Validation of MRI quantitative susceptibility mapping of superparamagnetic iron oxide nanoparticles for hyperthermia applications in live subjects

The use of magnetic fluid hyperthermia (MFH) for cancer therapy has shown promise but lacks suitable methods for quantifying exogenous irons such as superparamagnetic iron oxide (SPIO) nanoparticles as a source of heat generation under an alternating magnetic field (AMF). Application of quantitative susceptibility mapping (QSM) technique to prediction of SPIO in preclinical models has been challenging due to a large variation of susceptibility values, chemical shift from tissue fat, and noisier data arising from the higher resolution required to visualize the anatomy of small animals. In this study, we developed a robust QSM for the SPIO ferumoxytol in live mice to examine its potential application in MFH for cancer therapy. We demonstrated that QSM was able to simultaneously detect high level ferumoxytol accumulation in the liver and low level localization near the periphery of tumors. Detection of ferumoxytol distribution in the body by QSM, however, required imaging prior to and post ferumoxytol injection to discriminate exogenous iron susceptibility from other endogenous sources. Intratumoral injection of ferumoxytol combined with AMF produced a ferumoxytol-dose dependent tumor killing. Histology of tumor sections corroborated QSM visualization of ferumoxytol distribution near the tumor periphery, and confirmed the spatial correlation of cell death with ferumoxytol distribution. Due to the dissipation of SPIOs from the injection site, quantitative mapping of SPIO distribution will aid in estimating a change in temperature in tissues, thereby maximizing MFH effects on tumors and minimizing side-effects by avoiding unwanted tissue heating.


Results
We first examined the accuracy of QSM for estimating ferumoxytol nanoparticle concentrations in vitro. In imaging phantoms containing 89 Zr-ferumoxytol, MR QSM and PET were sensitive enough to detect as low as 1.8 μg Fe/mL, and their measurement increased linearly with ferumoxytol concentration (Fig. 1). QSM linearly responded to ferumoxytol at higher concentrations where the quality of the magnitude image began to decline (Fig. 1a). The conversion factors to relate QSM (ppm) and PET (MBq/cc) units to the concentration of ferumoxytol were derived from linear regression of phantom imaging data (Fig. 1b). The empirical conversion factor to relate QSM to ferumoxytol concentration was determined to be 11.6 ppm×L/g. This conversion factor closely agreed with a theoretical mass susceptibility value of 12.7ppm×L/g at 7 T MRI, which can be derived from the equation relating the mass susceptibility (χ Fe ) to mass magnetization (M Fe ) 21 where μ 0 is the permeability of free space and B 0 is the applied field. In comparison with the straightforward process of determining ferumoxytol by QSM in the imaging phantom, in vivo estimation of ferumoxytol or exogenous iron by QSM is much more complex due to tissue heterogeneity and uneven distribution of susceptibility sources. To overcome these challenges, we applied the graph-cuts based simultaneous phase unwrapping and chemical shift removal method (SPURS) 33 to correct phase and chemical shift discontinuities in the MRI gradient-echo data (Fig. 2a). QSM was then reconstructed from the unwrapped field map with the preconditioned total field inversion algorithm 34 . The steps for deriving QSM in this study differed from our prior studies 19 because of the need to correct for a significant chemical shift from the high fat present in mouse torso and the need to estimate ferumoxytol in a range of concentrations suitable for MFH applications. With this advanced QSM technique, the area injected with ferumoxytol in live mice was clearly discernible from the neighboring tissues (a yellow circled region in Fig. 2a). However, the high susceptibility region was not entirely confined to the injection site; one should therefore rely on the location of SPIO injection site or perform MR scans prior to and post SPIO injection to differentiate the specific susceptibility of SPIO from non-specific susceptibility noise. The difficulty with QSM for quantifying SPIO in vivo is also apparent in the coronal view (Fig. 2b). With algorithms for reducing most of the streaking artifacts, intravenously injected ferumoxytol distribution in the liver could be clearly distinguished from other susceptibility sources. In the same mouse that was implanted with tumor by subcutaneous injection of tumor cells into the upper flank, ferumoxytol accumulation in the tumor was also discernible, displaying higher susceptibility at the periphery of the tumor (yellow circled region in Fig. 2b). The distribution of nanoparticles such as ferumoxytol at the boundary between the tumor and neighboring tissues seemed to be consistent with prior observations, which was ascribed to a phenomenon referred to as the enhanced permeability and retention (EPR) effect [35][36][37] .
With the QSM algorithm fully optimized for in vivo detection of SPIO, we then examined the accuracy of QSM using 89 Zr-ferumoxytol by comparing QSM with PET estimates (Fig. 3a). Mice were scanned pre-and post-ferumoxytol injection to discern susceptibility changes due to SPIO injection. To minimize the dissipation of ferumoxytol from the injection site to outside of tumors and to test the accuracy of QSM at a lower range, we chose the injection volume of ferumoxytol to be 1% of the tumor volumes. The injection sites at tumors near the flank were clearly visible with QSM intensities. We also noted close resemblance of the peak intensities delineated by QSM and PET in both sides of tumors. After PET/CT and MRI scans were completed, mice were sacrificed and subcutaneous tumors excised for gamma counter measurement to validate imaging-based estimation of ferumoxytol in live subjects. The amount of ferumoxytol predicted by PET and QSM corresponded linearly to the estimate by gamma counter except for PET and QSM' underestimation of ferumoxytol (slope 0.77 and 0.62 for PET and QSM versus gamma counter, respectively, shown in Fig. 3c and Table 1). The percent of ferumoxytol retained within the tumors, determined by the difference between the actual injection amount and gamma counter measurement, was essentially 100% for 2 out of 4 tumors, while it was only about 50% for the other 2 tumors. The loss of retention was likely due to dissipation of ferumoxytol to outside of tumors or inaccuracies with delivering microliter volumes into tumors.
In comparison with subcutaneous injection, intravenous injection of ferumoxytol into mice with subcutaneous tumors resulted in uptake mainly by the liver (Fig. 3b). At 24 h post tail-vein injection of ferumoxytol, less than 0.1% of injected dose was found in tumor sites, while 11% of injected dose was accumulated in the liver ( Table 1). The uptake of ferumoxytol by the liver (~185 μg/ml) in mice was comparable to the levels reported for human subjects (~140 μg/ml) 38 . In comparison, the amount of ferumoxytol in our subcutaneous PC3 tumors (~10 μg/ml) was significantly lower than the levels in tumor lesion in the same study (median value ~ 34.5 μg/ ml) 38 . Dominant liver uptake of ferumoxytol is consistent with the clinical use of ferumoxytol (that is digested by macrophages in the liver to release iron ions to the blood) to treat iron deficiency anemia in chronic kidney disease patients 39 . Likewise, after imaging of live subjects by PET/CT and MRI for QSM, the liver was harvested and processed for gamma counter. The QSM and PET estimates of ferumoxytol accumulation in the liver were found to be in excellent agreement (<6% deviation) with the gamma counter estimate (Table 1).
We then examined the therapeutic efficacy of ferumoxytol at two different doses and AMF-induced hyperthermia on suppression of subcutaneous tumor growth (Fig. 4a). Immuno-deficient NSG mice were utilized as a host to accommodate the growth of human cell line, and to avoid possible hyperthermia-induced activation of immune cells and their effects on tumor killing. When tumors of PC3 prostate cell line reached approximately 100 mm 3 in size, mice were grouped into no treatment control or three treatment cohorts that were subjected to intratumoral injection of ferumoxytol, AMF, or both (Fig. 4b). To examine the effect of hyperthermia induced cell killing in its relation to ferumoxytol distribution, we chose to use a small dose of ferumoxytol (a single injection of 25 μl at 5 mg/ml, i.e., 125 μg) to avoid bulk heating-induced bystander killing. The combination of ferumoxytol and AMF led to measurable inhibition of tumor growth compared to no treatment or either the ferumoxytol-or The first echo of the magnitude image, the local field, and QSM images are shown. Note that a rectangular mask was applied to the field to eliminate the extremities of the falcon tube prior to QSM reconstruction. The tube was placed parallel to the main magnetic field in a 7T MR scanner. PET images of the same falcon tube were acquired immediately after the MRI scan. (b) Linear regression of 89 Zr-ferumoxytol phantoms measured by PET and QSM versus the known concentration of ferumoxytol. The slope of the liner curve was used to convert PET and QSM signals to ferumoxytol concentration in live mice and in harvested tissues.
www.nature.com/scientificreports www.nature.com/scientificreports/ AMF-only treatment cohorts. However, the single injection, low-dose ferumoxytol plus AMF combination was only marginally significant at reducing tumor size (p = 0.06 for +/+ vs. −/−) (Fig. 4b). At 24 h after the last treatment, mice were sacrificed and tumors were taken out for ex vivo fluorescence imaging of GFP intensity to assess tumor viability (Fig. 4c,e). Following fluorescence imaging, excised tumors were subsequently scanned by MRI for QSM to relate tumor killing to ferumoxytol concentration (Fig. 4d). The amount of ferumoxytol retained within the tumor was lower overall than the amount injected (125 μg), likely caused by dissipation of ferumoxytol during the four days after intratumoral injection, ferumoxytol degradation inside cells, and/or the loss of SPIO www.nature.com/scientificreports www.nature.com/scientificreports/ properties due to exposure to AMF. However, judging from a trend of lower tumor volume with increasing ferumoxytol concentrations within tumors (Fig. 4f), hyperthermia effects was indeed responsible for tumor killing even at this low dose of ferumoxytol injections. To further examine if tumor killing correlates with ferumoxytol dose, a repeated, high concentration of ferumoxytol (four injections of 25 μl at 30 mg/ml, i.e., 750 μg per injection) was used. As anticipated, the combination of high dose ferumoxytol and AMF led to significant reduction in tumor size (Fig. 4e).
The histology of subcutaneous tumors receiving intratumoral injections of ferumoxytol with or without AMF further confirmed the specific killing induced by MFH (Fig. 5). First, we noted that even with intratumoral injection, ferumoxytol (delineated by Prussian blue stain) appeared to distribute to the periphery of tumors rather than permeating evenly throughout tumor tissues. Due to a small volume of intratumoral injection of ferumoxytol, we noted that hyperthermia induced tumor killing (identified by TUNEL stain) appeared to be restricted to the region of ferumoxytol localization, validating a lack of temperature elevation in the bulk of the tumor. Cell death in the tumor stroma (a region marked with yellow asterisk in Fig. 5a and blue dotted circle in Fig. 5b) was also www.nature.com/scientificreports www.nature.com/scientificreports/ confirmed, identified by the absence of colocalized GFP staining with ferumoxytol distribution. Co-localization of ferumoxytol distribution and cell death by TUNEL was less apparent in tumors without AMF (a region marked with red asterisk; Fig. 5a).

Discussion
Magnetic susceptibility, a fundamental physical property of a SPIO contrast agent such as ferumoxytol is proportional to the concentration of contrast agent and independent of its surrounding medium. The molar susceptibility of the contrast agent, which can be predetermined using phantom calibration in vitro, can be used to convert a susceptibility map into a concentration distribution of SPIO contrast agent in vivo. Quantitative imaging of SPIO distribution in vivo is not only useful as a blood pool contrast agent but can potentially aid magnetic hyperthermia in cancer by providing a better prediction of temperature rise to ensure therapeutic activity in tumor tissues while minimizing unwanted side effects 40 . Previously, we reported the application of QSM to estimate SPIO distribution in animals after fixation, and validated the accuracy of QSM on harvested tissues 28,29 . In this study, we applied QSM technique to estimate and validate distribution of clinically approved SPIO nanoparticles in live subjects, which will be necessary for predicting temperature elevation in tissues in response to AMF in vivo.
QSM prediction of SPIO directly injected into solid tumors near the skin faces technical challenges due to the need to correct for a chemical shift arising from the high fat content near the skin, the proximity of tumor growth to the skin in contact with air, and a large variation of SPIO concentrations near the injection site. Tissue susceptibility due to high fat in the skin was eliminated by the SPURS method in this study 33 , which was specifically designed to correct for chemical shift contributions to the phase in the gradient-echo MRI data. SPURS has been shown to produce more accurate QSM maps as it avoids the need for additional field map smoothing 33 . Due to the proximity of tumor growth near the skin and air interface, we used QSM reconstruction that avoids the removal of the background field as this technique assumes that the susceptibility source is far from the tissue edge and its field is approximately orthogonal to the background field 18,34 . Another difficulty encountered in this study was the degree of susceptibility source variation that ranged from ~0.1 ppm for soft tissue to high ppm values (e.g., 348 ppm for 30 mg/ml SPIO) for intratumoral injections of SPIO. We used the recently reported preconditioned total field inversion QSM technique 34 to overcome these problems. The resulting QSM algorithms developed in this study enabled the mapping of SPIO distribution in the entire body of live subjects in vivo. From the QSM map, intravenously injected ferumoxytol in mice with subcutaneous tumor xenografts was found mainly in the liver. Ferumoxytol accumulation in the tumor was clearly discernable, but predominated at the tumor-stroma boundary. This finding is consistent with many prior studies demonstrating accumulation of nanoparticles and antibodies around the tumor periphery 41,42 . Our QSM therefore can be used for evaluating biodistribution and pharmacokinetics of SPIO nanoparticles in real-time and for exploring different designs of nanoparticles with improved tumor targeting.
SPIO nanoparticles are being investigated as therapeutic agents for hyperthermia-induced tissue heating under AMF 1,3,40 . Inhomogeneous distribution of SPIO after intratumoral injection has previously been attributed to uneven tumor growth inhibition 3 . Therefore, in vivo quantification of SPIO such as ferumoxytol by a clinical imaging modality can provide information for predicting temperature elevation under AMF in live subjects, and therefore insight into possible cellular and tissue damage. To correlate SPIO concentration with tumor killing by hyperthermia, two different doses of ferumoxytol were used for intratumoral injection. For a low-dose study, ferumoxytol was injected 24 h prior to AMF application, which was chosen to promote cellular internalization of ferumoxytol. For the high-dose, repeated ferumoxytol study, the first ferumoxytol injection was made 24 h prior to AMF, followed by three cycles of ferumoxytol injection and immediate application of AMF. The dependence of cell killing on ferumoxytol concentration and AMF application was apparent, judging from the trend of tumor size and ferumoxytol amounts within tumor, and more significant tumor killing effects seen in the high-dose www.nature.com/scientificreports www.nature.com/scientificreports/ cohort. Compared to more pronounced suppression of tumor growth in prior studies 1,3,43 , MFH effects seen in this study were more modest. This was likely due to a relatively low specific absorption rate of ferumoxytol compared to other SPIOs 31 , a delivery of small injection volume into the tumor, and the use of immune-compromised host that lacks in a hyperthermia-induced immune or macrophage-mediated killing of tumors 44,45 . It is also likely that further optimization including ferumoxytol dose, route of injection, timing of injection and AMF application, and AMF parameters will significantly influence the efficiency of tumor killing. Hyperthermia-induced tumor killing will be critically influenced by tumor and stroma interactions and the degree of tumor blood vessel formation, which will affect EPR effect for SPIO uptake by tumor both for intratumoral and intravenous routes of injection. The goal of our current study, however, was not achieving maximum MFH-induced tumor killing, but rather developing and demonstrating a clinical imaging tool to quantitatively assess SPIO distribution in live subjects, and correlating its distribution with cell killing directly caused by MFH.
Compared with PET imaging of radionuclides bound to iron-oxide nanoparticles for iron quantification, our QSM approach is advantageous in that it avoids the use of ionizing radiation and provides superior soft tissue contrast and spatial resolution. QSM can serve multiple purposes to improve MFH by mapping SPIO distribution prior to AMF. In particular, it can be used to optimize intratumoral delivery of SPIOs to ensure SPIO dispersion www.nature.com/scientificreports www.nature.com/scientificreports/ throughout the tumor while minimizing distribution into normal tissues for more accurately targeted hyperthermia. More broadly, the QSM algorithms developed in this study can facilitate preclinical and clinical studies to explore and optimize SPIO-based hyperthermia therapy in terms of material design, SPIO and AMF dose and frequency, as well as the route of SPIO delivery.

Methods
Preparation of dual-modality contrast agent 89 Zr-ferumoxytol. 89 Zr-conjugated ferumoxytol (AMAG Pharmaceuticals) was produced using 1-Ethyl-3-(3-dimethylaminopropyl) carbodiimide (Sigma) and DFO-p-SCN (Macrocyclics) following a procedure described previously 9 . For validation of QSM by PET, imaging phantom containing 89 Zr-ferumoxytol ranging from 1.8 to 117.0 μg/ml by mass of iron (Fe) was prepared by serial dilution of 89 Zr-ferumoxytol suspended in 1% agarose in PBS solution successively layered on beds of 1% agarose gel in a falcon tube. In vivo validation of QSM by PET and gamma counter. For validation of MR QSM in live subjects, 89 Zr-ferumoxytol was injected directly into tumors when tumor size reached approximately 100 to 250 mm 3 . The injection volume of 89 Zr-ferumoxytol (20 mg/ml) was set to 1% of the tumor volumes, which was chosen to test the lower limit of the imaging sensitivity and to minimize the loss of ferumoxytol due to dissipation from tumors. Live mice were scanned by PET/CT and MRI prior to and immediately after injection of ferumoxytol. After imaging, mice were euthanized, and subcutaneous tumors were harvested and processed for the measurement Image reconstruction and post-processing. QSM maps were computed offline from multi-echo GRE MRI data using custom MATLAB software (version R2012b, The Mathworks, USA). The susceptibility induced field inhomogeneity map, f, was estimated by a nonlinear voxel-wise fit to the phase of the complex data, unwrapped and corrected for chemical shift using the graph-cut based SPURS algorithm 25,33,48 . A binary mask, M, isolating the region of the anatomy with the tumor, was created by applying a region-growing algorithm to the MRI magnitude image in the ITK-Snap image processing software 49  where χ is the total susceptibility, * is its convolution with the dipole kernel d, f is the total field, w is a noise weighting, ∇ is the gradient operator, and M G is the binary edge weight for suppressing streaking artifacts. The values in the preconditioner matrix, P, which accounts for strong susceptibility differences between the high SPIO dose and surrounding tissue in the tumor, were set to 30 for air (region outside the mask) and to the signal to noise ratio matrix for tissue (region inside the mask) estimated from the phase fitting step 34 . The optimal regularization parameter, λ, was chosen empirically and set to 1000. MRI and PET/CT images were manually aligned using the AMIDE multimodality image analysis software (AMIDE software; http://amide.sourceforge. net/index.html). Volumes of interest (VOIs) with dimensions equal to the tumor volume were drawn on the PET/ CT and MRI images to cover the visible tumor volume. The mean voxel values and volumes from the VOIs were recorded for each image type. Using a fat mask created by binarization of the fat map from the fat/water separation procedure, the average QSM of fat in the mouse was estimated, used as reference for VOI estimates. All mean voxel values were converted into total iron mass using the coefficients determined from the phantom calibration experiments.
Gamma count. Organs were harvested from euthanized specimens and scanned individually using Wizard 2 Gamma Counter (Perkin Elmer). Harvested tumors and liver were weighed, and a fraction of these tissues (1/4-1/2 of the total volume) was used for gamma counter. Organs were not washed in order to maximize the retention of 89 Zr-ferumoxytol. A standard of 89 Zr-ferumoxytol was prepared to calculate absolute iron uptake in each organ.
Histology. Mice were anesthetized with an intraperitoneal injection of a ketamine/xylazine combination (150 mg/kg; 15 mg/kg) and underwent cardiac perfusion with PBS and subsequently 4% paraformaldehyde by a trained animal user according to IACUC guidelines. Subcutaneous tumors were fixed for 24 hours and embedded in paraffin and cut in 7 µm sections (Microtome, Leica). Sections were stained for iron accumulation by Prussian Blue staining 29 , to detect DNA breaks by TUNEL (terminal deoxynucleotidyl transferase dUTP nick end labeling) stain, and to localize tumor growth by GFP immunostaining (performed by Histowiz, Inc.).

Data availability
All data generated or analyzed during this study are included in the published article.