Noninvasive MRI Native T₁ Mapping Detects Response to MYCN-targeted Therapies in the Th-MYCN Model of Neuroblastoma

Neuroblastoma is a tumor arising from anomalies in the embryonic sympathoadrenal lineage of the neural crest in children (1). Despite intensive first-line multimodal therapy, neuroblastoma still accounts for 13% of all cancer-related deaths in children due to resistant, relapsing, and systemic disease. Promising novel targeted therapeutic approaches against neuroblastoma are being developed and include small-molecule inhibitors as well as epigenetic, noncoding RNA, and cell-based immunologic therapies (2–5). Amplification of the proto-oncogene MYCN is the most common genomic aberration, which defines a subgroup of children with a high-risk disease. MYCN plays a central role in the biology of high-risk neuroblastoma and as such represents a major therapeutic target.

The application of the mouse hospital and coclinical trial concept represents a clear paradigm shift in neuroblastoma translational research (2, 6). This approach integrates more advanced mouse modeling, including genetically engineered mouse (GEM) models, such as the Th-MYCN mouse (7), to accelerate the discovery and evaluation of novel therapeutic strategies, and helps shape the clinical trial pipeline priorities for children with high-risk disease. Small-molecule inhibitors targeting the stability of MYCN protein have shown strong antitumor activity in the Th-MYCN model and are being evaluated in early phase pediatric clinical trials (2, 8–10). These include the selective inhibitor of Aurora A kinase, alisertib (MLN8237, NCT01601535), and selective inhibitors of mTOR activity (NCT01331135, NCT01467986, NCT01625351, NCT02343718, NCT02574728, NCT02638428, NCT02813135).

Evaluation of response to treatment in children with neuroblastoma is based on Response Evaluation Criteria in Solid Tumours (RECIST) using noninvasive anatomic imaging such as CT or MRI. The revised International Neuroblastoma Response Criteria (INRC) guidelines now also include sensitive nuclear medicine-based functional imaging approaches such [metaiodobenzylguanidine (MIBG) scans and [F-18]2-fluoro-2-deoxyglucose positron emission tomography/CT (FDG PET/CT)] for the assessment of bone and bone marrow metastatic disease, present in 50% of cases (11). In addition to providing more accurate detection of active disease, functional imaging techniques may also provide biomarkers of response to novel therapies in neuroblastoma clinical trials, in which conventional pharmacodynamic biomarkers can be difficult to evaluate. 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 anatomic information at diagnosis and follow up while sparing exposure to ionizing radiation associated with CT. Advanced MRI-based functional imaging techniques can be used to define quantitative imaging biomarkers that inform on biologically relevant structure–function relationships in pediatric cancers in vivo (12).

The mouse hospital concept provides a unique opportunity to evaluate predictive and prognostic imaging biomarkers of response in neuroblastoma and to perform the close imaging–pathology correlation necessary to understand the biological processes underpinning the imaging measurement and provide the stringent validation needed before they can be deployed clinically. We have previously demonstrated that a reduction in the tumor native spin-lattice relaxation time T₁, measured using inversion recovery true fast imaging with steady-state precession (IR-TrueFISP) MRI, can provide a sensitive biomarker of response to cyclophosphamide, which is a usual component of various frontline protocols for neuroblastoma, and antivascular therapies in the Th-MYCN model (13).

In this study, we evaluate how a reduction in native tumor T₁ provides a robust biomarker of response to alisertib and the mTOR inhibitor vistusertib (AZD2014) in the Th-MYCN model. By comparing native T₁ maps with those derived from multiparametric MRI and computational pathology, we demonstrate that native T₁ mapping (the voxel-wise quantification of T₁) is sensitive to the rich histologic presentation of neuroblastoma, including regional differences in undifferentiated, differentiating and apoptotic neuroblast density. This study demonstrates the potential application of T₁ mapping for diagnosis/prognosis, surgical planning, and the evaluation of novel therapies for children with neuroblastoma.

All experiments were performed in accordance with the local ethical review panel, 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 (14) and the ARRIVE (animal research: reporting in vivo experiments) guidelines (15).

Transgenic Th-MYCN mice were genotyped to detect the presence of the human MYCN transgene (7). The study was performed using both male and female homozygous mice, which developed a single palpable abdominal tumor at 40–80 days old with 100% penetrance. Tumor development was monitored weekly by palpation by an experienced animal technician. A total of 46 mice were enrolled with a median tumor volume of 861 ± 86 mm³ (derived from T₂-weighted MRI; median ± 1 SEM, ranging from 280 to 2,557 mm³). MRI was performed prior to treatment (day 0). Mice were left to recover for 24 hours, and then treated (Day 1) with 30 mg/kg orally of alisertib (MLN8237, purchased from Selleckchem, n = 11) or vehicle (10% 2-hydroxypropyl β-cyclodextrin, 1% NaHCO₃, n = 9), or 25 mg/kg orally of vistusertib (AZD2014, obtained under material transfer agreement with AstraZeneca, n = 14) or vehicle (5% DMSO, 95% PEG300, n = 12). Posttreatment MRI was performed 24 hours after treatment started (Day 2). Mice were housed in specific pathogen-free rooms in autoclaved, aseptic microisolator cages (maximum of 4 mice per cage) and allowed access to sterile food and water ad libitum.

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). Core temperature was maintained at approximately 37°C with warm air blown through the magnet bore.

For all the mice, contiguous anatomical T₂-weighted transverse images were acquired through the mouse abdomen for the quantification of tumor volume, optimization of the local field homogeneity using the FASTmap algorithm, and for planning the subsequent multiparametric MRI measurements. In addition to IR-TrueFISP MRI for quantification of the spin-lattice (T₁) and spin-spin (T₂) relaxation times, these also included measurement of the apparent diffusion coefficient (ADC), the transverse relaxation rate R₂* and the magnetization transfer ratio (MTR) using the MRI sequences and parameters listed in Supplementary Table S1.

Tumor volumes were determined using segmentation from regions of interest drawn on each tumor-containing T₂-weighted MRI slice using OsiriX. All the multiparametric MRI data were fitted voxelwise using in-house software (ImageView, working under IDL, ITT) with a robust Bayesian approach that provided estimates of T₁, T₂, ADC and R₂*. MTR (%) was calculated as MTR = (1-M_25ppm/M_100ppm) × 100 and fitted voxelwise using in-house code written in Matlab (The Mathworks).

Guided by T₂-weighted MRI, tumors were carefully excised and orientated for histopathologic processing. Formalin-fixed and paraffin-embedded tumors were sectioned (3 μm) and stained with hematoxylin and eosin (H&E). Whole-slide H&E images were digitized using a Hamamatsu NanoZoomer XR scanner (20× magnification, 0.46 μm resolution, Hamamatsu). Histology images were subsequently split into tiles of 2,000 × 2,000 pixels (jpeg) for computational efficiency using Bio-Formats (https://www.openmicroscopy.org/bio-formats/).

For each tumor, the MRI slice of interest was visually aligned with the digitized whole-slide hematoxylin and eosin (H&E)-stained image using anatomic landmarks as described recently (16).

Image processing was carried out using CRImage (17). First, cell nuclei were extracted from H&E staining by Otsu thresholding (18). Noisy image structures were then deleted using morphologic opening. The clustered nuclei were separated by the Watershed algorithm. For every nucleus, 91 morphologic (19), three local context, and 46 cell-cytoplasm features were extracted. A support vector machine (SVM) with a radial basis function (RBF, γ = 1/number_of_features) kernel was trained with these features, based on annotations provided by a neuropathologist on 16,320 cells from 7 whole-slide samples. Cells were subsequently classified into five categories: undifferentiated neuroblasts, differentiating neuroblasts, apoptotic cells, lymphocytes, stromal cells.

Whole-slide images of cells were processed to match the MRI resolution (234 × 234 μm), with the number of segmented cells and classified cells within 518 × 518 pixel regions representing a single pixel in the final cell density maps. Density maps were normalized to their sample's maximum number of cells/classified cells to facilitate the evaluation intratumor heterogeneity.

This was performed as recently described using the automatic coherent point drift (CPD) algorithm (16, 20). First, density maps of all the segmented cells were nonrigidly registered to the T₁ images based on features extracted by a Canny edge detector. The same transformation was subsequently applied to the density maps of each classified cell category.

The first parametric maps were divided into subregions of high and low values using thresholds summarized in Supplementary Table S2. A binary mask was created for each subregion and applied to the second parametric map. This analysis was performed in 13 tumors across both vehicle cohorts for which precise MRI-Pathology registration was possible. Statistical comparison of subregional median values between the two parametric maps was performed and the process was repeated in reverse.

Statistical analysis was performed with GraphPad Prism 7 (GraphPad Software Inc.). The mean values for tumor volume, and the mean of the median values for all the quantitative MRI parameters were used. All the absolute and treatment-induced relative changes in MRI parameters were assumed to be normally distributed, which was confirmed using the D'Agostino-Pearson omnibus K2 normality test. Student two-tailed t test was used to assess any significant differences in quantitative MRI parameters and tumor volume upon treatment (paired), and in the magnitude of these changes compared with the control cohort (unpaired), with a 1% level of significance. Further statistical analysis was performed with the Bonferroni correction (n = 5). Any significant differences between groups identified in the subregional analysis were identified using the Wilcoxon signed rank test with a 5% level of significance. Significant correlations were determined using linear regression analysis, confirmed by using the robust regression and outlier removal approach (21).

The Th-MYCN GEM model of neuroblastoma recapitulates the aggressiveness of the clinical disease, with an observed average 31 ± 4% increase in tumor volume measured over the 48 h experimental timecourse (Table 1A and B; Fig. 1). Despite this, tumor median values for all the MRI parameters remained stable over 48 h in the vehicle treated cohorts (coefficients of variation CoV_T1 = 2.4%, CoV_T2 = 5.8%, CoV_ADC = 14.0%, CoV_R2* = 11.7%, CoV_MTR = 6.2%). There was no significant difference in tumor volume between the different treatment cohorts at the time of enrollment (Supplementary Fig. S1). Treatment with either alisertib or vistusertib led to a highly significant reduction in native T₁ (−9.3 ± 0.9% and −5.4 ± 1.1%, both P < 0.0001) and was associated with a significant reduction in tumor volume with vistusertib (−42 ± 5.1%, P < 0.0001) but not alisertib, although a reduction in tumor volume was seen in 9 of 11 treated mice (Supplementary Fig. S2A and S2B). Both the alisertib- and vistusertib-treated groups elicited significant antitumor activity when compared with their respective vehicle control cohorts (both P < 0.0001). No significant changes in tumor native T₂, ADC, R₂*, or MTR were determined following treatment with either alisertib or vistusertib, nor any treatment-induced relative changes compared with vehicle controls.

Tumors arising in the Th-MYCN model present a characteristically hemorrhagic phenotype with large areas of extravasated red blood cells (RBC). The transverse relaxation rate R₂* is sensitive to the concentration of paramagnetic deoxyhemoglobin associated with deoxygenated RBCs, hence neuroblastomas typically exhibit relatively high R₂* values. We recently validated R₂* as a robust biomarker for mapping RBC distribution in this tumor model (16).

Visual comparison of native T₁ and R₂* maps (Fig. 1) showed that regions of high R₂* colocalized with regions of low native T₁. Retrospective analysis of measurements made in 71 untreated tumors arising in GEM models of neuroblastoma (Supplementary Methods) revealed that the median native T₁ inversely correlated with native median R₂* (r = −0.59, P < 0.0001; Fig. 2A). Subregional analysis using established empirical R₂* threshold values [R₂*< 70 s⁻¹ as no hemorrhage (16, 22) and >250 s⁻¹ as purely RBC, and mixed regions of neuroblasts and RBC for the R₂* values between] identified significantly different values of T₁ associated with low (<70 s⁻¹), intermediate and high (>250 s⁻¹) R₂* (Fig. 2B). Comparison of the relative changes in median T₁ and R₂* with treatment revealed a significant negative correlation [r = −0.78, P = 0.002 with Bonferroni correction (n = 5), Fig. 2C]. Importantly, both positive and negative changes in tumor R₂* occurred with treatment, thereby accentuating the sensitivity of native T₁ to RBC deposition, but excludes changes in the content of paramagnetic RBCs or other such species as the main cause of reduction in T₁ upon treatment.

We then focused on the major histologic component of these tumors, i.e., the dense cellular network. We trained a cell classifier, which allowed the robust segmentation and classification of five different classes of cells with an overall accuracy of 95.3% (Fig. 3A and B, confusion matrix shown in Supplementary Table S3; Supplementary Fig. S3A and S3B). We generated parametric maps of undifferentiated neuroblasts and apoptotic cells density and compared them with spatially registered native T₁ maps. In vehicle control tumors, regions exhibiting high values of T₁ colocalized with dense regions of undifferentiated neuroblasts (Fig. 4). Threshold-based subregional analysis confirmed that regions with higher T₁ values corresponded to areas of increased density of undifferentiated neuroblasts and, reciprocally, regions with higher neuroblast density had higher native T₁ values (Fig. 5A and B; Supplementary Table S2). Interestingly, areas dense in apoptotic cells in vehicle control tumors also corresponded to regions of lower native T₁. The widespread reduction in T₁ seen in the vistusertib-treated tumors was associated with a widespread and significantly higher fraction of apoptotic cells (57% ± 3% compared with 16% ± 3% in vehicle control, P < 0.0001) and tissue damage, concomitant with a significantly lower fraction of undifferentiated neuroblasts (21% ± 3% compared with 64 ± 4% in vehicle control, P < 0.0001; Fig. 5C). The more modest but widespread reduction in T₁ in the alisertib-treated tumors was not associated with any detectable differences in the fraction of apoptotic or undifferentiated neuroblasts on corresponding H&E staining, as confirmed by cleaved caspase-3 staining (Supplementary Fig. S4). Note that this response was, however, associated with the reduction in tumor volume seen in 9 of the 11 mice treated (Supplementary Fig. S2), and the absence of any significant difference in tumor T₁ posttreatment between the alisertib and vehicle control cohorts (contrary to that seen with vistusertib, P < 0.0001).

Combining the MRI data from vistusertib- and alisertib-treated mice with matched histopathology revealed a significant negative correlation between treatment-induced reduction in T₁ over 24 hours and the proportion of apoptotic neuroblasts present in the tumor at the study endpoint (r = −0.55, P = 0.04, Fig. 5D). Combining the MRI data from vehicle control, vistusertib-, and alisertib-treated tumors with matched histopathology showed a positive correlation between median T₁ and the ratio of undifferentiated neuroblasts (r = 0.70, P < 0.0001, Fig. 5E) and a negative correlation with the fraction of apoptotic cells (r = −0.63, P = 0.006, Fig. 5F).

We identified three tumors exhibiting a significant amount of differentiating neuroblasts (yet with only very few mature ganglion cells). In these tumors (Fig. 6), previously shown to have very low levels of hemorrhage (16), regional differences in T₁ visually and spatially corresponded to differences in undifferentiated neuroblast density, with regions of low T₁ and low-density undifferentiated neuroblasts corresponding to hotspots of differentiating neuroblasts, arranged in islands separated by a large amount of neuropil or simply interspersed with undifferentiated neuroblasts.

In pediatric oncology, the difficulty of obtaining posttherapy surgical biopsies is hindering the development of robust predictive/prognostic pharmacodynamic biomarkers of response urgently needed to accelerate the clinical evaluation of more effective and safer therapeutic strategies. Recent large molecular profiling protocols at national level (23–25), advocate for biopsies at the time of relapse to identify actionable alterations in pediatric recurrent cancers. In this regard, advanced MRI-based functional imaging techniques that can define quantitative biomarkers to noninvasively visualize spatial variations and temporal evolution of tissue structure–function in vivo are being actively explored (12). Early imaging biomarker development demands close imaging–pathology correlation, to understand the biological processes underpinning the imaging measurement, before they can be routinely deployed in the clinic (26).

In this study, we demonstrate how T₁ mapping is sensitive to the rich histologic presentation of neuroblastoma, and can provide a sensitive biomarker of response to two clinically relevant MYCN-targeted therapeutics in the Th-MYCN GEM model of neuroblastoma. We have continued to exploit computational pathology methodologies to enable the precise comparison of MRI parametric maps with whole-slide digitized pathology (16). Importantly, the Th-MYCN GEM model recapitulates the chemosensitivity and pathophysiology of high-risk, MYCN-amplified neuroblastoma, including a dense and hemorrhagic vascular phenotype and undifferentiated or poorly differentiated tumor phenotype with a high mitosis–karyorrhexis index, indicative of both a high level of proliferation and apoptosis (27).

Using this approach, we have identified, and confirmed using quantitative subregional analysis, four major determinants of the regional heterogeneity observed on native T₁ maps: (i) regions with high T₁ values corresponded to hotspots of undifferentiated neuroblasts, characterized by a high level of proliferation, whereas (ii) regions rich in differentiating neuroblasts exhibited lower T₁ values, and both (iii) regions with large amounts of extravasated RBCs, and (iv) large areas of cell damage, with or without RBCs, were both associated with very low T₁ values. The association between T₁ and extravasated RBCs was an expected finding consistent with the linear relationship of blood T₁ with hematocrit level and hemorrhage (28).

Our data with vistusertib indicate that the reduction in native T₁ was associated with a shift in tumor composition characterized by rapid loss of tumor regions with higher T₁ values, a consequence of cell death, with the posttherapy tumor T₁ values determined by dying and remaining hemorrhagic fractions. A similar conclusion can be drawn on the contrast mechanism underpinning the reduction in T₁ upon treatment with alisertib, based on both the known mechanism of response to alisertib through apoptosis in this model and the observed reduction in tumor volume in our study (10). However, we could not confirm this using endpoint histopathologic assessment, potentially due to the high intertumor heterogeneity both in terms of the amount of apoptosis present at the time of enrollment (as shown by the endpoint histopathology in the vehicle cohorts) and in the actual response to alisertib treatment in this model as recently reported (10). The absence of any significant relative change in R₂*, a validated biomarker for RBCs (29), or T₂, ADC, and MTR, which all relate to tissue water content/binding, strongly suggests that the overall decrease in T₁ is being driven by the loss of the tissue fraction with high T₁ values, that is, regions with a high density of undifferentiated neuroblasts, rather than a gain of new MRI contrast, for example, that resulting from cell death–mediated release of paramagnetic ions (30, 31).

By definition, the spin-lattice T₁ relaxation time refers to the interaction or energy transfer between the excited ¹H spin and the molecules within the surrounding molecular structure. The T₁ value, that is, the efficacy of the spin-lattice relaxation, is dependent on molecular tumbling of the molecule in which the proton resides. For MRI applications, this molecule is primarily water, which can be present in three states associated with different T₁ values: (i) free water (free to move, high T₁), (ii) "structured" water (bound to a macromolecule by a single bond where molecular tumbling is still possible, lower T₁), (iii) "bound" water (found in solids, bound by multiple bonds, high T₁). The general consensus is that the reduced tissue T₁ of structured water is a consequence of its interaction with proteins and other macromolecules. Tissue T₁ thus depends on compartmentalization of structured water and the amount of molecular crowding within each different compartment. Cancer cells and tumor tissue typically have elevated T₁ values compared with normal tissues, the original observation that demonstrated the potential of MRI for cancer diagnosis. Elevated tumor T₁ remains attributed to a difference in intracellular water structure and order compared with normal cells (32, 33). T₁ has also been suggested to change during cell cycle and mitosis in vitro, a phenomenon also attributed to different levels of water–macromolecule interactions (34, 35). However, very early work in MRI-detectable isolated large cells such as Xenopus oocytes and Aplysia neurons confirmed that cell nuclei exhibit higher T₁ values than the cytoplasm (1.85 vs. 1.2 seconds, respectively, for Xenopus oocytes at 7T), and that degradation/permeabilization of the nuclear envelope causes an equilibration of T₁ values (36, 37). A more recent study reported anomalously rapid hydration water diffusion dynamics near DNA surfaces, which demonstrates that water interacts differently with DNA compared with protein. More precisely, water behaves like free water near DNA (38), which would explain both the higher nuclear T₁, and the change in T₁ observed during mitosis when the chromatin is condensed and DNA is less accessible to water molecules and the nuclear membrane completely disappears.

Poorly or undifferentiated neuroblastoma are defined as small round nuclei with stippled chromatin (diffuse open chromatin) with scant eosinophilic cytoplasm and indistinct cell borders. This definition is thus self-explanatory for the higher T₁ values reported here in areas of dense, undifferentiated neuroblasts (dense cells with a high nuclear/cytoplasmic ratio and minimal extracellular compartment). We can also assume that any reduction in undifferentiated cell density, or change in cell phenotype and/or intracellular compartmentalization, in a sufficiently large number of cells would thus result in a reduction in T₁ (39). The reduced native T₁ associated with dense areas of differentiating neuroblasts, characterized by lower nuclear-to-cytoplasmic ratio, lower cell density, and possibly surrounded by abundant eosinophilic neuropil, supports this hypothesis. Many of the events occurring during apoptosis, including water loss, pyknosis, and karyorrhexis, would also align with a reduction in T₁ if happening in a sufficient number of cells (40). Interestingly, both pyknosis and karyorrhexis are steps common to apoptosis, necrosis, and senescence, indicating a potential generic sensitivity of T₁ to cell death. As virtually all undifferentiated neuroblasts in this model are positive (and apoptotic cells negative) for the proliferation marker Ki67 (27), this hypothesis corroborates the studies by McSheehy and colleagues showing that a reduction in native T₁ positively correlates with Ki67 staining (30, 31).

The differential diagnosis and risk-stratification for children with neuroblastoma is based on criteria including histologic features such as the grade of tumor differentiation. The sensitivity of T₁ mapping to regions rich in undifferentiated, apoptotic, or differentiating neuroblasts seen in the Th-MYCN model herein suggests its potential to noninvasively classify tumors by favorable and unfavorable histology. It may also help identify anaplastic lymphoma kinase (ALK)-positive regions, mutations associated with poor outcome in neuroblastoma, and for which small-molecule inhibitors are currently being developed. Interestingly, ALK mutations have been shown to be associated with a differentiating molecular signature, confirmed at a pathologic level in several MYCN- and ALK-mutated GEM models (41–44). T₁ mapping may also afford additional prognostic value in confirmed cases of neuroblastoma, in which high cellular density of proliferative cells is associated with poor outcome, whereas a high density of apoptosis suggests a more favorable outcome (45). Finally, T₁ mapping may help identify the nature of tumors following the induction phase of first-line therapy, where it is uncertain if a mass is comprised of undifferentiated neuroblastoma or apoptotic or differentiated disease. In this regard, T₁ mapping would provide additional and complementary information to semiquantitative molecular imaging strategies such as MIBG and FDG-PET scans and help confirm the nature and heterogeneity of the disease associated with MIBG avid/nonavid and FDG (positive/negative) disease. This is important as discrepancies exist between the expression of the norepinephrine transporter (NET), responsible for the uptake of MIBG, and the presence of an aggressive cellular phenotype. These include MIBG nonavid disease that presents in approximately 10% of children, and reduced NET protein expression in high-risk MYCN-amplified disease (46). In addition, targeted therapies against MYCN or ALK can lead to the modulation of vascular perfusion (and hence the delivery of radiolabeled MIBG and FDG), glucose uptake, and NET expression, which may potentially lead to a change in MIBG avidity, which does not reflect, or makes it difficult to assess, changes in the extent of active disease using the current INRC guidelines. Treatment with the histone deacetylase inhibitor vorinostat has for example been shown to be effective against neuroblastoma while increasing NET transporter expression in neuroblastoma (47). T₁ mapping has the potential to help improve the accuracy of detection of active disease for enhance surgical tissue sampling, surgical resection planning, and response assessment.

The voxel-wise quantification of T₁ is an essential component of many MRI-based functional and molecular imaging techniques being developed to study the tumor microenvironment and for the evaluation of novel targeted therapies, including immunotherapy (48–53). DCE-MRI, arterial spin labeling (ASL-) MRI and oxygen-enhanced T₁-MRI are being evaluated clinically to assess tumor vascular perfusion/permeability and hypoxia. However, native tumor T₁ maps acquired in the clinic are often only estimated and seldom reviewed or interpreted (13). In contrast, the clinical adoption of native T₁ mapping has increased the potential for the noninvasive and differential diagnosis of cardiac pathology (54, 55) and the staging of chronic liver disease (56). The cardiac MR experience has shown that T₁ mapping is simple to perform and analyze, minimally subjective, and highly reproducible (∼2% CoV for a modified Look-Locker inversion recovery MOLLI sequence over 24 hours; ref. 57). However, there are many acquisition schemes available for T₁ mapping, and the measured T₁ will depend on the precision and reproducibility of each scheme, and how is it affected by motion, flow, and off-resonance effects. In our study, one of the advantages of the IR-TrueFISP technique, aside from its high accuracy, is that it is inherently flow compensated in the directions of slice selection and readout, especially at the blood velocity observed in tumors (58), allowing us to exclude changes in vascular flow as a source of reduction in native T₁. Moving forward, including T₁ mapping in an ethically approved clinical study within the standard-of-care first-line chemotherapy would provide the study to rapidly evaluate and validate T₁ mapping potential for the neuroblastoma clinic. Such a study would also inform on the potential of native T₁ mapping to help better define bone and bone marrow metastasis and its response to treatment.

The potential value of native T₁ reduction as a generic biomarker of early tumor response to therapy was first demonstrated by McSheehy and colleagues (30, 31). By understanding the spatial relationship of T₁ mapping with regional variations in neuroblastoma phenotype, our study sheds new light into the biology underpinning native T₁ contrast, based on cell anatomy. Our data strongly supports the use of T₁ mapping as a generic approach to assess early response to cancer treatment, especially since (i) the “small-blue-round-cell tumor” phenotype, characterized by monotonous proliferations of small, undifferentiated, or poorly differentiated cells with scant cytoplasm, is actually used to refer to the phenotype of a large group of highly aggressive tumors, including many high-risk pediatric malignancies such rhabdomyosarcoma and medulloblastoma (and adult cancers such as certain subtypes of sarcoma, carcinoma, lymphoma, and melanoma) and (ii) both pyknosis and karyorrhexis are common steps to the major cell death processes. However, it would be important to understand the disease or tissue-specific factors, which may also affect native T₁ including the presence of edema, fat, or melanin.

In summary, our study demonstrates that native T₁ mapping can precisely and quantitatively map the rich histopathology of neuroblastoma tumors and its modulation by MYCN-targeted therapeutics in the clinically relevant Th-MYCN model of neuroblastoma. By providing strong evidence for the sensitivity of native T₁ to dense areas of undifferentiated neuroblasts, our data suggest further application for diagnosis, risk stratification, and surgical planning, and that its potential as a biomarker of successful response to therapy could be extended to larger subsets of aggressive pediatric and adult tumors. Widely available on conventional clinical scanners, our study provides a strong rationale for the incorporation of T₁ mapping both at the time of diagnosis and in early phase clinical trials to guide clinical decision making and the delivery of precision medicine to children with neuroblastoma.

M.D. Blackledge reports personal fees from Bayer AG and personal fees from GenesisCare outside the submitted work. L. Moreno has participated in advisory boards for Novartis, AstraZeneca, Roche/Genentech, Mundipharma, Bayer, and Amgen and has received honoraria from Celgene and Novartis for an educational event and travel grants from Celgene, MundiPharma and Amgen. L. Moreno is also a member of data monitoring committees for clinical trials sponsored by Novartis, Actuate Therapeutics, Shionogi, Incyte, the University of Southampton, and the Royal Marsden NHS Foundation Trust and a member of the Executive Committee of SIOPEN (European neuroblastoma research cooperative group), which receives royalties for the sales of dinutuximab beta. L. Moreno's institution receives funding from sponsors for DMC participation, advisory role, or conducting industry-sponsored clinical trials. Y. Yuan reports personal fees from Merck Sharp & Dohme Limited (consultancy) outside the submitted work. No potential conflicts of interest were disclosed by the other authors.

K. Zormpas-Petridis: Conceptualization, data curation, software, formal analysis, validation, investigation, visualization, methodology, writing-original draft, writing-review and editing. E. Poon: Resources, investigation, writing-review and editing. M. Clarke: Data curation, validation, writing-review and editing. N.P. Jerome: Investigation, writing-review and editing. J.K.R. Boult: Investigation, writing-review and editing. M.D. Blackledge: Supervision, writing-review and editing. F. Carceller: Supervision, writing-review and editing. A. Koers: Investigation, project administration, writing-review and editing. G. Barone: Data curation, supervision, writing-review and editing. A.D.J. Pearson: Supervision, writing-review and editing. L. Moreno: Supervision, writing-review and editing. J. Anderson: Data curation, supervision, writing-review and editing. N. Sebire: Data curation, supervision, validation, writing-review and editing. K. McHugh: Resources, data curation, supervision, writing-review and editing. D.-M. Koh: Supervision, validation, writing-review and editing. L. Chesler: Resources, funding acquisition, writing-review and editing. Y. Yuan: Conceptualization, resources, software, supervision, funding acquisition, validation, visualization, methodology, project administration, writing-review and editing. S.P. Robinson: Conceptualization, resources, supervision, funding acquisition, validation, writing-review and editing. Y. Jamin: Conceptualization, resources, data curation, formal analysis, supervision, funding acquisition, validation, investigation, visualization, methodology, writing-original draft, project administration, writing-review and editing.

Y. Jamin received a Children with Cancer UK Research Fellowship (2014/176). Y. Jamin and S.P. Robinson received Rosetrees Trust grant M593. S.P. Robinson received Cancer Research UK grant C16412/A27725. E. Poon and L. Chesler received Children with Cancer UK Project Grant (2014/174). F. Carceller is partly supported by George and Giant Pledge via the Royal Marsden Cancer Charity. L. Chesler received Cancer Research UK Program Grant (C34648/A18339 and C34648/A14610). 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; grant nos.: C1060/A10334 and C1060/A16464), 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.