|Home | About | Journals | Submit | Contact Us | Français|
Children diagnosed at age ≥ 18 months with metastatic MYCN-nonamplified neuroblastoma (NBL-NA) are at high risk for disease relapse, whereas those diagnosed at age < 18 months are nearly always cured. In this study, we investigated the hypothesis that expression of genes related to tumor-associated inflammatory cells correlates with the observed differences in survival by age at diagnosis and contributes to a prognostic signature.
Tumor-associated macrophages (TAMs) in localized and metastatic neuroblastomas (n = 71) were assessed by immunohistochemistry. Expression of 44 genes representing tumor and inflammatory cells was quantified in 133 metastatic NBL-NAs to assess age-dependent expression and to develop a logistic regression model to provide low- and high-risk scores for predicting progression-free survival (PFS). Tumors from high-risk patients enrolled onto two additional studies (n = 91) served as independent validation cohorts.
Metastatic neuroblastomas had higher infiltration of TAMs than locoregional tumors, and metastatic tumors diagnosed in patients at age ≥ 18 months had higher expression of inflammation-related genes than those in patients diagnosed at age < 18 months. Expression of genes representing TAMs (CD33/CD16/IL6R/IL10/FCGR3) contributed to 25% of the accuracy of a novel 14-gene tumor classification score. PFS at 5 years for children diagnosed at age ≥ 18 months with NBL-NA with a low- versus high-risk score was 47% versus 12%, 57% versus 8%, and 50% versus 20% in three independent clinical trials, respectively.
These data suggest that interactions between tumor and inflammatory cells may contribute to the clinical metastatic neuroblastoma phenotype, improve prognostication, and reveal novel therapeutic targets.
The concept of tumor-promoting inflammation is a recognized enabling characteristic of cancers.1 Recent studies have demonstrated the prognostic significance of tumor-associated macrophages (TAMs) in some adult cancers, including Hodgkin's lymphoma and breast cancer.2–6 However, the prognostic significance of tumor-associated inflammatory cells in metastatic disease and in childhood cancers is unknown.
Neuroblastoma, an embryonal tumor of the sympathetic nervous system, is one of the most common solid tumors in children, with approximately 40% of patients presenting with metastatic disease at diagnosis.7 Among those with metastatic disease, amplification of the MYCN proto-oncogene (30% of tumors) is associated with high risk of disease relapse, whereas those lacking MYCN amplification (NBL-NA) have clinical behaviors that are distinctly associated with age at diagnosis.8–10 Patients diagnosed with metastatic NBL-NA at age ≥ 18 months often have tumors with recurrent segmental genomic alterations and have high-risk disease with only 45% long-term disease-free survival.10–16 In contrast, children diagnosed with metastatic NBL-NA at age < 18 months of age frequently have tumors with whole chromosomal alterations and have greater than 90% overall survival (OS) after receiving only moderate-intensity chemotherapy.17–20 Biologic mechanisms responsible for the age-dependent genomic and clinical phenotypes of metastatic NBL-NA and for different responses to treatment among those age ≥18 months at diagnosis have been unclear.
Our previous gene expression profiling study of metastatic NBL-NA tumors suggested that there may be age-dependent differences in expression of genes representing tumor-associated inflammatory cells.21 In the current study, we focused on intratumor inflammatory cells, especially TAMs, and their relationship with clinical behavior of metastatic NBL-NA. We examined the infiltration of macrophages in locoregional and metastatic tumors with immunohistochemistry (IHC). We also used a TaqMan low-density array (TLDA; Life Technologies, Carlsbad, CA) assay to assess expression of inflammatory and tumor cell–related genes in metastatic NBL-NA tumors diagnosed before and after 18 months of age. Our findings provide new insights about intratumor inflammation in metastatic NBL-NA tumors and provide the basis for constructing a novel 14-gene model that predicts risk of disease progression in those diagnosed at age ≥ 18 months.
Tissue microarray (TMA) details are provided in the Data Supplement. Macrophages were identified using IHC analysis of primary neuroblastoma tissues using antibodies directed against CD163 and allograft inflammatory factor 1 (AIF1). Tissue section scores ranged from 0 to 7 for each marker, with higher scores indicating a greater proportion of positive cells.
The details of the 48-gene TLDA assay are provided in the Data Supplement. All patients included in the gene expression study had metastatic NBL-NA tumors and were enrolled onto Children's Cancer Group (CCG), German Society for Pediatric Oncology-Hematology (GPOH), or Children's Oncology Group (COG) trials at diagnosis (Table 1 and Data Supplement). Details of treatment for the three cohorts were described previously and are provided in the Data Supplement.12,14,16,21 Informed consent was obtained in accordance with institutional review board policies.
The Data Supplement illustrates the flow of statistical and validation methods used in this report. Because our primary interest was to identify genes that are predictive of outcome in the cohort of patients older than 18 months of age, we first conducted a univarite logistic regression model based on TLDA gene expression data from 133 samples from the training cohort (CCG), which includes patients older and younger than 18 months of age at the time of diagnosis. Genes that were independent of age at diagnosis with a P value of ≤ .25 were included in a final multivariate logistic model to predict PFS. Our aim was to build a robust model that was predictive of disease progression in patients older than 18 months of age and that could be used as the basis for classification into signature-based low- and high-risk tumor-progression groups. Disease progression was defined a priori (Data Supplement). The effective period for risk of disease progression in the training cohort was 4 years from diagnosis. Because few patients were censored before the end of the effective period for risk of disease progression, ignoring this censoring had little practical effect on the logistic regression analysis of whether disease progression had occurred. Age was included as a continuous covariate in the final multivariate logistic regression analysis to assess for residual significance. Logit values, representing the tumor-progression scores, were computed for each patient. Measures of accuracy based on resubstitution analysis and leave-one-out cross validation (LOOCV) are presented. Classification accuracy was assessed using receiver operating characteristic curves and areas under the curve (AUC). External validation of the prediction model was performed using the independent GPOH and COG samples, with the tumor-progression score for each patient calculated using the regression coefficients from the prediction model derived from the training cohort. The relative contributions to the accuracy of the 14-gene NBL-NA prediction score of age at diagnosis, tumor cell–related genes, and inflammation-related genes were assessed using 5,000 permutations of the data set (Data Supplement).
Tumor-progression risk scores obtained from the multivariate logistic regression model were used to define signature-based risk groups. The median tumor-progression risk score from the training cohort (n = 133) was used as the cutoff point to define signature-based high-risk (tumor-progression score ≥ median score) or low-risk (tumor-progression score < median score) scores.
Details of the statistical analyses for developing the prognostic score are described here and summarized in the Data Supplement. In addition, survival analysis methods22 are used to describe outcome in low- and high-risk groups defined by the prognostic score. The primary end point for these analyses was progression-free survival (PFS), defined as the minimum interval from date of diagnosis to date of disease progression, date of death (four patients only), or date of last follow-up. Patients who did not experience progression or die were censored at the time of last follow-up. The Kaplan-Meier method was used to compute PFS probabilities and produce survival curves. CIs are based on Greenwood SEs. Unless otherwise stated, the reported probabilities are based on 5-year PFS rates. Tests of the difference in PFS between risk groups are based on the log-rank statistic. Other common statistics23 (eg, t test, Spearman rank correlation) were used where appropriate and are indicated in the text. Bonferroni adjustments to account for multiple comparisons were used where appropriate. Statistical computations were performed using STATA software (version 9.0; STATA, College Station, TX) or the R project (http://www.r-project.org).
We performed IHC analysis of 71 neuroblastoma tumors (29 patients with locoregional, 31 with metastatic disease [stage 4], and 11 with metastatic disease with special designation [stage 4S]; Data Supplement) using antibodies directed against two macrophage markers (CD163 and AIF1). There were significantly greater numbers of infiltrating macrophages observed only with CD163 staining, which identifies alternatively activated macrophage (M2), in samples of patients with metastatic (stage 4) compared with locoregional neuroblastoma (t test P = .003; Bonferroni-adjusted P < .017 considered significant; Fig 1A). There was no statistically significant difference in the number of intratumor CD163+ macrophages between metastatic tumors with special designation (stage 4S), which undergo spontaneous regression, and locoregional tumors (Fig 1B).
We also performed gene expression analysis of 133 metastatic NBL-NA tumors (CCG cohort: 94 children diagnosed at age ≥ 18 months and 39 diagnosed at age < 18 months; Table 1; Data Supplement) with a custom-built TLDA containing 31 tumor-related and 13 inflammation-related genes (Data Supplement). We identified greater expression of inflammation-related genes associated with monocyte/macrophage, myeloid, and B cells in tumors of children diagnosed at age ≥ 18 months compared with those diagnosed at age < 18 months (Fig 1C). Although inflammation-related genes CD33, FCGR3 (CD16), and IGKC showed significant association with PFS in univariate analysis (Data Supplement), we did not identify any single gene model that could accurately predict PFS in children diagnosed at age ≥ 18 months with AUC > 0.7. These data suggest that inflammatory cells within tumors, especially TAMs, contribute to the age-associated clinical behavior of metastatic NBL-NA tumors.
We further examined the expression of the 31 tumor-related and 13 inflammation-related genes in the CCG cohort and identified 14 genes that contributed to a model predictive of PFS (Table 2). Among the 14 genes used in our model, nine (64%) were tumor cell related, and five (36%) were inflammation related. The accuracy of the model for predicting PFS using LOOCV AUC estimates was 0.82 for patients in all age groups and 0.74 for patients age ≥ 18 months at diagnosis (Data Supplement).
Tumors from the CCG cohort were categorized as low or high risk based on their 14-gene tumor-progression risk score using LOOCV analysis. Figures 2A and and2B2B show that patients with a low-risk score had significantly better PFS (72% at 5 years; 95% CI, 60% to 82%) than those in the high-risk score group (16% at 5 years; 95% CI, 8% to 26%) using LOOCV analysis (P < .001). The overall 5-year PFS for the 94 patients who were age ≥ 18 months at diagnosis and treated on CCG high-risk protocols was 23%. Among these patients, 30 (32%) had a low-risk score with a 5-year PFS rate of 47% (95% CI, 28% to 63%), and 64 (68%) had a high-risk score with a 5-year PFS of 12% (95% CI, 5% to 22%; P = .002), demonstrating that a subset of patients at extremely high risk of disease progression can be identified by the 14-gene signature among these otherwise clinically indistinguishable patients. Five-year OS for the 94 patients whose tumors had low- or high-risk scores also was significantly different (60% v 23%; P = .003; Data Supplement). Classification by resubstitution analysis showed higher accuracy in prediction and more divergent Kaplan-Meier curves (Data Supplement), reflecting the optimistically biased approach of this analysis and the need for cross validation.24–28
Independent validation of the 14-gene signature to predict PFS was obtained from analysis of metastatic NBL-NA tumors from two independent cohorts of patients (GPOH, n = 39; COG, n = 52) diagnosed at age ≥ 18 months (Table 1; Data Supplement). Those patients whose tumors had signature-based high-risk scores had significantly worse 5-year PFS than those with low-risk scores (Figs 2C and and2D).2D). Among the 39 GPOH patients, 21 (54%) had low-risk scores with a 5-year PFS rate of 57% (95% CI, 34% to 75%), and 18 (46%) had high-risk scores with a 5-year PFS of 8% (95% CI, 1% to 29%; P = .002). Nineteen COG patients (36%) had low-risk scores with a 5-year PFS rate of 50% (95% CI, 26% to 70%), and 33 (64%) had high-risk scores with a 5-year PFS of 20% (95% CI, 8% to 35%; P = .009). Five-year OS for patients in these two cohorts whose tumors had low- or high-risk scores also was significantly different (GPOH: 65% v 31%; P = .012; COG: 51% v 31%; P = .039; Data Supplement). These data show the validity of the 14-gene signature and the prognostic information obtained from inclusion of inflammation-related genes to identify subsets of patients with different outcomes in a clinically indistinguishable population.
The five inflammation-related genes in our 14-gene model included CD14, CD33, FCGR3 (CD16), interleukin-6 receptor (IL6R), and interleukin-10 (IL10), which are mainly expressed by macrophages and myeloid cells and, along with CD163, signify intratumor macrophage polarization to the anti-inflammatory M2-like phenotype.28,29 Our previous research demonstrated that expression of these inflammation-related genes in neuroblastoma tumors correlates with microscopic presence of IL6-producing CD68+ cells, which are considered to be TAMs.30 Comparison of levels of expression of inflammation-related genes in neuroblastoma tumors with five neuroblastoma cell lines demonstrated that these markers are expressed 150-fold (range, five to 309; t test P < .001) higher on average in tumors than in cell lines (Fig 3B), suggesting that these genes are primarily expressed by tumor-associated inflammatory cells and consistent with IHC findings.
We next investigated the contribution of gene categories and age at diagnosis to the predictive accuracy of the NBL-NA signature. Using a permutation strategy (Data Supplement), we discovered that on average, the inclusion of inflammation-related genes explained 25% of the accuracy of the 14-gene model in predicting PFS and added to the 63% provided by tumor cell–related genes. Age at diagnosis explained an additional 12% of the accuracy.
Inflammation-related genes were also found to be highly correlated to one another in the CCG cohort, a pattern that was also observed in the GPOH and COG cohorts (Fig 3A; Data Supplement). The strongest gene-gene correlations were observed between IL6R and CD14 (Spearman r = 0.77; P < .001; Fig 3C) and between IL6R and CD33 (Spearman r = 0.75; P < .001). IL6R is primarily expressed on cells of the monocytic lineage, but it has also been reported to be expressed on some neuroblastoma cell lines.31
The nine genes categorized as tumor cell–related included neurotrophic kinase receptor 2 (NTRK2) and calmodulin-binding transcription activator 1 (CAMTA1), genes known to be associated with neuroblastoma growth and suppression, respectively.24–27 Interestingly, IL6R was also found to be moderately correlated with the expression of NTRK2 (Spearman r = 0.58; P < .001; Fig 3D). Together, these data suggest that inflammation-related genes, especially genes related to polarization of TAMs, contribute to prognosis and are associated with a poor clinical outcome.
The diverse outcomes of children with NBL-NA have been largely unexplained. Our study suggests for the first time that infiltrating inflammatory cells, especially TAMs, may contribute to this diversity. We demonstrate that TAMs are more prevalent in tumors of children with metastatic rather than locoregional neuroblastoma. Furthermore, we show that expression of inflammation-related genes is higher in tumors of children diagnosed at age ≥ 18 months and that a subset of these genes representing TAMs is associated with an extremely poor outcome in this group. Including expression of both inflammatory and tumor cell genes in a 14-gene signature enables prediction of disease progression for the first time in the clinically indistinguishable group of patients diagnosed at age ≥ 18 months with metastatic NBL-NA. The novel finding that five inflammation-related genes contribute to 25% of the accuracy of the 14-gene model emphasizes the role of inflammation in neuroblastoma and uncovers previously unrecognized potential targets for therapy. This 14-gene expression scoring model, which was validated in two independent cohorts of patients, has clinical applicability and may be of used for managing high-risk patients.
In recent years, the concept of inflammatory cells in the tumor microenvironment as critical participants in tumor progression has gained acceptance.1,32 A tumor-infiltrating macrophage and T-cell signature (CD68high/CD4high/CD8low) has been reported to predict PFS and OS in patients with breast cancer.4 In our study, IL6R expression was found to be highly correlated with expression of CD14 (macrophage marker) and CD33 (myeloid marker), and their expression, along with that of IL10 and FCGR3/CD16 (M2 polarization), contributed to the accuracy of our model in predicting disease progression. This novel finding provides a validated clinical context for the recently established role of TAMs and bone marrow mesenchymal stem cells in promoting neuroblastoma growth via activation of the IL6/IL6R pathway.30,31 Melanomas and myelomas, similar to neuroblastomas, have also been shown to co-opt their tumor microenvironment cells to produce IL6, leading to STAT3 activation and promotion of tumor growth.30,31,33,34 Studies using immunocompetent mouse cancer models have demonstrated that antibody responses against transformed cells could lead to recruitment and polarization of macrophages, leading to production of cytokines such as IL6, IL10, and IL4 that in turn stimulate tumor growth and angiogenesis.35,36 In the current study, using a highly specific and sensitive TLDA assay, expression of genes related to the humoral immune system was observed in children diagnosed at age ≥ 18 months but did not contribute to the final predictive model. Additional studies are needed to elucidate the role of antibody response in neuroblastoma and its relation to recruitment and polarization of TAMs.
Genes related to tumor cells contributed most to the accuracy of the 14-gene NBL-NA signature and included NTRK2, which binds brain-derived neurotrophic factor and plays an important role in the survival and differentiation of neuroblastoma cells.25,26,37,38 The association of high NTRK2 expression with aggressive behavior in metastatic NBL-NA was previously reported by our group21 and is further supported in this study. Our present work also reveals a novel association between expression of IL6R and NTRK2. Although additional mechanistic studies are required to understand interactions between these two pathways, our data point to the role of a pro-tumor inflammatory microenvironment in enabling a highly aggressive neuroblastoma phenotype.
Several gene expression and genomic studies of neuroblastoma tumors obtained at diagnosis have previously reported associations with patient outcome.20,39–43 However, these studies analyzed groups of patients who were heterogeneous with respect to MYCN gene amplification status, clinical stage, and age at diagnosis. Segmental genomic alterations identified in high-risk NBL-NA tumors have not been predictive of outcome in this high-risk group but have had clinical utility in children with intermediate-risk neuroblastoma.20,41,44–48 Gene expression–based models built using neuroblastoma samples from clinically heterogeneous groups of patients lack predictive accuracy in those diagnosed at age ≥ 18 months with metastatic NBL-NA.42,43 Similarly, our previously reported 55-gene NBL-NA–specific microarray signature21 was not predictive of outcome for patients with MYCN-amplified tumors, which represent a distinct molecular subgroup of neuroblastomas (unpublished data).49 Overall, these findings suggest that prognostic studies in neuroblastomas should focus on well-defined molecular and clinical subgroups (eg, using MYCN status). Our current finding also highlights the importance of assessing the neuroblastoma tumor microenvironment in prognostic studies.
Our current study defines a clinically applicable 14-gene expression signature that identifies two subsets of patients with different PFS. Children with high-risk tumor-progression scores uniformly had a poor outcome, with 8% to 20% PFS at 5 years after diagnosis, whereas those with low-risk scores had 47% to 57% PFS. It is possible that the addition of treatment response evaluations such as imaging with 123I-metaiodobenzylguanidine and quantification of bone marrow disease with polymerase chain reaction assays will further improve risk classification, especially for patients with a low-risk tumor-progression score.50,51
In summary, our study reports the first evidence of a role for intratumor inflammation in metastatic neuroblastomas and provides a validated prognostic signature for children with metastatic NBL-NA. The increase in expression of inflammation-related genes in children age ≥ 18 months with poor outcome allows the identification of a subgroup of patients at extremely high risk who may benefit from treatments targeting the tumor microenvironment along with tumor cells. The recent success of therapies directed at tumor-associated immune system cells in adult cancers52–54 suggests opportunities for their application in children with neuroblastoma.
We acknowledge the Children's Oncology Group and the German Society for Pediatric Oncology-Hematology for providing neuroblastoma specimens. We thank Ronnie Houston, BS, for his technical assistance in immunohistochemistry and Janahan Gnanachandran, MS, for his data analysis. We would also like to thank Martine Torres, PhD, for her editorial review of the manuscript and Yves DeClerck, MD, for his comments and suggestions.
Supported by Grant No. 2R01 CA60104-16 from the National Cancer Institute (R.C.S.), a grant from Amgen (J.A.S.), Grant No. K12-CA60104 from the National Institute of Child Health and Human Development, grants from Alex's Lemonade Stand Foundation and St Baldrick's Foundation (S.A.), and grants from the Nautica Malibu Triathlon and Bogart Pediatric Cancer Research Program (S.A., R.C.S.).
Authors' disclosures of potential conflicts of interest and author contributions are found at the end of this article.
The author(s) indicated no potential conflicts of interest.
Conception and design: Shahab Asgharzadeh, Robert C. Seeger
Collection and assembly of data: Shahab Asgharzadeh, Jill A. Salo, André Oberthuer, Matthias Fischer, Frank Berthold, Michael Hadjidaniel, Cathy Wei-Yao Liu, Leonid S. Metelista, Roger Pique-Regi, Peter Wakamatsu, Judith G. Villablanca, Susan G. Kreissman, Katherine K. Matthay, Hiroyuki Shimada, Wendy B. London, Richard Sposto, Robert C. Seeger
Data analysis and interpretation: Shahab Asgharzadeh, Jill A. Salo, Lingyun Ji, Michael Hadjidaniel, Roger Pique-Regi, Hiroyuki Shimada, Wendy B. London, Richard Sposto, Robert C. Seeger
Manuscript writing: All authors
Final approval of manuscript: All authors