Validation of a prognostic multi‐gene signature in high‐risk neuroblastoma using the high throughput digital NanoString nCounter™ system

Abstract

Microarray‐based molecular signatures have not been widely integrated into neuroblastoma diagnostic classification systems due to the complexities of the assay and requirement for high‐quality RNA. New digital technologies that accurately quantify gene expression using RNA isolated from formalin‐fixed paraffin embedded (FFPE) tissues are now available. In this study, we describe the first use of a high‐throughput digital system to assay the expression of genes in an “ultra‐high risk” microarray classifier in FFPE high‐risk neuroblastoma tumors. Customized probes corresponding to the 42 genes in a published multi‐gene neuroblastoma signature were hybridized to RNA isolated from 107 FFPE high‐risk neuroblastoma samples using the NanoString nCounter™ Analysis System. For classification of each patient, the Pearson's correlation coefficient was calculated between the standardized nCounter™ data and the molecular signature from the microarray data. We demonstrate that the nCounter™ 42‐gene panel sub‐stratified the high‐risk cohort into two subsets with statistically significantly different overall survival (p = 0.0027) and event‐free survival (p = 0.028). In contrast, none of the established prognostic risk markers (age, stage, tumor histology, MYCN status, and ploidy) were significantly associated with survival. We conclude that the nCounter™ System can reproducibly quantify expression levels of signature genes in FFPE tumor samples. Validation of this microarray signature in our high‐risk patient cohort using a completely different technology emphasizes the prognostic relevance of this classifier. Prospective studies testing the prognostic value of molecular signatures in high‐risk neuroblastoma patients using FFPE tumor samples and the nCounter™ System are warranted.

Highlights

• Improved neuroblastoma risk classification is needed to refine treatment.

• Microarray signatures provide prognostic information for neuroblastoma patients.

• Complex technology has limited clinical integration of microarray signatures.

• Expression signatures can be obtained using nCounter™ system's simple technology.

Abbreviations

FFPE

formalin-fixed paraffin embedded

INRG

International Neuroblastoma, Risk Group

COG

Children's Oncology Group

H&E

hematoxylin and eosin

SD

standard deviations

EFS

event-free survival

OS

overall survival

SE

standard error

PCA

Principal Components Analysis

1. Introduction

Neuroblastoma, the most frequent extracranial solid tumor of childhood, exhibits a broad range of clinical behavior and varying response to treatment (Maris et al., 2007). To ensure that therapy is appropriately tailored, efforts to identify variables that accurately predict outcome have been ongoing for more than 35 years. Recently, an International Neuroblastoma Risk Group (INRG)⁴ Classification System has been established that utilizes 7 prognostic variables to stratify patients into very low‐, low‐, intermediate‐, or high‐risk cohorts (Cohn et al., 2009). Although outcome is excellent for non‐high‐risk patients, almost half of all patients are classified as high‐risk, and less than 50% will be cured with current treatment strategies. At the time of diagnosis, it is currently not possible to distinguish the subset of high‐risk patients who will achieve long‐term survival from those who will develop progressive disease.

The prognostic value of gene expression profiles has been recently evaluated using microarray technologies, and numerous low‐ and high‐risk neuroblastoma signatures have been discovered, tested, and validated (Abel et al., 2011; De Preter et al., 2010; Garcia et al., 2012; Schramm et al., 2005; Schulte et al., 2010; Vermeulen et al., 2009; Wei et al., 2004). Some of these signatures can sub‐stratify patients with high‐risk neuroblastoma, identifying a subset of high‐risk patients with significantly worse survival, or an “ultra‐high‐risk” group (Abel et al., 2011; Asgharzadeh et al., 2006; De Preter et al., 2010; Garcia et al., 2012). However, incorporating molecular signatures into diagnostic classification systems has proven to be difficult because microarray analysis requires frozen tumor tissue to yield high quality RNA, and these samples are rarely available. The microarray‐based assays are also expensive and require complex statistical analyses. The nCounter™ System overcomes these limitations. This system utilizes simple technology that captures and counts individual mRNA transcripts with high levels of sensitivity, linearity, multiplex capability, and digital readout. Importantly, different sample types, including highly degraded RNA from FFPE material, can be utilized (Fernandez et al., 2012; Kulkarni, 2011; Payton et al., 2009; Stricker et al., 2013), and recent studies using the nCounter™ System have validated expression signatures obtained using microarray technology (Kulkarni, 2011; Malkov et al., 2009; Northcott et al., 2012).

Here we describe the first use of a high‐throughput digital system to assay the expression of a published “ultra‐high risk” neuroblastoma molecular classifier (De Preter et al., 2010). The 42‐gene signature was built from a meta‐analysis of neuroblastoma studies performed using oligonucleotide microarray platforms. We show that the prognostic value of this microarray‐derived gene signature was validated in a cohort of 107 high‐risk neuroblastoma patients using nCounter™ technology and RNA isolated from FFPE high‐risk neuroblastoma tumor samples.

2. Materials and methods

2.1. Tumor samples

Neuroblastoma patients from Comer Children's Hospital, St. Jude Children's Research Hospital, Ann & Robert H. Lurie Children's Hospital of Chicago, Texas Children's Hospital, Seattle Children's Hospital, and Children's Hospital Los Angeles were diagnosed from 1993 to 2011 and classified as high‐risk based on Children's Oncology Group (COG) criteria (Maris, 2010). Medical records were abstracted at each institution and clinical data including age at diagnosis, stage, MYCN status, ploidy, histology, and outcome were collected. The patients were staged according to the International Neuroblastoma Staging System (Brodeur et al., 1993) and tumor histology was defined as favorable or unfavorable using the International Neuroblastoma Pathologic Classification (INPC) System (Shimada et al., 1999). A hematoxylin and eosin (H&E) stained section and two to four FFPE scrolls of diagnostic tumor tissue were sent to the University of Chicago. The stained tissue sections were reviewed by a pathologist (PP), and the percentage of viable and necrotic tumor, lymphoid infiltrates and other tissue elements including connective tissue and stroma were assessed. Only cases with >50% viable neuroblastoma tumor cells were selected for expression profiling. This study was approved by the Institutional Review Board at the University of Chicago and at each of the collaborating institutions.

2.2. RNA isolation

RNA was isolated using the RNeasy^® FFPE kit (Qiagen, Valencia, CA) from two 10‐μm sections from each sample. RNA concentration was quantified using UV spectroscopy (Nanodrop Technologies, Wilmington, DE) and integrity assessed using a Bioanalyzer 2100 and RNA Nano Chip assay (Agilent Technologies, Wilmington, DE).

2.3. Codeset design and expression quantification

Details of the nCounter™ technology (NanoString Technologies, Seattle, WA) have been reported previously (Geiss et al., 2008; Kulkarni, 2011). Briefly, NanoString designed and manufactured customized probes corresponding to the 42 genes in a previously reported prognostic signature (De Preter et al., 2010) (Table 1). A codeset specific to a 100‐base region of the target mRNA was designed using a 3′ biotinylated capture probe and a 5′ reporter probe tagged with a specific fluorescent barcode; creating two sequence‐specific probes for each target transcript. Probes were hybridized to 100 ng of total RNA for 19 h at 65 °C and applied to the nCounter™ Preparation Station for automated removal of excess probe and immobilization of probe‐transcript complexes on a streptavidin‐coated cartridge. Data were collected using the nCounter™ Digital Analyzer by counting the individual barcodes.

Table 1.

Sequence‐specific probes constructed for the analysis of 107 high‐risk neuroblastoma tumors using the nCounter™ System.

De Preter et al.	Refseq ID	Targeted region	Reference genes	Refseq ID	Targeted region
AHCY	NM_000687.2	1805–1905	ACSM3	NM_005622.3	955–1055
AKR1C1	NM_001353.5	1210–1310	BRD4	NM_014299.2	745–845
ARHGEF7	NM_001113511.1	1275–1375	CLTC	NM_004859.2	290–390
BIRC5	NM_001012270.1	1095–1195	CRYBB3	NM_004076.3	230–330
CADM1	NM_001098517.1	1071–1171	DHX57	NM_198963.1	3888–3988
CAMTA2	NM_015099.2	1573–1673	EIF5A2	NM_020390.5	450–550
CDCA5	NM_080668.3	320–420	GAPDH	NM_002046.3	35–135
CDKN3	NM_001130851.1	390–490	GNAI2	NM_002070.2	1000–1100
CLSTN1	NM_001009566.1	935–1035	GUSB	NM_000181.1	1350–1450
DDC	NM_000790.3	1195–1295	HDAC3	NM_003883.2	1455–1555
DPYSL3	NM_001387.2	4790–4890	HPRT1	NM_000194.1	240–340
ECEL1	NM_004826.2	2443–2543	N4BP1	NM_153029.3	2495–2595
EPB41L3	NM_012307.2	2455–2555	OAZ1	NM_004152.2	313–413
EPHA5	NM_004439.4	2855–2955	PGK1	NM_000291.2	1030–1130
EPN2	NM_014964.4	285–385	PTK6	NM_005975.2	1715–1815
FYN	NM_002037.3	765–865	RPL23A	NM_000984.5	611–711
GNB1	NM_002074.2	2165–2265	RPS6KB2	NM_003952.2	980–1080
HIVEP2	NM_006734.3	5715–5815	SDHA	NM_004168.1	230–330
INPP1	NM_001128928.1	1515–1615	SUFU	NM_016169.2	4380–4480
MAP7	NM_003980.3	1085–1185	TRAPPC3	NM_014408.3	212–312
MAPT	NM_001123066.2	5605–5705	TUBB	NM_178014.2	320–420
MCM2	NM_004526.2	2945–3045	UBC	NM_021009.3	1875–1975
MRPL3	NM_007208.2	350–450
MYCN	NM_005378.4	1545–1645
NCAN	NM_004386.2	4330–4430
NME1	NM_000269.2	500–600
NRCAM	NM_001037132.1	4955–5055
NTRK1	NM_001012331.1	1365–1465
ODC1	NM_002539.1	950–1050
PAICS	NM_001079524.1	2604–2704
PLAGL1	NM_001080951.1	64–164
PMP22	NM_000304.2	1644–1744
PRKACB	NM_002731.2	835–935
PRKCZ	NM_002744.4	771–871
PTN	NM_002825.5	80–180
PTPRN2	NM_002847.3	2865–2965
SCG2	NM_003469.3	1400–1500
SLC24A5	NM_205850.2	410–510
SNAPC1	NM_003082.3	143–243
TYMS	NM_001071.1	555–655
ULK2	NM_014683.2	2340–2440
WSB1	NM_015626.8	1410–1510

2.4. Data processing and class prediction analysis

Each codeset included probes for the 42‐gene signature, spiked‐in External RNA Control Consortium positive and negative controls, and reference housekeeping genes (Table 1). Housekeeping genes were selected by analyzing published neuroblastoma microarray datasets (Asgharzadeh et al., 2006; Oberthuer et al., 2010; Wang et al., 2006), binning genes into low‐ medium‐ and highly‐expressed, and then selecting 3 genes for each category with minimal variance across samples.

Background hybridization was determined using spiked‐in negative controls. All signals below mean background plus 2 standard deviations (SD) were considered to be below the limits of detection, and set to mean background. A normalization factor was calculated from the spiked in exogenous positive controls in each sample and applied to the raw counts from the nCounter™ output data. Then, a content normalization factor was calculated from the Geomean of the reference genes and applied to the data previously normalized by the positive control. Probesets were produced in two batches, and several samples were run with both sets of probes to generate a per gene batch correction factor that was applied across the entire data set. Each sample was run in duplicate, and for most analyses, the mean of the sample pairs was used.

2.5. Statistical analysis

The methods of Kaplan–Meier (Kaplan and Meier, 1958) were used to generate event‐free survival (EFS) and overall survival (OS) curves for the entire cohort and for subgroups stratified by established risk factors using Prism 5 (GraphPad Software, Inc., La Jolla, CA). The curves were compared using a log‐rank test. Five‐year EFS and OS rates were expressed as the estimate ± the standard error (SE).

To test the prognostic signature described by De Preter and colleagues, cross‐platform data for the 42 genes from the 250 patients in their training set were obtained (De Preter et al., 2010). Following methods reported by these investigators (De Preter et al., 2010), a prognostic vector was created by subtracting the mean expression value in patients with EFS of at least 1000 days from the mean expression value from the patients who died of disease for each of the 42 genes. For classification of each of the 107 individuals in our study, the Spearman's correlation coefficient was calculated between the log transformed, normalized nCounter™ data, and the vector from the De Preter training data. Patients with a negative correlation coefficient were predicted to have low‐molecular risk and a favorable prognosis, whereas patients with a positive correlation had high‐molecular risk and an unfavorable prognosis. Survival functions (low versus high‐molecular risk) were compared with the log‐rank test using the survival package in R. Heatmaps were generated using Heatmap.2, and Principal Components Analysis (PCA) was performed using the pcaMethods packages in R. MYCN targets were identified by taking the union of the following gene sets in MdbSig3.0: MYC_amplified_leukemia, MYC_amplified, MYC_targets_up, and MYC_targets_down.

3. Results

3.1. Analytic cohort of 107 high‐risk neuroblastoma patients

A total of 138 high‐risk patients with available paraffin‐embedded material were identified at the 6 participating institutions. Review of the medical records revealed that 5 cases were misclassified as high‐risk. Tumor tissue was not available in 5 cases, as the diagnosis was made using bone marrow samples. In 10 cases, tumor cellularity was <50%. RNA was isolated from 118 tumors. For 11 samples the data quality was poor as determined by correction factors >10, >20% of the genes were expressed below the threshold, and/or overall means were 3 or more SD above or below the mean for all samples following correction and normalization. Although many of samples with poor data quality were needle‐core biopsies, we were able to generate high quality expression data from 8 core needle biopsy samples. Thus, the size of the core needle biopsies and the amount of viable tumor in these samples are important determinants of data quality. We were able to generate high quality expression data for 107/118 cases (90.7%), using FFPE material, and these patients comprised the analytic cohort for this study.

Patients ranged in age from 3 days to 184 months, with a mean of 38 months. Eighty‐three (77.6%) patients were older than 18 months (Table 2). Eighty‐seven (81.3%) patients had stage 4 disease, and MYCN amplification was detected in tumors from 61 patients (57%). All patients were either enrolled on or treated according to a cooperative group clinical trial or institutional high‐risk neuroblastoma protocols (Furman et al., 2012; Kletzel et al., 2002; Kreissman et al., 2013; Matthay et al., 2009; Santana et al., 2005; Wagner et al., 2004). Five‐year EFS and OS rates ±SE for the entire cohort were 43% ± 5.2% and 52% ± 5.3%, respectively (Figure 1). Age at diagnosis, stage of disease, tumor histology, MYCN status, and ploidy were not statistically significantly predictive of outcome (Table 2).

Table 2.

Clinical and biological characteristics of patient cohort.

Characteristic	No	%	5 year EFS +/− SE	P value	5‐year OS +/− SE	P value
Age
≤18 months	24	22.4	35.7 +/− 10.1	0.0761	48.6 +/− 10.5	0.3985
>18 months	83	77.6	44.9 +/− 6.0		52.8 +/− 6.1
INSS stage
2,3	20	18.7	50.5 +/− 12	0.5448	49.8 +/− 12.0	0.7994
4	87	81.3	40.8 +/− 5.7		52.4 +/− 5.9
Histologya
Favorable	10	9.3	33.3 +/− 17.2	0.3083	45.0 +/− 17.4	0.4633
Unfavorable	72	67.3	40.8 +/− 6.4		49.4 +/− 6.5
Unknown	25	23.4	49.6 +/− 10.3		61 +/− 0.1
MYCN Status
Unamplified	38	35.5	46.5 +/− 8.7	0.2901	54.0 +/− 8.8	0.6345
Amplified	61	57.0	42.6 +/− 6.8		53.0 +/− 6.9
Unknown	8	7.5	30.0 +/− 17.5		36 +/− 0.2
Ploidy
Hyperdiploid	45	42.1	44.4 +/− 7.9	0.3956	52.4 +/− 8.1	0.0923
Diploid	36	33.6	33.1 +/− 8.9		42.6 +/− 9.1
Unknown	26	24.3	51.7 +/− 10.6		63.6 +/− 10.3

^aInternational Neuroblastoma Pathologic Classification (Shimada et al., 1999).

Figure 1.

Kaplan–Meier curves of OS and EFS of the entire cohort of patients with high‐risk neuroblastoma (n = 107).

3.2. Integrity of RNA from FFPE blocks is sufficient for nCounter™ analysis

The integrity of the RNA determined by the Bioanalyzer varied from a RIN of 1.2–2.8, confirming that the RNA purified from FFPE blocks is highly degraded. Although RNA with RINs of 7.5 or greater is generally recommended for microarray experiments, chains that are at least 100 base pairs long were suitable for hybridization with the capture and reporter probes using the nCounter™ system.

3.3. Gene expression cluster analysis shows correlation with MYCN status

The nCounter™ measurements were highly reproducible. Pearson correlation coefficients were calculated for each of the duplicate samples and ranged between 0.925 and 0.997 (mean of 0.98) (Figure 2A). Furthermore, a linear regression was performed on all pairs, and the R‐squared values ranged between 0.855 and 0.995 with a mean of 0.96 (Figure 2B). Unsupervised clustering of the 42 genes tested within our cohort showed that samples clustered into three groups, largely defined by expression of MYCN (Figure 3A). The MYCN high expressing group consisted of cases with MYCN amplified tumors, but there were some non‐amplified cases in this group as well. Interestingly, although these tumors clustered with the MYCN amplified tumors, their MYCN expression levels were lower than that observed in the amplified tumors, indicating that they are similar to amplified tumors on the basis of other genes in the signature. The other two groups had lower levels of expression of MYCN. Of note, while these were mostly non‐amplified tumors, there were some amplified tumors in these clusters that, while they expressed MYCN, they did so at levels closer to non‐amplified tumors. The heat map demonstrates that duplicate samples were highly correlated; for every sample, the most closely related sample was the replicate (Figure 3A). Thus, the mean expression level for each replicate was calculated and used in all subsequent experiments.

Figure 2.

A) Boxplots of Pearson Correlation Coefficients for all sample duplicates show that all pairs are highly correlated. B) Boxplot of the R‐squared for all duplicates derived from a linear regression shows that all duplicates are highly correlated. C) Unsupervised hierarchical clustering of the mean of duplicates for all 107 samples by expression levels of 42 genes as measured by the nCounter™ System. The bar at the top denotes MYCN amplification status, with red = amplified, blue = non‐amplified, and gray = unknown. D) Histogram plotting the MYCN expression levels for each sample. Samples clearly segregate into low MYCN and high MYCN expression.

Figure 3.

Unsupervised clustering of 42‐genes shows a strong association with MYCN amplification status. A) Unsupervised hierarchical clustering of all independent samples by expression levels of the 42‐genes in the high‐risk signature. The bar at the top denotes MYCN amplification status, with red = amplified, blue = non‐amplified, and gray = unknown. B) First 2 principal components in a principal components analysis, based on the mean level of expression of 42 genes in 107 samples. Color denotes MYCN amplification status as above. C) Boxplots of gene expression levels for known MYC targets ODC1, PAICS, and NME1 show that they are strongly differentially expressed between MYCN amplified versus non‐amplified cases. D) Boxplots of gene expression levels for genes in the 42‐gene signature that are not associated with MYCN expression or MYCN status.

A heat map of the mean values showed similar segregation of tumors by MYCN expression ( Figure 2C). PCA of the expression levels for 42 genes also demonstrated a strong separation between samples with versus without MYCN amplification ( Figure 3B). MYCN gene expression clearly segregated into high and low expression groups, and the level of expression correlated strongly with MYCN amplification status ( Figure 2D). Eighty‐nine percent (54/61) of MYCN amplified cases also had high expression of MYCN. Likewise, of the samples with high MYCN expression, only 4 did not have MYCN amplification. In addition, high MYCN gene expression correlated with expression of known MYC target genes (i.e. ODC1, PAICS, NME1) (Ben‐Yosef et al., 1998) (Figure 3C). De Preter and colleagues noted that 12 of the 42 genes in the signature are known targets of MYCN (De Preter et al., 2010). We found that expression of all 12 of these genes strongly correlated with MYCN status in our cohort. To specifically determine the correlation between each of the genes in the signature and MYCN status, differentially expressed genes in the MYCN amplified versus non‐amplified cases were quantified via linear model, followed by correction for multiple hypothesis testing using Storey's q‐value method (Storey, 2002). Twenty‐eight of the 42 genes in the signature were statistically significantly differentially expressed in the MYCN amplified cases compared to the non‐amplified tumors, including the known MYC targets CAMTA1, NME1, and NTRK1. However, 14 additional genes, including PLAGL1, AKR1C1, ARHGEF7, CADM1, CAMTA2, EPB41L3, and FYN, showed little correlation with MYCN status (Figure 3C and D).

3.4. The multi‐gene expression signature analyzed with the nCounter™ system sub‐stratifies high‐risk neuroblastoma

The nCounter™ 42‐gene panel sub‐stratified the cohort of 107 high‐risk patients into two groups with statistically significantly different outcome, replicating the results of De Preter et al. (2010) Patients with low‐ versus high‐molecular risk tumors had OS rates of 57.3% ± 7.8% versus 36.7% ± 7.7%, respectively (p = 0.0027) (Figure 4A), and EFS of 48.8% ± 7.7% versus 29.3% ± 7.4% (p = 0.028) (Figure 4B). The 42‐gene signature was also prognostic within the subset of 38 high‐risk patients with non‐MYCN amplified tumors, with OS of 55.8% ± 10% versus 28.6% ± 17% (p = 0.0056) for patients with low vs high‐risk molecular tumors (Figure 4C), and EFS of 54.6% ± 9.8% versus 14.3% ± 13.2%, respectively (p = 0.011) (Figure 4D).

Figure 4.

Kaplan–Meier curves of OS (A) and EFS (B) for 107 high‐risk neuroblastoma patients classified using the 42‐gene prognostic signature analyzed using the nCounter™ System. OS and EFS were statistically significantly worse for patients with high‐molecular risk neuroblastomas compared to those with low‐molecular risk tumors (p = 0.0027 and p = 0.028, respectively). Kaplan–Meier curves of OS (C) and EFS (D) for non‐MYCN amplified cases. OS and EFS were statistically significantly worse for patients with high‐molecular risk neuroblastomas compared to those with low‐molecular risk tumors (p = 0.0056 and p = 0.011, respectively).

4. Discussion

In this study, we validated the prognostic significance of a published neuroblastoma microarray signature that was previously shown to sub‐stratify high‐risk patients into cohorts with significantly different survival (De Preter et al., 2010) using high throughput digital technology and RNA isolated from 107 FFPE high‐risk tumors. The cohort of high‐risk patients with worse outcome is considered at “ultra‐high risk” for relapse. The nCounter™ system requires very small amounts of RNA, and reliable results can be obtained even if the RNA is highly degraded. This technology has numerous advantages over microarray methodology including the ability to analyze gene expression using FFPE tumor tissue as starting material. The workflow does not involve enzymatic and amplification steps, thus reducing the risk of additional technical errors and the introduction of bias in the results. Furthermore, this digital technology is less time consuming, requires only a simple data analysis and normalization strategy, and is less expensive and more reproducible (Malkov et al., 2009).

The multi‐gene signature analyzed in our study originated from an analysis of 933 neuroblastoma patients from eight independent studies (De Preter et al., 2010). This molecular classifier was established based on re‐analysis of published studies with updated clinical information, re‐annotation of the probe sequences, common risk definitions for training cases, and a single method for gene selection (prediction analysis of microarray) and classification (correlation analysis). Using new digital technology, we replicated the findings of De Preter et al. (2010) and show that the 42‐gene classifier sub‐stratified high‐risk neuroblastoma patients into subsets with significantly different outcome. The signature was also prognostic within the subset of patients with tumors that lacked MYCN amplification. In contrast, none of the other established prognostic markers was associated with outcome.

Interestingly, although 28 of the 42 genes were statistically differentially expressed between MYCN amplified and non‐amplified cases, the signature determined prognosis in the whole cohort as well as the non‐MYCN amplified group. In fact, this signature was prognostic while MYCN status was not in our cohort. It is important to note that not all genes are associated with MYCN status, and thus this signature doesn't just represent MYCN copy number status. Indeed, it can be seen from the heatmap that while amplified and non‐amplified tumors mostly cluster separately, there are non‐amplified tumors clustering with the amplified tumors and vice versa. Future studies will explore the hypothesis that the expression signature, rather than MYCN amplification status, is a more accurate measurement of MYCN pathway activation, and thus, a better discriminator for outcome.

Despite the prognostic value of this 42‐gene ultra‐high risk signature (De Preter et al., 2010), it has not been prospectively studied or integrated into a neuroblastoma classification system, due in part to the requirement for frozen tumor and high quality RNA isolation for the microarray assay. Although frozen tumors are not readily available, the vast majority of patients with high‐risk neuroblastoma have FFPE tumor samples. RNA extracted from FFPE material is typically highly degraded, and therefore of limited use in other molecular studies. The nCounter™ technology overcomes this obstacle by using relatively short 50mer probes that will hybridize efficiently with RNA chains as short as 100 base pairs (each target gene hybridizes with one capture and one reporter probe). As we show, the highly degraded RNA from FFPE tumor samples is sufficient to obtain strong gene expression signals that are highly reproducible.

The ability to identify patients who will not adequately respond to standard high‐risk treatment strategies in the diagnostic setting is critical for informed treatment decisions. Prospective clinical trials will be needed to identify the most powerful prognostic classifiers for high‐risk neuroblastoma patients. Using the nCounter™ System, data can be generated in ∼24 h and analytic results can be obtained within a few weeks. Thus, this technology is well suited for evaluating expression signatures in newly diagnosed patients. With more precise prognostication, “ultra‐high‐risk” patients could be spared unnecessary acute and long‐term toxicities associated with current, intensive multi‐modality therapy approaches. Transitioning patients who are not likely to survive with standard therapy to alternative treatments and/or early phase clinical studies may provide the clinical benefit of more targeted, less toxic treatments and improved quality of life.

Conflict of interest

Kirsteen H. Maclean is a NanoString Technologies employee.

Stricker Thomas P., Morales La Madrid Andres, Chlenski Alexandre, Guerrero Lisa, Salwen Helen R., Gosiengfiao Yasmin, Perlman Elizabeth J., Furman Wayne, Bahrami Armita, Shohet Jason M., Zage Peter E., Hicks M. John, Shimada Hiroyuki, Suganuma Rie, Park Julie R., So Sara, London Wendy B., Pytel Peter, Maclean Kirsteen H. and Cohn Susan L., (2014), Validation of a prognostic multi‐gene signature in high‐risk neuroblastoma using the high throughput digital NanoString nCounter™ system, Molecular Oncology, 8, doi: 10.1016/j.molonc.2014.01.010.