Impact of Genomic and Clinical Factors on Outcome of Children ≥18 Months of Age with Stage 3 Neuroblastoma with Unfavorable Histology and without MYCN Amplification: A Children’s Oncology Group (COG) Report

Abstract

PURPOSE:

Patients ≥18 months of age with INSS Stage 3 unfavorable histology (UH), MYCN-non-amplified (MYCN-NA) tumors have favorable survival rates compared to other high-risk neuroblastoma populations. The impact of select clinical and biological factors on overall (OS) and event-free survival (EFS) were evaluated.

PATIENTS AND METHODS:

Patients enrolled on COG A3973 (n=34), ANBL0532 (n=27), and/or biology protocol ANBL00B1 (n=72) were analyzed. Tumors with available DNA (n=65) and RNA (n=42) were subjected to whole exome sequencing (WES) and RNAseq. WES analyses and gene expression profiling were evaluated for their impact on survival. Multivariate analyses of EFS/OS using significant factors from univariate analyses were performed.

RESULTS:

5-year EFS/OS for patients treated with high-risk therapy on A3973 and ANBL0532 were 73.0±8.1%/87.9±5.9% and 61.4±10.2%/73.0±9.2%, respectively (p=0.1286 and p=0.2180). In the A3973/ANBL0532 cohort, patients with <partial response (PR;n=5) at end-Induction had poor outcomes (5-year EFS/OS: 0%/20.0±17.9%. Univariate analyses of WES data revealed that subjects whose tumors had chromosome 1p or 11q loss/LOH and chromosome 5 or 9 segmental chromosomal aberrations had inferior EFS compared to those with tumors without these aberrations. Multivariate analysis revealed that 11q loss/LOH was an independent predictor of inferior OS [HR 3.116(95%CI 1.034–9.389), p=0.0435].

CONCLUSION:

Patients ≥18 months of age at diagnosis who had tumors with UH and MYCN–NA INSS Stage 3 neuroblastoma assigned to high-risk therapy had an 81.6±5.3% 5-year OS. Less than PR to induction therapy and chromosome 11q loss/LOH are independent predictors of inferior outcome and identify patients who should be eligible for future high-risk clinical trials.

Introduction

Neuroblastoma is a malignancy of the sympathetic nervous system that is marked by biologic and clinical heterogeneity. Patients with neuroblastoma are classified as having low-, intermediate- or high-risk disease based on clinical (age, clinical stage, histopathology) and biologic (amplification of the MYCN proto-oncogene, tumor cell ploidy) factors(1–5). Treatment is assigned based on this risk stratification, and outcomes range from very high rates of spontaneous regression without intervention for subsets of patients with low-risk neuroblastoma to a significant risk of death despite aggressive multimodal therapy for patients with high-risk neuroblastoma(4). Patients with International Neuroblastoma Staging System (INSS) Stage 3 neuroblastoma (unresectable but localized tumors that cross the midline with or without regional lymph node involvement or localized tumors that involve contralateral lymph nodes) have widely variable clinical outcomes(6). Because of this, risk classification and treatment strategies for Stage 3 patients, representing 14.8% of neuroblastoma patients in a recent analysis of the COG biobanking study ANBL00B1(6), differ widely and international consensus does not yet exist regarding their optimal risk classification and therapy. Prior cooperative group studies have identified amplification of MYCN, unfavorable histopathology (UH), and elevated serum ferritin as adverse prognostic factors in this group of patients. Intensification of therapy for Stage 3 patients based on factors such as MYCN amplification status has led to improvements in outcome(7–10).

Treatment for patients with MYCN-non-amplified (MYCN-NA) Stage 3 tumors varies widely from observation for spontaneous regression for infants to moderate or dose intensive multi-agent chemotherapy(11–14). A large retrospective analysis of 1,483 INSS Stage 3 patients diagnosed between 1990–2002 included in the International Neuroblastoma Risk Group Database (INRGdb), an international repository of patient biological data and clinical outcomes, evaluated the impact of baseline clinical factors and treatment received on EFS and OS. Similar to previous reports, in this analysis, age ≥18 months, tumor MYCN amplification and unfavorable histopathology were found to be independent predictors of inferior outcomes regardless of assigned treatment(15). Within the cohort of patients ≥18 months of age at diagnosis who had MYCN-NA but UH tumors (n=359), the type of treatment received [ranging from no treatment to high-dose chemotherapy with autologous peripheral blood stem cell rescue (HDC-ASCR)] was not correlated with outcome(15).

These findings, in part, have led to significant variability worldwide in the management of patients ≥18 months with Stage 3, MYCN-NA, UH tumors. The Société Internationale d’Oncologie Pédiatrique European Neuroblastoma (SIOPEN) Low and Intermediate Risk Neuroblastoma European Study (LINES) demonstrated a 5-year EFS and OS of 59.8% and 76.1 %, respectively, for patients ≥12 months of age with unresectable, MYCN-NA neuroblastoma receiving combination chemotherapy (alternating cycles of carboplatin/etoposide and vincristine/doxorubicin/cyclophosphamide for a total of 4 neo-adjuvant and 2 adjuvant cycles) and surgical resection in the majority of patients (NCT00025428)(12). Based on these results, the current SIOPEN LINES2009 trial (NCT01728155) is evaluating whether the addition of radiotherapy to the primary tumor and oral 13 cis-retinoic acid to chemotherapy and surgery will improve outcome.

In contrast, in the Children’s Oncology Group (COG), patients with INSS Stage 3 neuroblastoma ≥18 months (≥547 days) of age at diagnosis with MYCN-NA but UH tumors are eligible to enroll onto clinical trials for high-risk disease utilizing intensive, multi-modality therapy. In the Children’s Cancer Group (CCG) 3891 randomized Phase 3 trial for patients with high-risk neuroblastoma evaluating myeloablative chemotherapy with autologous bone marrow transplant (ABMT) versus continuing chemotherapy, Stage 3 patients randomized to ABMT (n=20) had a 5-year EFS of 65±11% and 5-year OS of 65±11% compared to 41±11% (P=0.21) and 46±11% (P=0.23) for patients randomized to continuing chemotherapy (n=23). The small group of patients randomized to both ABMT and 13-cis-RA (n=6) had a 5-year EFS of 80±11% and OS of 100%. The differences in survival between the ABMT and continuing chemotherapy groups were not statistically significant, likely at least in part due to the fact that these sub-analyses of the larger high-risk cohort were not powered to detect differences in this relatively rare group of high-risk patients (14).

This analysis reports the outcome of INSS Stage 3 patients ≥18 months with MYCN-NA, UH tumors when treated with COG high-risk therapy on two successive Phase 3 studies, COG A3973 (NCT00004188) and ANBL0532 (NCT00567567), consisting of induction chemotherapy, surgical resection, a single myeloablative chemotherapy course with peripheral blood stem cell rescue and local radiotherapy (16, 17). Patients on A3973 received 6 cycles of induction therapy consisting of alternating cycles of cyclophosphamide, doxorubicin and vincristine (cycles 1,2 4 and 6) with cisplatin and etoposide (cycles 3 and 5) while patients on ANBL0532 had cycles 1 and 2 replaced with cycles of cyclophosphamide and topotecan. Radiotherapy differed between A3973 and ANBL0532 in that patients on ANBL0532 with residual primary tumor after an attempt at resection received boost radiotherapy to 3600 cGy while patients with a complete resection received 2160 cGy, whereas all patients on A3973 received 2160 cGy, regardless of degree of resection. Eligible patients on both studies received a single myeloablative course of carboplatin, etoposide and melphalan. These patients were also eligible to receive dintuximab-based post-consolidation on either COG ANBL0032 (NCT00026312) or ANBL0931 (NCT01041638). In addition, newly diagnosed patients in this group were eligible to provide biologic specimens on the ANBL00B1 biology protocol (NCT00904241) regardless of therapeutic trial enrollment status. We sought to determine whether additional clinical or biological factors could further refine outcome prediction for this group of patients, as identification of such factors could help inform treatment intensity in ongoing or future clinical trials.

Patients and Methods

Patients with INSS Stage 3 disease, age ≥18 months of age at diagnosis whose tumors were MYCN non-amplified and had unfavorable histological features who enrolled on COG high-risk therapeutic protocols A3973 (n=34) (16) and ANBL0532 (n=27) (17), or biology protocol ANBL00B1 only (n=72) were analyzed. Of these, 101 patients (ANBL00B1 n=72, A3973 n=6, ANBL0532 n=23) had biospecimens available for analysis, referred to as cohort ANBL14B8. All patients or their legal guardians provided informed written consent for the included studies in accordance with the Declaration of Helsinki, human investigations were performed after approval by an institutional review board and in accordance with an assurance filed with and approved by the U.S. Department of Health and Human Services.

Response was assessed at end-Induction as per the 1993 version of the International Neuroblastoma Response Criteria (INRC)(18). If patients were designated as having progressive disease (PD) at a timepoint prior to end-Induction and were missing end-Induction response, they were coded as having PD at end-Induction. End-induction response was not available for patients who were not treated on a therapeutic trial.

EFS and OS estimates were calculated per Kaplan-Meier with standard errors(19). Survival was compared using a log-rank test. EFS was defined as time from diagnosis to first episode of relapse, progression, death, or secondary malignancy; patients without an event were censored at the date of last contact. OS was defined as time from diagnosis to death; surviving patients were censored at the date of last contact. Tests for violations of the proportional hazards assumption were performed. P-values <0.05 were considered statistically significant.

A series of univariate analyses were performed to evaluate the impact of the evaluated clinical and biological variables on EFS and OS. A series of multivariable Cox proportional hazards models using the chromosomes that were statistically significant (p<0.05) from the univariate analyses were fit for EFS and OS. For both the univariate and multivariable analyses, age was included in the models in two separate ways: as a continuous variable or as a categorical variable (18 months to <5 years vs. ≥5 years). Separate EFS models were fit for chromosome 1p and chromosome 1p36. Backward selection was used to determine the most parsimonious model.

Genomic studies

For whole-exome sequencing (WES) the quality and quantity of DNA was assessed using the Genomic DNA Screen Tape Assay (Agilent Technologies). The concentration of gDNA greater than 200bp was calculated and 100ng was sheared in 50μL of nuclease-free water with the Covaris E220 (Covaris). Dual-indexed adapters were ligated using the KAPA Hyper Prep Kit (Roche) according to manufacturer’s recommendations. Exome capture was performed using Agilent’s SureSelect XT Low-Input Reagent Kit with SureSelect V6 (Agilent) and custom probes capture libraries. Each library was normalized and paired-end sequenced on Illumina’s NovaSeq 6000 (Illumina). Copy number analysis was performed utilizing Nexus Copy Number v10 (BioDiscovery, El Segundo, CA). Reference normal copy number was generated using ten normal tissues (tonsils) per BioDiscovery’s recommended settings and used to perform somatic copy number analysis on WES tumor data. FASTQ files were converted using industry standard BCL2FASTQ v1.8.4. WES data were aligned to GRCh37 (hs37d5) by BWA(v0.7.8-r455), followed GATK’s Base Recalibrator (v3.5.0). Next, Picard Tools(v1.128) merged aligned bams and marked duplicate reads. Germline calling was performed by GATK’s Haplotype Caller and Freebayes (v1.1.0–6-gf069ec6). Somatic variant calling was performed by Manta (v.1.1.1), Strelka (v2.7.1), and Seurat (v2.6). SnpEff (v3.6h) was used to annotate gene variant effects. Copy number analysis was completed using tCoNuT (https://github.com/tgen/tCoNuT).

The segmental copy number assignment was manually curated by three independent investigators (RP, SA and RK). Segmental copy number calls were made based on algorithms utilized by the COG Biopathology Center for specific loci for copy number alterations (1p loss, 1q gain, 2p gain, 3p loss, 4p loss, 11q loss and 17q gain), or loss of heterozygosity (LOH in 1p36.22-p36.32, 11q14q23, or 11p). Segmental copy numbers were called for regions with ≥1 megabase (Mb) alteration, whole chromosome arm alteration, and any regions of LOH (≥5 Mb LOH region). Known amplification regions (MYCN, ALK, MYC, MDM2) were also evaluated. Additionally, chromosomes 5–10, 12–16, 18–22 were also assessed using similar criteria without regard to the recurrence of any SCA regions in these chromosomes. Whole chromosome gain or loss with consistent pattern of LOH without any of the segmental abnormalities were not defined as SCA. Regions and chromosomes were annotated as having no SCA or SCA. Samples were categorized for having quiet genome in all chromosomes, only having whole chromosome alterations, or assigned the number of SCA observed in their genomes.

For RNA-sequencing, 150ng of total RNA was used to enrich for poly-adenylated RNA, which were heat-fragmented to a target size of 180bp and converted to cDNA using random primers with Superscript II (Invitrogen), then used for library prep using the Illumina TruSeqRNA library kit. RNAseq’s FASTQ’s were aligned to GRCh37(hs37d5) by STAR (v2.5.3a). Star-Fusion (v0.8.0) and Fusion catcher (v0.99.7b) were used to identify potential fusion transcripts. Salmon (v0.7.2) was used to quantify transcripts.

For gene set analyses, RNA-seq data were aligned using Hisat2 (version 2.2.0), reads were counted using featureCounts (version 2.0.1)(20), and TMM method in the edgeR package was used for the gene count normalization(21). EdgeR (version 3.34.1 ) was used to perform the differential analysis and gene sets were scored using Gene Set Variation Analysis, as implemented in the GSVA R (version 1.40.1)package(22). Gene sets were tested for enrichment in rank ordered lists using the fgsea R package. Unsupervised clustering was performed using non-negative matrix factorization (NMF)(23) on the 1000 gene counts with highest coefficient of variation across samples. Adrenergic (ADR) and mesenchymal (MES) gene signatures have been previously described(24). The R v3.6.0 package was utilized for statistical and graphical processes.

Data Availability Statement

Sequence data are available under Gene Expression Omnibus (GEO: GSE218527) and Sequence Read Archive (SRA: PRJNA903956).

Results

Clinical Outcomes

Characteristics of patients enrolled on therapeutic studies and those enrolled on ANBL00B1 are shown (Table 1). Of these, 101 patients comprised ANBL14B8 (only ANBL00B1 n=72, A3973 n=6, ANBL0532 n=23) and had available biospecimens submitted through ANBL00B1 for analysis. The proportional hazards assumption was found not to be violated in any analysis. The 5-year EFS and OS for children ≥18 months with INSS stage 3, MYCN–NA, UH disease in this cohort (N=101) were 67.9±5.3% and 77.7±4.7%, respectively. For the subset treated on therapeutic trials A3973 (n=34) and ANBL0532 (N=27), 5-year EFS and OS were 73.0±8.1% and 87.9±5.9%; 61.4±10.2% and 73.0±9.2%, respectively, with no statistically significant differences in EFS or OS between the two therapeutic cohorts (p=0.1286 and p=0.2180, respectively; Figure 1a-b).

Table 1.

Characteristics of Eligible Patients.

Characteristic		No. (%)
	A3973 (n=34)	ANBL0532 (n=27)	A3973 + ANBL0532 (n=61)	ANBL00B1 [not in A3973 or ANBL0532] (n=72)	ANBL14B8 Cohort (n=101)
Age at diagnosis, months (median, range)	40.49 (18.27, 189.08)	33.51 (18.04, 235.30)	37.55 (18.04, 235.30)	36.99 (18.14, 240.23)	36.86 (18.04, 240.23)
A3973 Enrollments	34 (100.00)	0 (0.00)	34 (55.74)	--	6 (5.94)
ANBL0532 Enrollments	0 (0.00)	27 (100.00)	27 (44.26)	--	23 (22.77)
Sex
• Female	18 (52.94)	17 (62.96)	35 (57.38)	36 (50.00)	53 (52.48)
• Male	16 (47.06)	10 (37.04)	26 (42.62)	36 (50.00)	48 (47.52)
Primary site
• Adrenal	6 (18.18)	6 (23.08)	12 (20.34)	13 (18.84)	18 (18.56)
• Abdominal, other	20 (60.61)	14 (53.85)	34 (57.63)	38 (55.07)	55 (56.70)
• Neck	--	--	--	3 (4.35)	3 (3.09)
• Thorax	4 (12.12)	3 (11.54)	7 (11.86)	6 (8.70)	8 (8.25)
• Other	3 (9.09)	3 (11.54)	6 (10.17)	9 (13.04)	13 (13.40)
• Unknown	1	1	2	3	4
Response after induction therapy
• Complete/Very Good Partial Response	20 (64.52)	19 (73.08)	39 (68.42)
• Partial Response	10 (32.26)	3 (11.54)	13 (22.81)
• No/Mixed Response	--	1 (3.85)	1 (1.75)
• Progressive Disease	1 (3.23)	3 (11.54)	4 (7.02)
• Not Evaluated or Missing	3	1	4
Immunotherapy on ANBL0032 or ANBL0931	3 (8.82)	20 (74.07)	23 (37.70)	16 (22.22)	34 (33.66)
Relapse	7 (20.59)	10 (37.04)	17 (27.87)	12 (16.67)	23 (22.77)
• Local	1 (2.94)	4 (14.81)	5 (8.20)	4 (5.56)	8 (7.92)
• Metastatic	5 (14.71)	4 (14.81)	9 (14.75)	7 (9.72)	12 (11.88)
• Local + Metastatic	1 (2.94)	2 (7.41)	3 (4.92)	1 (1.39)	3 (2.97)
5-year EFS ± standard error	73.0 ± 8.1	61.4 ± 10.2	68.0 ± 6.4	71.1 ± 6.2	67.9 ± 5.3
5-year OS ± standard error	87.9 ± 5.9	73.0 ± 9.2	81.6 ± 5.3	78.8 ± 5.5	77.7 ± 4.7

Figure 1. Clinical Outcomes and Impact of Clinical Features on Outcome.

(A) EFS and (B) OS of all patients and by trial cohort. (C) EFS and (D) OS by end-induction response. (E) EFS and (F) OS by receipt of dinutuximab.

In the combined cohort of patients enrolled on A3973 and ANBL0532 with end-Induction response data available (n=57), Table 1, statistically significant differences in survival outcomes were detected in patients based upon end-Induction response category. Patients with CR/VGPR (n=39) and PR (n=13) had better outcomes than those with <PR (n=5) [(5-year EFS: 74.0±7.6% vs. 75.0±12.5% vs. 0%; p<0.0001) and 5-year OS: 84.4±6.2% vs. 100% vs. 20.0±17.9%; p<0.0001); Figure 1c-d].

There were no differences in EFS and OS for patients who received dinutuximab-based immunotherapy on ANBL0032 or ANBL0931 (n=23) compared with patients who did not receive immunotherapy (n=38) [(p=0.8390 and p=0.2575, respectively); Figure 1e-f].

Impact of Histology on Outcome

For the ANBL14B8 cohort with available biospecimens (n=101), centrally-reviewed histologic subtypes of neuroblastoma were evaluated for their impact on outcome. No significant differences in EFS/OS were found between patients with ganglioneuroblastoma, nodular (GNB-N) versus neuroblastoma (including Schwannian stroma-poor neuroblastomas). Five-year EFS for patients with GNB-N was 73.4±10.9% (p=0.3316) vs 66.1±6.0% for those with neuroblastoma; 5-year OS was 86.6±8.2% vs 74.9±5.5% (p=0.1653).

Impact of SCA on Outcome

Of the 101 patients with available diagnostic pre-treatment biospecimens collected via ANBL00B1, 64/101 had high-quality DNA available for WES. EFS for patients whose tumors were found by WES to harbor SCAs were compared to EFS for patients whose tumors either had no copy number change or had whole chromosomes loss or gain. A summary of evaluated clinical and biologic factors stratified by EFS is shown (Figure 2) demonstrating heterogeneity in number of SCAs in this population, with the most common alterations observed in chromosomes 17q, 11q, and 1p (52%, 46%, and 40%, respectively). Univariate analysis showed significant differences in EFS based on copy number status (normal copy number vs copy number change) of chromosomes 1p (5-year EFS: 52.6±13.7% vs 70.6±7.7%, p=0.0484), 5 (40.6±14.0 vs 74.2±7.3, p=0.0118), 9 (36.4±14.5% vs 73.5±7.2%, p=0.0177), and 11q (50.8±10.7% vs 78.1±8.0%, p=0.0108). SCA involving chromosome 1p36 specifically (a previously identified smallest region of overlap on this chromosome(25)) was also associated with inferior EFS (48.5±15.6% vs 70.5±7.4%, p=0.0160) in univariate analysis. Chromosome 11q loss was associated with a significant difference in OS (5-year OS: 62.4±10.2% vs 90.4±5.6%, p=0.0336; Figure 2). The presence of other SCAs did not independently predict outcome in this cohort.

Fig 2. Summary Heatmap of Clinical and Biological Factors Evaluated Stratified by EFS.

An association between the number of SCAs identified and survival was not detected, however patients who had tumors with no chromosomal aberrations or numerical chromosomal aberrations had superior EFS when compared to subjects whose tumors harbored 1–2 or ≥3 SCAs [(log-rank p=0.0195); Figure 3]. An SCA number cutoff that identified subjects with an inferior OS could not be defined.

Figure 3. Impact of 11q SCAs on Outcome.

(A) EFS and (B) OS by chromosome 11q status

Multivariate Analysis of Survival

11q LOH was an independent predictor of OS, whether age was considered to be a continuous or categorical (<5 years vs ≥5 years) variable [HR= 3.116 (95% CI 1.034, 9.389; p=0.0435)]. Using age as a continuous variable, whether including 1p or 1p36, only age and any chromosome 9 aberration (HR= 3.797; 95% CI 1.417, 10.175; p=0.0080) remained independent predictors of EFS (Supplemental Figure 1). Using age as a categorical variable, whether including 1p or 1p36, only age and any chromosome 5 aberration (HR= 3.344; 95% CI 1.359, 8.230; p=0.0086) remained independent predictors of EFS (Supplemental Figure 2).

Impact of Gene Expression on Outcome

Of the 101 patients with available pre-treatment biospecimens, 47 had high-quality RNA available for RNA sequencing (42/47 of these subjects were also included in WES analyses). TERT expression, using a pnorm cutoff of 0.9, was not associated with EFS or OS. A previously published 157-gene expression signature predictive of outcome(26) was not associated with EFS or OS. NMF clustering of patients based on highly variable gene expression identified two clusters (cluster 1, n=35, cluster 2, n=11; Supplemental Figure 3). Differential gene expression and gene set enrichment analyses identified significant enrichment in cluster 2 for pathways involved in TNFα and NFKβ extracellular matrix and integrins, and epithelial to mesenchymal transition. In contrast, cluster 1 tumors were enriched in pathways involved in cell cycle regulation and MYC activation. These results led us to analyze enrichment of these clusters for previously published gene expression signatures of adrenergic and mesenchymal profiles in neuroblastoma(24). We observed significant enrichment for the mesenchymal signature in patients in cluster 2 (Supplemental Figure 3C). Tumors from patients in cluster 1 also showed a strong upregulation of EGR2 and SOX10 mediated initiation of Schwann cell myelination and significant overexpression of SOX10 and PLP1 genes (Figure 5). SOX10 and PLP1 are associated with Schwann Cell Precursors (SCP), but signatures associated with presence of SCP-like genes are associated with favorable outcome in children with non-MYCN amplified neuroblastoma(27, 28). Interestingly, samples from patients in cluster 2 were also highly enriched for the Schwann cell precursor signature (Supplemental Figure 3C) identified in previously studies(29). While the number of patients in cluster 2 was small (n=11), 5-year EFS and OS for this group was 100± 0.0% compared to 71.8±8.3% for patients in cluster 1 (p=0.0664; Figure 4), while 5-year OS were 100± 0.0% and 84.1±6.7%, respectively (p=0.1620).

Figure 5. RNA Sequencing Data.

(A) Top 50 differentially expressed genes by NMF Cluster (B) Top differentially expressed hallmark pathways by NMF cluster (C) EFS by NMF cluster (D) OS by NMF cluster

Figure 4. Impact of SCA count on EFS.

EFS by number of segmental chromosomal aberrations.

Discussion

Patients ≥18 months of age with INSS stage 3, MYCN–NA, UH neuroblastoma represent a rare subset of children with outcomes that are inferior to those of most patients with advanced locoregional disease, and superior to those of most high-risk patients when high-risk disease is defined by the presence of tumor MYCN amplification or metastatic disease. Our study represents the first comprehensive assessment of outcomes for this cohort of patients in the COG context, in which these patients were classified as having high-risk disease and received therapy appropriate for this risk designation. While the EFS for this cohort treated with high-risk therapy is numerically superior to EFS previously reported following less intensive therapy (12), a prospective trial to compare these approaches has not been performed. A limitation of this study is the use of the staging system used at the time of the conduct of the relevant trials (INSS) which relied on degree of surgical resection to assign stage. Current prospective trials utilize the International Neuroblastoma Risk Group Staging System (INRGSS), which uses diagnostic imaging and bone marrow assessments to assign stage. Unfortunately, the image-defined risk factors included in the INRGSS were not routinely collected on the trials relevant to this analysis, rendering the use of this contemporary staging system impossible.

Consistent with prior data in other neuroblastoma patient cohorts, SCAs were shown to adversely impact outcome in this cohort. A SIOPEN analysis revealed that the presence of any SCA was associated with an older age at diagnosis (27). In addition, the presence of any SCA was associated with inferior outcome, regardless of age(30). In this SIOPEN study, the number of SCAs was inversely correlated with OS. In patients ≥18 months with tumors that had no SCAs, 5-year OS was 100%, while children with tumors with ≥1 SCA had a 5-year OS of 61% (P=0.018)(30). Loss of chromosome 11q was the only SCA independently predictive of inferior outcomes in our cohort, consistent with findings in other clinical trials for neuroblastoma. In a retrospective analysis of high-risk patients treated by the SIOPEN, MYCN-A and 11q loss appeared to be mutually exclusive events, and patients with tumors harboring MYCN-A or 11q loss fared equally poorly, with an 8-year OS of ~35%(31). This is in stark contrast to the outcomes for high-risk patients whose tumors harbored numerical chromosomal aberrations or 17q gain, as 8-year OS for these patients were 100% and 60%, respectively(31). High-risk patients with tumors with 11q loss are also less likely to respond favorably to induction chemotherapy(32). The mechanism by which 11q loss mediates adverse outcomes in neuroblastoma is largely unknown, though our results suggest that 11q loss may be a marker of general chromosomal instability and chemoresistance. Proposed mechanisms underlying the impact of 11q loss on outcomes include loss of tumor suppressor genes (such as CADM1, ATM or H2AFX) or microRNAs (such as miR4301, miR125b-1, let-7a or miR100)(33). One potential strategy to improve outcomes for this cohort would be to uniformly classify patients ≥18 months with stage 3, MYCN–NA, UH and 11q loss as having high-risk disease and to include these patients in ongoing or future high-risk trials evaluating novel therapies.

As in many studies in neuroblastoma, we have also found response to therapy to be a powerful predictor of overall outcome(32, 34–36). Patients in this analysis who failed to achieve a partial response or better (in this cohort a reduction of > 50% in primary tumor volume, as patients by definition did not have metastatic disease) to 6 cycles of induction chemotherapy and an attempt at resection of their tumors fared dismally. While only 5 patients met this criterion, all of them had an event (2 local recurrences, 2 metastatic recurrences and 1 death as first event), and only 1 patient in this group was alive 5 years after diagnosis. Patients who do not achieve a post-induction PR or better may be candidates for novel approaches, such as those that are being evaluated in high-risk patients with first relapse or primary refractory disease.

Depending on the statistical model used, segmental chromosomal aberrations of chromosomes 5 and 9 were also found to predict inferior EFS, but not OS, in this group of patients. Previous analyses of the impact of loss of chromosomes 5 and 9 have yielded conflicting results. In a retrospective analysis of 24 tumors from patients of all stages and clinical risk groups who had tumors available in the Pediatric Oncology Group Neuroblastoma Tissue Bank, 9 were found to have allelic imbalance via loss of a segment of chromosome 5q (specifically the APC gene locus)(37). With a median follow up of 46 months, patients with tumors harboring a loss at this locus had more favorable outcomes (OS 100%) compared to patients with tumors with intact copy number (OS 47%, p=0.018)(37). In this cohort, 5q allelic imbalance was more common in tumors with normal MYCN copy number and a hyperdiploid DNA index, both previously identified as favorable biologic features in neuroblastoma(37). Similarly, in another analysis of 177 neuroblastoma tumors, LOH of chromosome 9p21–23 (including the tumor suppressors CDKN2A and CDKN2B) was associated with a significantly better OS in patients with Stage 4 disease (80% OS in LOH group vs 40% OS in retained heterozygosity group, p=0.0273)(38). In contrast, in a cohort of 80 patients with neuroblastoma of various stages, tumor 9p21 LOH was associated with inferior OS (p=0.023), while 9q34-qter LOH was not associated with a difference in OS (p=0.551)(39). Our findings warrant further validation in an independent cohort, as the SCAs on these chromosomes were not necessarily recurrent or localized to a single arm of the chromosome, and because the impact of chromosome 5 and 9 SCAs on EFS may be unique to older patients with localized, unfavorable histology and MYCN-non-amplified tumors.

RNA sequencing identified a subset of patients in this cohort with an extremely favorable outcome (n=11, 5-year EFS/OS 100± 0.0%/100± 0.0%). Surprisingly, these tumors were marked by low expression of genes associated with an adrenergic profile and high expression of genes associated with a mesenchymal profile and Schwann cell precursors. Previous gene expression profiles have identified two predominant super-enhancer states in neuroblastoma, the more undifferentiated and chemo-resistant mesenchymal state and the more differentiated and chemo-sensitive adrenergic state(24). Preclinical models have shown the adrenergic and mesenchymal states to be plastic and that they may cooperate to drive malignancy(40). The impact of these gene signatures has not been widely evaluated in clinical cohorts of patients treated for high-risk neuroblastoma, and these findings warrant validation in an independent cohort of similar patients. It is interesting to note that 9/11 patients in the low-adrenergic/high-mesenchymal group were found to have ganglioneuroblastoma, nodular (GNB-N) histology, and further assessment of of SCP-like signature in this cohort should be investigated for prognostic significance. However, the presence of GNB-N histology was not associated with differences in outcome in our ANBL00B1 cohort, and validation of the impact of this expression profile could potentially lead to the identification of patients with better outcomes for whom intensive treatment may not be warranted. However, given the small numbers (n=11) of patients in cluster 2 enriched in mesenchymal and Schwan cell precursor signature, caution should be used when interpreting these results until they can be replicated. Although we were able to evaluate for some telomere maintenance mechanisms (ATRX mutations, TERT expression) in this cohort, not all telomere maintenance mechanisms were evaluated due to the constraints of specimen collection in this retrospective cohort. Future studies should comprehensively evaluate the impact of telomere maintenance mechanisms on outcome in this patient group.

To achieve consensus on the optimal management of this rare group of patients, international collaborations to conduct prospective studies are required. In the SIOPEN LINES trial, patients in this group are treated with an intermediate strategy, taking elements from non-high-risk treatment (lower dose chemotherapy, surgery) and high-risk treatment (13-cis-RA, external beam radiotherapy). The results of this trial are awaited with great interest, as the role of intensive high-risk therapy in this cohort of patients is still not clear. Evaluating tumors using techniques including WES, RNAseq analyses, and telomere maintenance pathway studies in the equivalent LINES2 cohort may provide additional information and lead to identification of molecularly distinct cohorts, allowing for a molecularly based risk classification. The success of chemo-immunotherapy for high-risk neuroblastoma(41), and the detection of remarkable effects on bulky tumors suggest that chemo-immunotherapy may be a particularly attractive strategy in this group of patients. Evaluation of the effects chemo-immunotherapy rather than high-dose chemotherapy with stem cell rescue in this rare cohort could be evaluated in an international multi-center setting.

Statement of Translational Relevance.

Patients ≥18 months of age with stage 3, MYCN–non-amplified, unfavorable histology neuroblastoma represent a rare subset of children with outcomes that are inferior to those of most patients with advanced locoregional disease, and superior to those of most high-risk patients. Our study represents the first comprehensive assessment of outcomes for this cohort in the COG context. We evaluated the impact of clinical variables on outcome and used available tumor samples from subjects within this subgroup to correlate whole exome sequencing and RNA sequencing profiles with outcome. We found that subjects with tumors harboring chromosome 11q losses had inferior outcomes and describe an expression profile with universally favorable outcomes. These findings provide a framework for a refined risk stratification where patients with 11q loss could be included with traditional high-risk groups and patients with a favorable expression profile could be considered to have non-high-risk disease.