Diagnostic FDG and FDOPA positron emission tomography scans distinguish the genomic type and treatment outcome of neuroblastoma

Neuroblastoma (NB) is a heterogeneous childhood cancer that requires multiple imaging modalities for accurate staging and surveillances. This study aims to investigate the utility of positron emission tomography (PET) with 18F-fluorodeoxyglucose (FDG) and 18F-fluoro-dihydroxyphenylalanine (FDOPA) in determining the prognosis of NB. During 2007–2014, forty-two NB patients (male:female, 28:14; median age, 2.0 years) undergoing paired FDG and FDOPA PET scans at diagnosis were evaluated for the maximum standardized uptake value (SUVmax) of FDG or FDOPA by the primary tumor. Patients with older age, advanced stages, or MYCN amplification showed higher FDG and lower FDOPA SUVmax (all P < 0.02). Receiver operating characteristics analysis identified FDG SUVmax≥ 3.31 and FDOPA SUVmax < 4.12 as an ultra-high-risk feature (PET-UHR) that distinguished the most unfavorable genomic types, i.e. segmental chromosomal alterations and/or MYCN amplification, at a sensitivity of 81.3% (54.4%–96.0%) and a specificity of 93.3% (68.1%–99.8%). Considering with age, stage, MYCN status, and anatomical image-defined risk factor, PET-UHR was an independent predictor of inferior event-free survival (multivariate hazard ratio, 4.9 [1.9–30.1]; P = 0.012). Meanwhile, the ratio between FDG and FDOPA SUVmax (G:D) correlated positively with HK2 (Spearman's ρ = 0.86, P < 0.0001) and negatively with DDC (ρ = −0.58, P = 0.02) gene expression levels, which might suggest higher glycolytic activity and less catecholaminergic differentiation in NB tumors taking up higher FDG and lower FDOPA. In conclusion, the intensity of FDG and FDOPA uptake on diagnostic PET scans may predict the tumor behavior and complement the current risk stratification systems of NB.


INTRODUCTION
Neuroblastoma (NB) is an embryonal cancer originating from the sympathetic nervous system [1]. The clinical course of NB is highly variable [1]. For the purposes of staging and surveillance, a combination of bone marrow studies, urinary catecholamine levels, and multiple imaging modalities have been used to accurately define the extensiveness of disease [2].
Traditionally, the tumor biology of NB has been defined by age, stage, histopathology, MYCN amplification, and ploidy [1]. More recently, overall genomic patterns of copy number alterations were proven to have independent prognostic value [3,4]. In addition, CT or MR imagedefined risk factors (IDRFs) such as tumor encasement of major structures predict worse outcome in localized tumors [5]. Whether modern molecular imaging tools performed at diagnosis can help to predict tumor biology or treatment outcome has been less studied. 123 I-Metaiodobenzylguanidine ( 123 I-MIBG) scintigraphy is the standard molecular imaging of NB [6,7]. By targeting the norepinephrine transporter (encoded by SLC6A2) [8], MIBG is specifically accumulated in NB cells, not only providing diagnostic and prognostic values [9,10] but also serving as a prelude to targeted radiotherapy with 131 I-MIBG in ultra-high-risk patients [11].
Due to its increasing availability, positron emission tomography (PET) with 18 F-fluorodeoxyglucose (FDG) has been explored in NB [12][13][14][15]. By targeting aerobic glycolysis of cancer cells, known as the Warburg effect [16], FDG PET provides additional diagnostic information for NB that fails to accumulate or weakly condenses MIBG [12], and complements 123 I-MIBG scan in demonstrating localized NBs and soft-tissue lesions [13]. The international guidelines for NB recommend FDG PET as an option to imaging with 123 I-MIBG [6].
More recently, 18 F-fluoro-dihydroxyphenylalanine (FDOPA) PET has shown high accuracy compared with CT/MR imaging [17] and good sensitivity compared with 123 I-MIBG scintigraphy [18,19]. By targeting the ubiquitous expression of aromatic l-amino acid decarboxylase (encoded by the DDC gene) in NB [20], which catalyzes l-DOPA to l-dopamine in the catecholamine biosynthesis pathway, the whole-body metabolic burden of FDOPA also confer a prognostic role in relapsed/refractory NBs [21]. However, the prognostic value of FDOPA PET scan at initial diagnosis of NB remains to be elucidated.
As the commercial supply of 123 I-MIBG has been limited in Taiwan, we have utilized FDG and FDOPA PET in diagnosing and following NB patients. This study aims to investigate the association between the tumor uptake pattern of FDG and FDOPA on diagnostic PET scans and the clinical features, genomic types, as well as treatment outcome in NB.

Clinical and imaging features
During the study period from June 2007 to July 2014, 88 patients with clinical diagnosis of NB were enrolled for PET scans. Forty-six patients were excluded from analysis: Forty-two patients had PET scans performed after receiving the second chemotherapy cycle or during post-treatment follow-up only; two patients only had FDG PET at diagnosis; one patient only had FDOPA PET at diagnosis; one patient had gross total resection (GTR) of primary NB tumor prior to PET imaging. The rest 42 patients who had paired FDG and FDOPA imaging performed on different days at initial diagnosis were eligible for analysis (Supplementary Figure S1). There were 28 boys and 14 girls. The median age at diagnosis was 2.0 years (interquartile range [IQR], 0.5-4.9 years). Most patients were older than 18 months (n = 24; 57%), had stage 4 disease (n = 25; 60%), and belonged to the high-risk group (n = 30; 71%) ( Table 1).
Whole-body PET scans using FDG or FDOPA provided good spatial resolution and clear contrast in the bony compartments and complement each other ( Figure 1). Although the strong physiologic FDG uptake by the brain and nasopharynx interfered the interpretation of skull lesions, FDOPA PET helped to identify lesions in the head and neck region more accurately. These imaging features are consistent with our previous findings [18].

FDG and FDOPA uptake by primary tumors and their clinical characteristics
Reading the FDG and FDOPA PET images of NB, we noted that primary NB tumors frequently took up FDG and FDOPA at different intensity. In this study, the maximum standardized uptake value (SUV max ) of each primary tumor was measured as the indicator of tumor uptake of FDG or FDOPA. The SUV max was used because of its lower sensitivity to partial-volume effects and higher reproducibility between observers. Table 1 shows the tumor uptake values compared by clinical characteristics. The SUV max of FDG and the ratio correlated positively with HK2 (Spearman's ρ = 0.86, P < 0.0001) and negatively with DDC (ρ = −0.58, P = 0.02) gene expression levels, which might suggest higher glycolytic activity and less catecholaminergic differentiation in NB tumors taking up higher FDG and lower FDOPA. In conclusion, the intensity of FDG and FDOPA uptake on diagnostic PET scans may predict the tumor behavior and complement the current risk stratification systems of NB. www.impactjournals.com/oncotarget between FDG and FDOPA uptake (G:D) were significantly higher in patients with high-risk features, including older age, stage 4, MYCN amplification, and anatomical imagedefined risk factors (IDRFs) [5] (all P < 0.05). By contrast, the SUV max of FDOPA was significantly higher in patients with younger age, lower stage, and low-or intermediaterisk groups (all P < 0.02).

Tumor uptake distinguishes NB genomic types
In the study cohort, 31 of 42 patients had their genomic type of NB determined by array-based comparative genomic hybridization (array-CGH) and/ or chromogenic in situ hybridization. Twenty-seven (87%) out of the 31 patients with available genomic data showed copy number alterations that fall into one of the three major genomic types in NB [22]: Numerical (wholechromosomal) alterations (Num + ; n = 11), segmental chromosomal alterations (Seg + ; n = 7), or MYCNamplification (MNA + ; n = 9). In the other four patients with genomic data available, three had no significant copy number alterations (i.e. silent or "flat") and one had Beckwith-Wiedemann syndrome with IGF2 microdeletion.  Comparing tumor uptake of FDG and FDOPA by the three major genomic types, FDG uptake of the primary tumor is highest in MNA + , intermediate in Seg + , and lowest in Num + . By contrast, FDOPA uptake is highest in Num + and lowest in Seg + and MNA + ( Figure 2B and Table 1). Receiver operating characteristics (ROC) analysis identified that FDG hi with SUV max ≥ 3.31, FDOPA lo with SUV max < 4.12, and G:D ratio ≥ 1.09 are the cutoff values that best distinguish the unfavorable Seg + and MNA + genomic types from the favorable Num + and other genomic type of NB ( Figure 2C-2E). Comparing the area under ROC curves (AUC) of FDG uptake, inversion of FDOPA uptake (1/SUV max ), and G:D ratio in classifying the unfavorable genomic types of NB (Seg + and MNA + ), G:D showed a significantly better performance than FDG or FDOPA (P = 0.03; Figure 2F) with a sensitivity of 87.5% (61.7%-98.4%) and specificity of 86.7% (59.5%-98.3%). Combining the criteria of FDG SUV max ≥ 3.31 and FDOPA SUV max < 4.12 to define an ultra-high-risk group by PET (PET-UHR; Supplementary Figure S2B) yields a sensitivity of 81.3% (54.4%-96.0%) and specificity of 93.3% (68.1%-99.8%) with an estimated AUC of 0.87 in predicting Seg + or MNA + genomic types of NB (Table 2).

Prognostic value of FDG and FDOPA uptake by the primary tumors
At a median follow-up of 39.2 months (range, 8.8-100.3 months), the 42 patients have a 5-year event-free survival (EFS) rate of 41% (25%-57%) and a 5-year overall survival (OS) rate of 50% (28%-68%). Patients with older age, stage 4 disease, MYCN amplification, high-risk disease, and unfavorable genomic types (Seg + and MNA + ) had significantly worse EFS and OS rates. By contrast, neither IDRFs nor GTR had an impact on survival (Supplementary Figure S3).
Comparing by the values of tumor uptake, strikingly distinct outcomes were found ( Figure 3). Specifically, tumors showing higher FDG avidity (SUV max ≥ 3.31; Figure 3A and 3B), lower FDOPA avidity (SUV max < 4.12; Figure 3C and 3D), or higher G:D ratio (ratio ≥ 1.09; Figure 3E and 3F) were associated with significantly worse survival rates (all P < 0.01). Cox proportional hazard modeling showed that, on univariate analysis, age, stage, MYCN status, FDG hi , FDOPA lo , G:D ≥ 1.09, and PET-UHR were associated with inferior EFS on univariate analysis (Table 3). On multivariate analysis, PET-UHR was confirmed as a poor prognostic factor (hazard ratio, 4.9 [1.4-16.9]; P = 0.012) that was independent from age, stage, IDRF, and MYCN status (Table 3). When FDG uptake, FDOPA uptake, or G:D ratio was analyzed with the traditional risk factors respectively, FDG hi or FDOPA lo each predicted inferior EFS independently, while G:D was not significant (Supplementary Table 1).

Tumor uptake of FDG and FDOPA and gene expression
To evaluate the probable mechanisms of tumor uptake, we analyzed the expression of selected PET imaging-related genes in 16 primary tumor samples (Figure 4 and Supplementary Figure S4). G:D correlated strongly with hexokinase 2 (HK2) expression (Spearman's ρ = 0.86, P < 0.0001) and negatively with DDC expression (ρ = −0.58, P = 0.02). Although hexokinase 1 (HK1) also phosphorylates FDG and is expressed at a much higher levels, there was no correlation between HK1 expression and tumor uptake. Surprisingly, FDOPA uptake had no significant correlation with TH (encoding tyrosine hydroxylase upstream of AADC; P = 0.09) or DDC (P = 0.34) expression levels, but showed positive correlation with the expression of SLC6A2, the target of 123 I-MIBG scintigraphy (ρ = 0.68, P = 0.004), supporting our previous finding that 123 I-MIBG avidity was associated with higher FDOPA uptake [18].

DISCUSSION
The background of this study is the lack of commercially available 123 I-MIBG in Taiwan during the study period, which has driven us to explore the clinical use of FDG and FDOPA PET in NB. Although we have demonstrated a complementary role of these PET scans in the diagnosis of NB, we must stress that 123 I-MIBG scan, in countries with a stable supply, remains the single most important molecular imaging of NB. The more recent application of PET with 124 I-MIBG [25] or 18 F-labeled MIBG analogs [26,27] may further enhance the diagnostic power of the "MIBG family" of molecular imaging techniques.    To the best of our knowledge, this is the first study showing significant prognostic impact of both FDG and FDOPA PET at initial diagnosis of NB. We showed that the molecular imaging phenotype of FDG hi , FDOPA lo is associated with unfavorable clinico-genomic features and worse treatment outcome in NB. The clinical implications of our findings are two-fold.
First, FDG hi , FDOPA lo is associated with unfavorable clinical characteristics but also with the poor-risk Seg + and MNA + genomic types, indicating a strong relationship between metabolic patterns on PET and underlying tumor biology. Although data on age, stage, histopathology, and MYCN amplification can be confidently obtained in most clinical settings [24], the detection of Seg + , that accounts for 37% of NB and confers a dismal outcome [3], requires time-and resource-consuming pangenomic methods or multiple in situ hybridizations [22]. The proposed "PET-UHR" criteria yielded a positive predictive value of 92.9% in predicting Seg + or MNA + , providing a convenient parameter of the underlying tumor biology for patients undergoing diagnostic PET scans. Furthermore, area under the ROC curve of G:D is larger than that of FDG  or FDOPA uptake alone in classifying the unfavorable genomic types, indicating that combining the information of FDG and FDOPA uptake may serve as a useful biomarker to predict the underlying tumor biology in NB. Interestingly, while the FDG SUV max was sequentially higher in the three genomic types (Num + < Seg + < MNA + ), the FDOPA SUV max is equally lower in the two unfavorable types (Num + > Seg + = MNA + ). Since both the Seg + and MNA + genomic types are associated with the FDG hi , FDOPA lo phenotype and a very poor outcome [3], the MYCN oncoprotein itself or its downstream signals may further contribute to a metabolic phenotype towards hyperglycolysis [28], causing significantly higher FDG uptake and more aggressive behavior of MNA + tumors.
The second clinical implication of our study is that the signal intensity of tumor uptake, i.e. FDG hi and/ or FDOPA lo , serves as independent predictors of inferior prognosis. Based on 123 I-MIBG scintigraphy, semiquantitative scoring systems have been developed to assess the total extent of MIBG-avid lesions and predicted the treatment outcome in NB [9,10]. Adopting these scoring systems of tumor extensity to FDG [14] and FDOPA PET [21] was also prognostic in relapsed/refractory NB. In addition to assessing disease extensiveness, measuring the intensity of tumor uptake may further expand the usefulness of PET scans in NB.
On FDG PET, higher SUV max correlated with higher stage at diagnosis [15,29] and worse survival at relapse [14]. Interestingly, the FDG uptake values in this study cohort are relatively similar to the results from relevant subgroups in two recent Korean studies (Supplementary Table 2) [15,29], although the measurement of SUV max may vary with different instruments and imaging protocols. Appropriate instrumentation calibration, standardization of protocols for patient preparation and scanning, and stricter uptake-time control with effective correction algorithms [30] may eventually enable FDG SUV max to become a reproducible imaging biomarker in multi-center and multinational trials of NB.
On FDOPA PET, Piccardo et al. showed that higher "whole-body metabolic burden", defined as the sum of bony metastatic extent plus the product of tumor uptake and tumor volume, correlated with very poor outcome in relapsed/refractory NB; however, the prognostic impact of the mean or maximal uptake of FDOPA by the primary tumor was not reported [21]. More studies are needed to further delineate the impact of FDG and FDOPA uptake across patient populations.
The biological basis of FDG hi , FDOPA lo may largely be explained by the Warburg effect [16]. Both FDG hi and FDOPA lo phenotypes correlated with higher HK2 expression, which is required for oncogenic transformation in vitro and tumor initiation in vivo [31] and is enhanced by MYCN and HIF-1α [32]. In breast cancer, the gene expression profile of tumors with FDG SUV max > 10 showed an "FDG signature" that was enriched with glycolysis-related genes and associated with activation of the MYC transcription factor (c-Myc), a functional counterpart of MYCN [33]. In this study, we demonstrated that NB with MYCN amplification correlated with higher FDG uptake.
In addition to its prognostic impact, the glycolysis pathway per se may serve as a therapeutic target. It has been recently found that treating NB cells and mouse models with 2-deoxyglucose, an analog of FDG and inhibitor of glycolysis, decreased HK2 expression and induced apoptosis [34]. Preclinical testing has also proved the in vivo efficacy of 2-deoxyglucose in NB [35]. Based on these findings, PET imaging may be further studied as a biomarker that identifies patients with NB or other cancers who would most likely to benefit from novel therapies targeting MYC/MYCN [28,36] or the glycolytic pathway [34,35,37].
Surprisingly, we found that lower, rather than higher, FDOPA uptake significantly correlated with very poor prognosis. The high uptake of FDOPA in NB and its association with better outcome may be explained by its characteristic catecholaminergic differentiation. Although FDOPA uptake did not correlate with the degree of differentiation on histopathology, its negative correlation with HK2 and positive correlation with SLC6A2 expression levels suggests that FDOPA lo tumors may confer a hyper-glycolytic and catecholaminergically dedifferentiated phenotype at the molecular level. Characterization of an "FDOPA signature" of gene expression is warranted.
Our study has some limitations. As the tumor uptake values are continuous variables, they may have continuous impact on prognosis, similar to the age effect in NB [38]. Our sample size precluded extensive statistical analyses to search for the best cutoffs that predict the highest risk of incomplete resection or poor prognosis. We postulate that PET scans may add value to anatomical imaging and conventional risk stratification systems by identifying a subgroup of patients with excellent prognosis, who can truly benefit from modern multimodal therapy, from the ultra-high-risk patients who should be enrolled in novel therapeutic trials. However, this observation cannot yet be made with assurance. Our data support that the utility of diagnostic FDG and/ or FDOPA PET should be further evaluated in a larger cohort of high-risk NB, ideally through multi-center or international collaborations.
In conclusion, we showed that the tumor uptake intensity of FDG and FDOPA on PET scans at diagnosis distinguish the tumor biology and treatment outcome in NB. The unfavorable PET imaging phenotype with FDG hi , FDOPA lo uptake patterns were associated with poor-risk clinico-genomic features and worse prognosis. The prognostic value of PET-UHR, using FDG SUV max ≥ 3.31 and FDOPA SUV max < 4.12 as criteria, may be incorporated in future risk stratification systems and be validated in prospective trials.

Patient enrollment
From June 2007 to July 2014, patients diagnosed with NB were enrolled. FDG and FDOPA PET scans were performed at diagnosis and during follow-up at National Taiwan University Hospital, Taipei, Taiwan at an interval of 3-12 months with other standard evaluations. Patients with paired FDG and FDOPA PET scans performed before completion of the first chemotherapy cycle were included for analysis ( Figure 1). The study was approved by National Taiwan University Hospital Research Ethics Committee. Informed consent was obtained from each participant's guardian.

Standard evaluations for neuroblastoma patients
In addition to PET imaging, routine evaluations included complete blood count, basic biochemistry, serum lactate dehydrogenase and ferritin levels, urinary vanilylmandelic acid level, CT/MR imaging, and bone marrow smear and biopsy. The presence of CT/MR image-defined risk factor was retrospectively retrieved from patients' radiological reports at diagnosis, based on the International Neuroblastoma Risk Group staging system criteria [5]. All patients were staged according to International Neuroblastoma Staging System [39]. The histological classification of resected tumors was based on the International Neuroblastoma Pathology Classification [40]. MYCN amplification was determined by chromogenic in situ hybridization [41]. DNA ploidy and urinary homovanillic acid level were not routinely evaluated.

Acquisition of PET images
FDG and FDOPA were produced with commercial systems (TRACERlab, GE Healthcare, USA) [18,45]. FDG and FDOPA PET scans were performed separately, at least one day apart. Before FDG PET, patients were fasted with avoidance of glucose-containing intravenous fluids for 6 hours. Before FDOPA PET, patients received 2 mg/kg of carbidopa orally 1 hour before injection [46]. Forty-five minutes after FDG (5 MBq/kg) or 90 minutes after FDOPA (4 MBq/kg) injection, whole-body images were acquired on a PET/CT scanner with low-dose CT (Discovery ST-16, GE Medical Systems, Milwaukee, WI). We began sedation 30 minutes before scanning if necessary.
Images were reconstructed using an iterative algorithm provided by the machine manufacturer. Analysis was performed on attenuation-corrected images. The maximum standardized uptake value (SUV max ) was determined by manually drawing elliptical regions of interest around areas of abnormal uptake, using the software from GE Medical System on a Xeleris 2 ® workstation (GE Healthcare). For patients with two or more primary tumors (n = 3), tumor with the largest volume was analyzed.

Quantitative real-time PCR
Total RNA was extracted from frozen tumor tissue with Trizol reagent following the manufacturer's instructions (Invitrogen). cDNA was produced from 5 mg of RNA using Superscript III with random hexamer primers (Invitrogen). Analysis of gene expression was performed in Applied Biosystems 7500 Fast Real-Time PCR System by using specific TaqMan primers (Applied Biosystems) for HK1 (Hs00175976_m1), HK2 (Hs00606086_m1), TH (Hs00165941_m1), DDC (Hs01105048_m1), SLC6A2 (Hs00426573_m1), HPRT1 (Hs99999909_m1), and SDHA (Hs00188166_m1). Expression levels of PET-related genes were averaged from two replicates and normalized to the geometric mean of HPRT1 and SDHA, which are control genes with the least expression variability across NB samples [47]. www.impactjournals.com/oncotarget

Statistical analysis
Outcome data were frozen on October 9, 2015. The Fisher exact, Kruskal-Wallis, and Spearman's nonparametric correlation tests were used to evaluate the association across variables. The cutoff value of PET imaging parameters was determined by ROC curve analysis. EFS was calculated from diagnosis to the first occurrence of relapse, progression, secondary malignancy, or death, or to the last contact if no event occurred. OS was calculated until the time of death or until the last contact that the patient was alive. Kaplan-Meier curves were generated and compared by log-rank tests. Cox proportional hazard models were built to test for prognostic values. The statistical analyses were performed with Small Stata 11.0 software (StataCorp, College Station, TX). All tests were two-sided. P values < 0.05 were considered statistically significant.