MRI Imaging of the Hemodynamic Vasculature of Neuroblastoma Predicts Response to Antiangiogenic Treatment

Neuroblastoma is a tumor of childhood that arises within the embryonic sympatho-adrenal lineage of the neural crest. Neuroblastoma in high-risk form is chemoresistant, and metastatic relapse still accounts for 13% of all cancer-related death in children. Amplification of the proto-oncogene MYCN is the commonest genomic aberrations and defines a subgroup of children at high-risk of developing uncontrollable refractory or relapsing disease despite aggressive multimodal therapy (1).

In neuroblastoma, a high vascular index is associated with increased disseminated disease, amplification of MYCN, unfavorable histology, and overall poor prognosis (2). Increased microvascular proliferation and other specific vascular morphological patterns are associated with even poorer prognosis, highlighting the pivotal role of angiogenesis in determining the clinical behavior of neuroblastoma (3, 4). The VEGF family is a key regulator of angiogenesis in neuroblastoma and high VEGF expression at the time of diagnosis is associated with poor outcome (5). Numerous antiangiogenic therapies are being evaluated in early phase pediatric clinical trials in children with solid tumors, including the anti-VEGF monoclonal antibody bevacizumab in combination with temozolomide (BEACON, NCT02308527), and the tyrosine kinase inhibitors regorafenib (NCT02085148) and pazopanib as a single agent (NCT01956669) or in combination with metronomic oral topotecan (TOPAZ), as well as the panVEGFR inhibitor cediranib currently being evaluated in children with metastatic alveolar soft part sarcoma (NCT00942877).

MRI is becoming the preferred clinical imaging technique for the management of children with neuroblastoma because of its exquisite soft tissue contrast. MRI provides excellent anatomical information at diagnosis and follow up while sparing exposure to ionizing radiation associated with CT (6). Advanced functional MRI techniques can additionally be used to define quantitative imaging biomarkers that inform on biologically relevant structure–function relationships in tumors in vivo. Perfusion MRI (pMRI) methodologies such as dynamic contrast enhanced (DCE) MRI using low molecular weight gadolinium chelates are often used to evaluate vascular response to VEGF signaling inhibitors in adult oncology clinical trials. However, these techniques suffer from marked measurement variability and are challenging to perform in young children (7, 8). Arterial spin labeling (ASL) MRI is an emerging and attractive contrast-free approach for the evaluation of cerebral perfusion, yet many challenges need to be overcome before its utility to assess vascular perfusion in extracranial tumors can be evaluated. Two complementary magnetic susceptibility-based MRI approaches are being actively exploited to assess tumor vascular function and response. Tumor vasculature has been studied using (i) intrinsic susceptibility (IS) MRI, which measures the intrinsic transverse relaxation (R₂*) contrast produced by paramagnetic deoxyhemoglobin within tumor blood vessels, and (ii) susceptibility contrast (SC) MRI, which relies on the intravenous administration of ultrasmall superparamagnetic iron oxide (USPIO) particles. Distribution of USPIO particles within tumors causes regional increases in tumor R₂* (ΔR₂*), from which the fractional blood volume (fBV, %) can be derived (9, 10). The long intravascular half-life of USPIO particles enables steady-state acquisition and high-resolution mapping of regional variations in tumor perfusion. Both R₂* and fBV has been shown to be sensitive imaging biomarkers of response to vascular-targeted therapies in vivo in the preclinical setting (11–13).

Imaging biomarkers must undergo stringent validation before they can be deployed clinically (14). Early imaging biomarker development demands close imaging-pathology correlation, to understand the biological processes underpinning the imaging measurement, which can be meaningfully studied using animal models. The Th-MYCN genetically engineered mouse (GEM) model of neuroblastoma has been shown to faithfully recapitulate the major genetic and pathophysiological features of the childhood disease (12, 15). The primary abdominal tumors in the Th-MYCN mice present with multiple para- and intratumoral anatomical landmarks detectable by conventional MRI, which can be used to ensure the accurate registration between the functional MRI-derived parametric maps and histopathology crucial for the validation of imaging biomarkers. The GEM models of neuroblastoma thus represent an information-rich platform with which to evaluate promising therapeutic strategies and associated noninvasive imaging biomarkers.

In this study, we investigated the utility of SC- and IS-MRI to inform on the functional tumor vasculature, and its response to the potent pan-VEGFR inhibitor cediranib, in tumors arising in the Th-MYCN GEM model of neuroblastoma. We then compared the regional distribution of tumor fBV and R₂* with aligned whole-slide digitized pathology stained for vessel density and red blood cell (RBC) aggregation respectively, and validated, via hotspot mapping and spatial statistics, to determine whether the MRI biomarkers can robustly characterize the vascular phenotype of neuroblastoma in vivo. We also demonstrate here that both baseline fBV and R₂* are predictive imaging biomarkers of tumor response to cediranib. The pathologic attributes associated with the differential sensitivity to cediranib treatment in the Th-MYCN model are discussed and positioned against available clinical MRI findings of MYCN-amplified neuroblastoma, including our initial experience with IS-MRI.

All experiments were approved by The Institute of Cancer Research Animal Welfare and Ethical Review Body and performed in accordance with the UK Home Office Animals (Scientific Procedures) Act 1986, the United Kingdom National Cancer Research Institute guidelines for the welfare of animals in cancer research (16) and the ARRIVE (animal research: reporting in vivo experiments) guidelines (17).

Transgenic Th-MYCN mice were genotyped to detect the presence of the human MYCN transgene (18). The study was performed using both male and female hemizygous mice, which developed palpable tumors at 50 to 130 days with a 25% penetrance. Tumor development was monitored weekly by palpation by an experienced animal technician. Mice with palpable tumors were then enrolled (day 0) and their tumor volume was subsequently monitored by MRI. A total of 68 mice were enrolled with a median tumor volume of 801 ± 63 mm³ (median ± 1 SEM, ranging from 143 to 2,055 mm³). Mice were housed in specific pathogen-free rooms in autoclaved, aseptic microisolator cages (maximum of four mice per cage).

We first evaluated the sensitivity of MRI to cediranib-induced acute modulation of neuroblastoma vasculature. IS- and SC-MRI were performed prior to (day 0) and 24 hours after treatment started (day 2). Mice were treated on day 1 with 6 mg/kg of cediranib orally (obtained under material transfer agreement with AstraZeneca, n = 10) or vehicle (n = 8). One mouse was excluded due to failed remote contrast injection. The 24-hour imaging timepoint was chosen based on (i) preliminary evidence that cediranib does not elicit any significant volume reduction at this timepoint in the highly chemosensitive Th-MYCN model (12) and (ii) cediranib caused significant reductions in DCE-MRI parameters at this timepoint in the adult phase I clinical trial (19).

In an additional cohort (n = 12) we further evaluated the effect of sustained daily treatment, with IS- and SC-MRI performed prior to, 24 hours and 7 days after daily treatment with 6 mg/kg cediranib. MRI data were not collected from two mice at the 24-hour timepoint.

Guided by the results of Study 2, we subsequently acquired IS-MRI data from additional Th-MYCN mice (n = 25) prior to daily treatment with cediranib for 7 days (bringing the total number of mice from which pretreatment R₂* data were acquired to 37). The volumetric response to cediranib over 7 days treatment was monitored by T₂-weighted MRI only, and compared with that from mice treated daily with vehicle (n = 12).

All MRI studies were performed on a 7T Bruker horizontal bore MicroImaging system (Bruker Instruments) using a 3 cm birdcage volume coil. Anesthesia was induced by an intraperitoneal 5 mL/kg injection of a combination of fentanyl citrate (0.315 mg/mL) plus fluanisone (10 mg/mL; Hypnorm, Janssen Pharmaceutical) and midazolam (5 mg/mL; Roche) and water (1:1:2). A lateral tail vein was cannulated with a 27G butterfly catheter (Hospira) for remote administration of USPIO particles. Core temperature was maintained at approximately 37°C with warm air blown through the magnet bore.

For all the mice, anatomical T₂-weighted transverse images were acquired from 20 contiguous 1-mm-thick slices through the mouse abdomen, using a rapid acquisition with refocused echoes (RARE) sequence with four averages of 128 phase encoding steps over a 3 × 3 cm field of view, an echo time (TE) of 36 milliseconds, a repetition time (TR) of 4.5 seconds and a RARE factor of 8. These images were used to determine tumor volumes, and for planning the subsequent functional MRI measurements, which included optimization of the local field homogeneity using FASTmap algorithm and the measurement of the baseline transverse relaxation rate R₂* (second⁻¹), which is sensitive to the concentration of paramagnetic species, principally deoxyhemoglobin. R₂* was quantified using a multiple gradient-recall echo (MGE) sequence with eight averages, and an acquisition time of 3 minutes 20 seconds. Images were acquired using eight echoes spaced 3 milliseconds apart, an initial TE of 6 milliseconds, a flip angle α  =  45° and a TR of 200 milliseconds. A dose of 150 μmol Fe/kg of the USPIO particle preparation P904 (overall particle size ∼25–30 nm diameter; Guerbet) was then administered intravenously. After 3 minutes to allow for equilibration, a second set of identical MGRE images acquired.

All the MRI data were acquired with a matrix size of 128 × 128 over a 3 × 3 cm field of view. Tumor volumes were determined using segmentation from regions of interest drawn on T₂-weighted images for each tumor-containing slice using OsiriX. Tumor R₂* maps were calculated from regions of interest drawn for each tumor-containing slice from the MGRE images acquired prior to and following administration of USPIO particles by fitting a single exponential to the signal intensity echo time curve on a voxel-by-voxel basis using a robust Bayesian approach using in-house software (ImageView, developed in IDL; ITT Visual Information Systems). Parametric maps of tumor fBV (%) were subsequently calculated using the USPIO-induced change in R₂* (ΔR₂*), as described previously (20).

Guided by T₂-weighted anatomical images, tumors were carefully excised and orientated for subsequent histopathologic processing. Formalin-fixed and paraffin-embedded tumors were cut in 3 μm sections and were stained with hematoxylin and eosin (H&E) and for the murine vascular endothelial marker endomucin (rabbit EP3095; Millipore). Whole-Slide images were digitized using a Hamamatsu NanoZoomer XR scanner (×20 magnification, 0.46 μm resolution).

The MRI slice of interest were visually aligned with digitized whole-slide H&E stained images using anatomical landmarks from T₂-weighted images, including the shape of the tumor, the position of the kidneys, the position and/or orientation of tumor-displaced, and circumferentially surrounded abdominal aorta and renal aorta/vein and the presence of lymph nodes.

Both RBCs and endomucin staining were automatically extracted from T₂-weighted MRI corresponding digitized whole-slide histopathologic images for the validation of R₂* and fBV as biomarker of RBCs aggregation and vascular perfusion. Histologic images (0.46 × 0.46 μm pixel resolution) were first split into tiles of 2,000 × 2,000 pixels (jpeg) for computational efficiency.

A macro was written in Fiji to extract RBCs from each tile by applying color deconvolution to extract the eosin color channel followed by the application of Otsu's automatic threshold detection method, both using ImageJ/Fiji plugins (with Java 8) (21). Algorithm accuracy was tested using independent annotation of 561 RBC and 591 non-RBC points in nine samples.

A macro was written in Fiji to extract endomucin from each tile by applying color deconvolution to extract the brown color channel followed by the application of maximum entropy threshold detection method, both using ImageJ/Fiji plugins (Java 8).

Algorithm accuracy was tested using independent annotation of stained/nonstained points (688/734) on nine different samples.

Whole-slide images of RBC and endomucin staining were converted into binary and processed to match MRI resolution (234 × 234 μm), with the fraction of pixels occupied by RBC or endomucin staining within 518 × 518 pixel-regions representing a single pixel in the final RBC map.

Comparison of R₂* versus RBC and fBV versus endomucin parametric maps was performed visually following the application the kernel density estimation (KDE) to display hotspots of high, above the 85th percentile for each tumor sample, R₂* and RBC (%), fBV (%), and endomucin staining (%) values using the “MASS” package in R (22, 23).

Accurate registration between histology and MRI images is essential to derive robust conclusions and apply spatial statistics. We applied the automatic coherent point drift (CPD) algorithm for both (i) rigid and (ii) nonrigid registration to corresponding pairs of images (R₂* or fBV maps and computed maps of RBC and endomucin, respectively) in Matlab (24). This algorithm was primarily selected because it preserves topologic structures due to the coherent motion of the point sets. To avoid any bias, the point sets (features) from each map were edges extracted using a simple Canny edge detector. Independently, we also performed (iii) a manual registration by rotation and scaling to match.

Statistical analysis was performed with GraphPad Prism 6 (GraphPad Software Inc.). The mean of median values for all the quantitative MRI parameters, the mean values for tumor volume, RBC, and endomucin were used for statistical analysis with a 5% level of significance. Any significant differences in tumor volume and quantitative histopathologic parameters were identified using Student two-tailed unpaired t test, with a 5% level of significance. Significant correlations between the mean values. All linear correlations were determined by Pearson's correlation method. Level of significance was 5%.

The Mantel test, a nonparametric analysis of associations between corresponding positions of two distance matrices, was perfomed on registered pairs of MRI maps and computed maps of histopathologic features using the “ade4” library in R. Distance matrices were calculated using the “dist” function.

Based on the results of the preclinical investigations we aimed to (i) evaluate the clinical relevance of the vascular phenotype observed in the Th-MYCN tumors by reviewing available diagnostic MRI data from patients with MYCN-amplified neuroblastoma, and (ii) evaluate the feasibility of acquiring IS-MRI in children with neuroblastoma.

Informed written consent was obtained from all parents or patients following IRB approval of the prospective study with IS-MRI at the Royal Marsden Hospital. For the retrospective review of routine diagnostic imaging at Great Ormond Street Hospital, anonymized data were provided to the researchers.

Patients (n = 19, 13 male/6 female) with confirmed MYCN-amplified disease included in this retrospective analysis were 1 day to 3.6 years old and imaged between May 2008 and July 2017 at Great Ormond Street Hospital. Exclusion criteria were (i) no MRI at diagnosis or (ii) MRI performed externally. Diagnostic imaging was performed on a 1.5 T Magnetom Avanto system (Siemens Healthcare) and included short-time inversion recovery (STIR) T₂-weighted sequences (TE = 60 milliseconds, TR = 8,000 milliseconds) and spectral attenuated inversion recovery (SPAIR) T₁-weighted BLADE sequence (TE = 20 milliseconds, TR = 600 milliseconds), prior and after the injection of Gd chelate-based contrast agent, diffusion-weighted MRI sequence (TE = 90 milliseconds, TR = 2,700 milliseconds, 5 b-values, b = 0, 50–800 s/mm²). Slice thickness was 6 mm and the field of view was adapted to each patient.

The three patients (n = 2 male/1 female) with refractory/relapsing disease included in this study were 5 to 7 years old and scanned between July 2013 and March 2015. MRI was performed on a 1.5 T Magnetom Avanto system using a phased-array body coil and included T₂-weighted sequence (turbo spin echo sequence TE = 80 milliseconds, TR = 7,000 milliseconds) and T₁-weighted (TE = 3 milliseconds, TR = 10,000 milliseconds) prior and after (last scan of the session) the injection of Gd chelate-based contrast agent, and a diffusion-weighted MRI sequence (TE = 70 milliseconds, TR = 3,400 milliseconds, 6 b-values, b = 0, 50–800 s/mm²). IS-MRI was performed in 2.5 minutes using a 2D MGE sequence (initial TE 5 milliseconds, five echoes, 10 milliseconds apart, TR = 102 milliseconds, flip angle = 40°). Images were analyzed and R₂* quantified in Matlab.

We first assessed changes in vascular function in Th-MYCN mice following treatment with cediranib. At the time of enrolment, the mean tumor fBV quantified using SC-MRI in the Th-MYCN mice (n = 19) was 19.6 ± 1%, consistent with the high vascular index of neuroblastoma. Treatment with cediranib caused an acute and highly significant reduction at 24 hours in tumor fBV (ΔfBV_{cediranib_24h} = −28 ± 4% vs. ΔfBV_{vehicle_24h} = −6 ± 8%, P = 0.0008), which was accompanied by a highly significant cytolentic response (ΔVolume_{cediranib_D0-D2} = 18 ± 8% vs. ΔVolume_{vehicle_D0-D2} = 53 ± 11%, P = 0.0006; Fig. 1A–F). Sustained daily treatment over 7 days caused a further significant reduction in fBV and antitumor activity (Fig. 1G–I). The average baseline fBV strongly correlated with the cediranib-induced reduction in fBV (r = −0.83, P = 0.0008; Fig. 1J), and the value of fBV at baseline correlated with the subsequent cediranib-induced tumor volumetric response at day 7 (r = −0.65, P = 0.02; Fig. 1K).

The analysis of the cohort of mice in which IS-MRI was routinely performed pretreatment confirmed that cediranib effectively suppress of the aggressive tumor growth that is typical of this model (ΔVolume_{cediranib_D0-D7} = −16 ± 5% vs. ΔVolume_{vehicle_D0-D7} = 142 ± 11%, P < 0.0001, n = 37 and 12). However, a range of volumetric response, from progressive disease to partial response, was observed (Fig. 2A). Examination of our pretreatment T₂-weighted anatomical images and native R₂* parametric maps revealed that responsive tumors exhibited a characteristically heterogeneous appearance, with areas of hypointense T₂ signal and relatively fast R₂* (R₂* = 113 ± 4 second⁻¹; Fig. 2B and C), whereas progressive tumors typically demonstrated a more homogeneous, isointense appearance on T₂ and significantly slower native R₂* values (R₂* = 68 ± 5 second⁻¹, P < 0.0001). Cediranib-induced changes in tumor volume over 7 days treatment correlated with native tumor R₂* measured before treatment (r = −0.72, P < 0.0001; Fig. 2D).

The numerous anatomical landmarks (kidney, spleen, abdominal arteries, lymph nodes) clearly evident on T₂-weighted MRI were used to guide the careful excision and subsequent orientation of digitized pathology with SC- and IS-MRI-derived parametric maps. Intratumoral regional differences in fBV visually reflected spatial variations in computed maps of vascular endothelial endomucin staining, automatically extracted from high-resolution IHC images with an accuracy of 97%, after automatic registration using the CPD algorithm (Fig. 3A and B). This was corroborated by the location, size, shape, and orientation of hotspots identified on KDE maps of high values (above the 85th percentile of each tumor sample) of fBV and endomucin (Fig. 4A). Application of the Mantel statistical test revealed significant (P < 0.05) correlation (0.10 < r < 0.58; Supplementary Table A1) between the distance matrices of fBV and endomucin in 16 of 23 tumors. Endomucin staining fraction corroborated the decrease in fBV measured longitudinally over 7 days, with significantly lower tumor values determined in the cohort treated with cediranib for 7 days than in the cohort of animals treated with a single dose of cediranib; both cohorts demonstrated lower values of fBV compared with the vehicle control group (Fig. 4B and C). Calculated tumor median values of fBV strongly correlated with median values of endomucin fraction area across the cohorts (r = 0.85, P < 0.0001; Fig. 4D).

Regional differences in parametric R₂* maps visually reflected spatial variations in extravasated RBC aggregation or RBC-filled necrotic areas (defined as large areas of cell damage) in the digitized maps, which was corroborated by the location, size, shape, and orientation of hotspots on KDE maps of high values (above the 85th percentile of each tumor sample) of R₂* and RBC, automatically extracted from high-resolution H&E-stained images with an accuracy of 99.7%, after automatic registration using the CPD algorithm (Fig. 5A). The Mantel statistical test revealed significant (P < 0.05) spatial correlation (0.10 < r < 0.76; Supplementary Table S2) between the distance matrices of R₂* and RBC in 9 of 12 tumors. Calculated tumor median values of R₂* strongly correlated with the mean values of RBC fraction area (r = 0.87, P < 0.0003). Tumor median R₂* and mean value of RBC fraction also correlated with endomucin staining (r = 0.53, P = 0.08 and r = 0.64, P = 0.02, respectively; Fig. 5B–D).

We then performed a semiquantitative analysis to calibrate the tumor R₂* phenotype against pathologic evaluation of aligned H&E stained sections (Fig. 6A–I; Supplementary Table S3). For tumors with a median R₂* between ∼80 and ∼110 second⁻¹, increasing tumor median R₂* reflected increasing area/proportion of the tumor presenting with a hemorrhagic and hypervascular phenotype, characterized by large sinusoidal-shaped vessels. In comparison, nonhemorrhagic regions remained well-vascularized but with markedly reduced hemorrhage and smaller capillary-like shape vessels. In these tumors, the extravasated RBCs conserved their biconcave shape and appeared intact. Tumors with a median R₂*>110 second⁻¹ additionally presented with large necrotic regions filled with both intact and damaged RBCs. Tumors exhibiting a median R₂* < 70 second⁻¹ (comparable to tumors that progressed while on cediranib), presented a very different phenotype, characterized by the absence of hemorrhage (including in necrotic areas), reduced microvessel density with smaller more regular capillaries-like vessels, and large regions of differentiating neuroblasts (yet with only very few mature ganglion cells) arranged in islands, separated by a large amount of neuropil.

To understand the relevance of the tumor MRI phenotypes observed in the Th-MYCN GEM mouse to MYCN-driven childhood neuroblastoma, we reviewed anatomical T₂-weighted, T₁-weighted and contrast enhanced (CE-) T₁-weighted images, as well as diffusion-weighted (DW) MRI-derived apparent diffusion coefficient (ADC) maps, acquired from 19 patients with MYCN-amplified neuroblastoma at the time of diagnosis. All patients presented with abdominal tumors showing a high degree of heterogeneity (both hyper- and hypointensity) on T₂-weighted MRI and regional hyperintensity on T₁-weighted MRI (Fig. 7A–E). Based on the well-established appearance of ageing blood on both T₂- and T₁-weighted MRI in hematoma (Fig. 7F; ref. 25), this appearance suggests the presence of blood products at different stage of hemoglobin degradation. Most tumors presented a central region, which did not enhance on CE-MRI and showed elevated ADC values, suggesting the presence of large areas of nonviable tissue. Additionally, a few tumors presented with cystic components, with/without blood. Interestingly, 5 of the 19 patients also presented with contiguous or satellite masses showing homogeneous signal on both T₁ (isointense) and T₂ (hyperintense) MRI, homogeneous contrast enhancement and low ADC values, indicative of water restriction (Fig. 7B and C). This MRI phenotype suggests a nonhemorrhagic and cellular-dense tumor phenotype within these lobes.

We also compared the radiology of two cases of ganglioneuroblastoma: a ganglioneuroblastoma intermixed (GNB) and ganglioneuroblastoma nodular (GNBn), two entities that are uncommonly MYCN-amplified. Yet GNB nodular are characterized by the presence of macroscopic undifferentiated neuroblastoma nodule(s) of Schwannian stroma-poor components coexisting with ganglioneuroblastoma-intermixed of stroma-rich component or ganglioneuroma (fully differentiated) of stroma-dominant component. The neuroblastoma nodule in the GNBn was easily identifiable with low ADC values, consistent with a stroma-poor neuroblastoma phenotype, and demonstrated marked hypointensity on T₂-weighted images, hyperintensity on T₁-weighted images and high gadolinium avidity, overall suggesting that the neuroblastic nodule presented a highly vascular and hemorrhagic phenotype.

Based on the initial results obtained in the Th-MYCN model, we have introduced IS-MRI into clinical trials in patients with relapsed/refractory neuroblastoma as a proof of concept. Our initial experience based on three patients scanned at 1.5T (Fig. 7G) showed that tumors (n = 5) presented with a heterogeneous regional distribution of R₂*, with tumor median R₂* values ranging from 20 to 41 second⁻¹ (Fig. 7H), a range similar to the one measured in mice when corrected for the difference of magnetic field strength. Regionally, higher R₂* values were associated with hypointensity on T₂-weighted images, higher gadolinium avidity, and lower ADC values. Combined with the apparent hemorrhagic nature of MYCN-amplified neuroblastoma, this pilot study suggests that IS-MRI should be prospectively evaluated in children with neuroblastoma to determine its clinical utility as an imaging biomarker.

There is a current paradigm shift toward a stratified precision medicine approach for children with cancer. Central to this approach is the integration of both genomic and biological information including pharmacodynamic and predictive biomarkers of response to incorporate urgently needed novel therapies into treatment schedules. With increasing evidence for the contribution of angiogenesis in determining and predicting the biological behavior of neuroblastoma, therapeutic approaches targeting the unique vascular phenotype of neuroblastoma are being accelerated in early phase pediatric clinical trials. However, there are currently no validated biomarkers predicting the clinical response or therapeutic benefits to anti-angiogenic therapy.

Although MRI is becoming the preferred clinical imaging method in the management of children with neuroblastoma particularly at diagnosis and for routine follow-up, the unique ability of advanced functional MRI techniques to noninvasively quantify changes in tumor vascular architecture and function has not yet been fully exploited (26). In this study, we have utilized the multiple intra- and para-anatomical tumor landmarks detectable on conventional T₂-weighted MRI in the Th-MYCN mouse, together with the CDP algorithm for registration to develop a robust MRI-histopathology correlation pipeline. The use of KDE hotspot mapping and nonparametric Mantel statistical test demonstrated that fBV, measured in vivo by SC-MRI, spatially correlates with stained area for the endothelial cell marker endomucin in tumors arising in the Th-MYCN mouse model of neuroblastoma. This approach thus validates fBV as a sensitive and but also a specific imaging biomarker of tumor vascular perfusion, and its therapeutic modulation by the potent VEGFR inhibitor cediranib. Furthermore, we found a negative correlation between tumor fBV and cediranib-mediated changes in fBV, suggesting baseline fBV as a potential predictive biomarker of vascular response to VEGFR inhibition. Importantly, our data suggest that both baseline tumor fBV and R₂* are predictive of the longer-term tumor response to VEGFR-targeted therapies such as cediranib.

Several preclinical studies have demonstrated that quantitation of tumor fBV using SC-MRI provides a sensitive biomarker of response to vascular-targeted agents in vivo (10, 11, 13). The data herein highlight the potential of SC-MRI to assess neuroblastoma vascular response to VEGF signaling inhibitors (27–30), and to MYCN-targeted therapeutics whose mechanism of action is predicted to elicit an antiangiogenic effect (31–34). For example, direct silencing of MYCN through bromodomain extra terminal (BET) domain inhibition, destabilization of MYCN protein via the selective targeting of Aurora A kinase or mTOR/PI3k signaling, and inhibition of anaplastic lymphoma kinase (ALK) and RET kinases (reviewed in ref. 35) have all been linked with angiogenic blockade in aligned preclinical trials against neuroblastoma or other pediatric solid tumors (32, 33, 36–38).

The high vascular permeability and density present in the Th-MYCN tumors and detected by IS- and SC-MRI illustrate the archetypal VEGF-driven vascular remodeling triggered by members of the MYC family oncogenes (32, 39). There is increasing evidence that enhanced angiogenic signaling is part of a wider MYC-driven adaptive program to nutrient deprivation, to which MYC-driven tumors rapidly develop an addiction to maintain their growth (39, 40). As a result, any blockade of this program via MYC deactivation, metabolic inhibition, or vascular blockade triggers a wave of apoptosis, resulting in the unique rapid tumor debulking observed in MYC-driven tumors (39, 40). The correlation between baseline fBV and R₂* and cediranib-induced tumor debulking established in this study illustrates a similar dependence on MYCN-driven angiogenic and vascular remodeling in the Th-MYCN GEM model. In contrast, tumors at the lower range of baseline R₂* and fBV values, which progress despite cediranib treatment, presented with a more normalized vasculature. We also identified tumors with large areas of differentiating neuroblasts, a phenotype similar to that in MYCN-driven GEM models of neuroblastoma harboring constitutive activation of ALK, and in which reduced VEGFR signaling has been reported (31, 41, 42). Clinically, differentiated neuroblastoma displays low vascularity (2, 3), partly due to the release of strong antiangiogenic factors that directly antagonize the effect of VEGF (43, 44). Additionally, in differentiating, Schwannian-stroma poor neuroblastoma, VEGF-mediated cross-talk between differentiating neuroblasts and endothelial cells has been identified, which mimics normal vascular maturation in the developing nervous system (45). Overall in this study, reduced tumor sensitivity to cediranib was associated with a more stable vascular phenotype, suggesting a reduced dependence of these nonresponsive tumors on VEGF signaling to sustain growth.

Genetically engineered models of cancer have been shown to faithfully recapitulate the major molecular and histopathologic features of human cancer, and to have a strong predictive power for clinical drug response and resistance (46). Neuroblastoma, and pediatric cancers in general, are particularly amenable for genetically modeling as they are only driven by a few genetic aberrations. Tumors arising in the Th-MYCN mouse model of neuroblastoma are stroma-poor, undifferentiated, or poorly differentiated with a high mitosis–karyorrhexis index, closely resembling the histology of MYCN-amplified tumors in the >18-month age group (15). The high tumor fBV values of ∼20% reported here are consistent with the unique hypervascular phenotype of childhood neuroblastoma. The median endomucin staining area in the Th-MYCN tumors (2.3%) is consistent with the mean stained area of endothelial cell marker CD31 (1.7%) in a cohort of 458 primary childhood neuroblastomas (4). Structurally, the large areas of irregular sinusoidal vessels found in the hemorrhagic regions of Th-MYCN tumors emulate the major unstable vascular patterns associated with unfavorable clinical prognosis. In contrast, tumors with lower R₂*, differentiated or not, presented a more normalized vascular pattern associated with favorable prognosis factors in primary neuroblastoma. Our retrospective radiological analysis of 19 cases of MYCN-amplified neuroblastoma revealed that all tumors presented with a MRI phenotype consistent with hemorrhage. Additionally, our initial experience with IS-MRI in children with refractory/relapse neuroblastoma demonstrated a range of tumor median R₂* values, comparable to those measured in the Th-MYCN model when corrected for the difference in magnetic field strength (47). Collectively, these data demonstrate that the Th-MYCN GEM model of neuroblastoma recapitulates the pathophysiology of childhood neuroblastoma and provide a unique clinically relevant and information-rich platform to identify and validate imaging biomarkers of the neuroblastoma microenvironment.

DCE-MRI and CT-based functional imaging are commonly used to evaluate tumor perfusion in the clinic. Yet their associated biomarkers are difficult to interpret as they reflect both vascular perfusion and permeability. More importantly, the acquisition of robust DCE-MRI data can be challenging in young children (7) and CT-based functional imaging methodologies should be preferentially avoided due to the higher inherent sensitivity of children to the negative effects of ionizing radiation (6). The data herein provide a strong rationale for the incorporation of both IS- and SC-MRI into functional imaging-embedded clinical trials for new therapeutic strategies that directly or indirectly modulate neuroblastoma vascular phenotype. IS-MRI is a rapid, quantitative, safe and noninvasive R₂* relaxometry measurement that has become the reference measurement to assess both liver and cardiac iron content and a primary endpoint in pediatric clinical trials in highly transfused children suffering from hematological diseases who are at risk of liver and cardiac iron overload (48). The multigradient recalled echo MRI sequence is universally available on clinical MRI scanners and IS-MRI can be easily incorporated in the neuroblastoma clinic, as demonstrated herein, adding only two and half minutes to the current routine scanning session. IS-MRI has been incorporated in an ancillary functional imaging study of the BEACON Neuroblastoma trial. This study also demonstrates that hemorrhage detectable with IS-MRI, or the absence thereof, could hold prognostic/diagnostic value. Clinically, the hemorrhagic nature of undifferentiated neuroblastoma is well described, yet its routine reporting remains only anecdotal (49). Its value for risk-stratification could be explored by introducing noninvasive IS-MRI at the time of diagnosis. As a matter of fact, introducing susceptibility-weighted imaging (SW-MRI), a variant of IS-MRI, which combines the information from both the magnitude and the phase of the MR signal, would quantify hemorrhage with the added value of detecting calcification (opposite phase to hemorrhage). SW-MRI can assist in the differential diagnosis of neuroblastoma (frequently calcified) versus Wilms' tumors (seldom calcified) as this not easily detected on conventional MRI. SC-MRI, which combines R₂* mapping with the use of USPIO particles, does not require bolus timing or complex image acquisition schemes, and provides an attractive steady-state imaging alternative to DCE-MRI. Recent clinical studies highlighting the safe off-label use of the USPIO preparation ferumoxytol for MRI in adults and children is a promising step-forward, warranting further evaluation of the approach in clinical trials (50).

In summary, the robust MRI-pathology cross-validation methodology used in this study validates IS- and SC-MRI as robust noninvasive imaging techniques to characterize, quantify, and map the unique vascular phenotype of neuroblastoma and its therapeutic modulation. With the central role of angiogenesis in determining and predicting the clinical behavior of neuroblastoma, baseline fBV and R₂*, have the potential to provide diagnostics and prognosis information at the time of diagnosis and provide currently unavailable predictive and pharmacodynamic biomarkers for antiangiogenic therapy against neuroblastoma. Finally, this study highlights the potential role of functional MRI in delivering precision medicine to children with neuroblastoma.

M.D. Blackledge has ownership interest (including stock, patents, etc.) in patent. A.D.J. Pearson is a consultant/advisory board member of Advisory. L. Moreno is a consultant/advisory board member of Novartis, Astra Zeneca, Roche Genentech, Mundipharma, Bayer, Amgen; received Celgene Educational event's travel grant; Novartis Educational event's member of DMC; received Mundipharma travel grant and Amgen travel grant; and is also a member of SIOPEN Executive Commitee, an organization that receives royalties from dinutuximab beta (Eusa Pharma). J. Anderson is a Chief Medical and Scientific Advisor at TC Biopharm; has received speakers bureau honoraria from Eusa Pharma; has ownership interest (including stock, patents, etc.) in Autolus Ltd.; and is a consultant/advisory board member of Roche. Y. Yuan is a consultant/advisory board member of Consultancy for Merck. No potential conflicts of interest were disclosed by the other authors.

Conception and design: K. Zormpas-Petridis, A.D.J. Pearson, J. Anderson, N. Sebire, Y. Yuan, L. Chesler, S.P. Robinson, Y. Jamin

Development of methodology: K. Zormpas-Petridis, M.D. Blackledge, E. Poon, A. Koers, S.P. Robinson, Y. Jamin

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): K. Zormpas-Petridis, F. Carceller, E. Poon, C.M. McErlean, G. Barone, A. Koers, S.J. Vaidya, L. Moreno, N. Sebire, K. McHugh, L. Chesler, S.P. Robinson, Y. Jamin

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): K. Zormpas-Petridis, N.P. Jerome, E. Poon, M. Clarke, A.D.J. Pearson, L. Moreno, K. McHugh, D.-M. Koh, Y. Jamin

Writing, review, and/or revision of the manuscript: K. Zormpas-Petridis, N.P. Jerome, M.D. Blackledge, F. Carceller, E. Poon, G. Barone, S.J. Vaidya, L.V. Marshall, A.D.J. Pearson, L. Moreno, J. Anderson, N. Sebire, K. McHugh, D.-M. Koh, S.P. Robinson, Y. Jamin

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): K. Zormpas-Petridis, M.D. Blackledge, F. Carceller, L. Chesler, Y. Jamin

Study supervision: M.D. Blackledge, Y. Yuan, Y. Jamin

We thank AstraZeneca for the supply of cediranib. Y. Jamin received a Children with Cancer UK Research Fellowship (2014/176). Y. Jamin and S.P. Robinson received Rosetrees Trust grant M593. E. Poon and L. Chesler received Children with Cancer UK Project Grant (2014/174). L. Chesler received Cancer Research UK Program Grant (C34648/A18339 and C34648/A14610). L. Moreno received Instituto de Salud Carlos III through Rio Hortega (CM12/00260) and Juan Rodes (JR15/00041) contracts. J. Anderson received a GOSHCC research leadership award. This work was supported in part by a Cancer Research UK and EPSRC to the Cancer Imaging Centre at ICR, in association with the MRC and Department of Health (England) (C1060/A10334 and C1060/A16464), NHS funding to the NIHR Biomedicine Research Centre and the Clinical Research Facility in Imaging, the NIHR GOSH Biomedical Research Centre, the Oak Foundation to the Royal Marsden.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.