Myers Squibb outside the submitted work. A. Gajjar reports grants from NCI during the conduct of the study. No disclosures were reported by the other authors.

S.A. Upadhyaya: Resources, data curation, formal analysis, investigation, writing–original draft, writing–review and editing. G.W. Robinson: Resources, data curation, formal analysis, investigation, writing–original draft, writing–review and editing. A. Onar-Thomas: Conceptualization, data curation, software, formal analysis, supervision, methodology, writing–original draft, writing–review and editing. B.A. Orr: Resources, formal analysis, investigation, writing–review and editing. P. Johann: Data curation, formal analysis, writing–review and editing. G. Wu: Resources, data curation, formal analysis, writing–review and editing. C.A. Billups: Data curation, software, formal analysis, methodology, writing–original draft, writing–review and editing. R.G. Tatevossian: Resources, data curation, writing–review and editing. S.K. Dhanda: Data curation, formal analysis, writing–review and editing. A. Srinivasan: Conceptualization, resources, investigation, methodology, writing–review and editing. A. Broniscer: Resources, investigation, writing–review and editing. I. Qaddoumi: Resources, investigation, writing–review and editing. A. Vinitsky: Resources, investigation, writing–review and editing. G.T. Armstrong: Resources, investigation, writing–original draft, writing–review and editing. A.E. Bendel: Resources, investigation, writing–review and editing. T. Hassall: Resources, investigation, writing–review and editing. S. Partap: Resources, Investigation, writing–review and editing. P.G. Fisher: Resources, investigation, writing–review and editing. J.R. Crawford: Resources, investigation, writing–review and editing. M. Chintagumpala: Resources, investigation, writing–review and editing. E. Bouffet: Resources, investigation, writing–review and editing. S. Gururangan: Resources, investigation, writing–review and editing. R. Mostafavi: Resources, data curation, formal analysis, writing–review and editing. R.P. Sanders: Conceptualization, resources, investigation, writing–review and editing. P. Klimo Jr: Resources, investigation, writing–review and editing. Z. Patay: Resources, data curation, investigation, writing–review and editing. D.J. Indelicato: Resources, investigation, writing–review and editing. K.E. Nichols: Formal analysis, supervision, writing–review and editing. F.A. Boop: Resources, investigation, writing–review and editing. T.E. Merchant: Conceptualization, resources, data curation, supervision, investigation, writing–review and editing. M. Kool: Data curation, formal analysis, writing–review and editing. D.W. Ellison: Conceptualization, resources, data curation, supervision, investigation, writing–review and editing. A. Gajjar: Conceptualization, resources, data curation, formal analysis, supervision, investigation, methodology, writing–review and editing.

The authors thank Aksana Vasilyeva for technical support; Vani Shanker for editing the article; Mathew Lear, technical director, St. Jude biorepository, and the staff of Diagnostic Biomarkers Shared Resource, St. Jude Children's Research Hospital, for technical support; staff in the Hartwell Center, St. Jude Children's Research Hospital, for generating the tumor DNA methylation data; and Joshua Stokes, Department of Biomedical Communications, St. Jude Children's Research Hospital, for the graphics.

This work was supported by grant CA21765 from the NIH. The content is solely the responsibility of the authors and does not necessarily represent the official views of the NIH. This study was also supported by the American Lebanese Syrian Associated Charities (ALSAC). Funding was provided to St. Jude Children's Research Hospital cancer center for conduct of the study and not to any individual researcher.

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.