Imaging iron-loaded mouse glioma tumors with bSSFP at 3 T

Authors

  • Lisa M. Bernas,

    1. Imaging Research Laboratories, Robarts Research Institute, University of Western Ontario, London, Ontario, Canada
    2. Department of Medical Biophysics, University of Western Ontario, London, Ontario, Canada
    Search for more papers by this author
  • Paula J. Foster,

    Corresponding author
    1. Imaging Research Laboratories, Robarts Research Institute, University of Western Ontario, London, Ontario, Canada
    2. Department of Medical Biophysics, University of Western Ontario, London, Ontario, Canada
    3. Department of Diagnostic Radiology and Nuclear Medicine, University of Western Ontario, London, Ontario, Canada
    • Imaging Research Laboratories, Robarts Research Institute, P.O. Box 5015, 100 Perth Drive, London, Ontario, Canada, N6A 5K8
    Search for more papers by this author
  • Brian K. Rutt

    1. Imaging Research Laboratories, Robarts Research Institute, University of Western Ontario, London, Ontario, Canada
    2. Department of Medical Biophysics, University of Western Ontario, London, Ontario, Canada
    Search for more papers by this author

Abstract

The poor prognosis for patients with high-grade glioma is partly due to the invasion of tumor cells into surrounding brain tissue. The goal of the present work was to develop a mouse model of glioma that included the potential to track cell invasion using MRI by labeling GL261 cells with iron oxide contrast agents prior to intracranial injection. Two types of agents were compared with several labeling schemes to balance between labeling with sufficient iron to curb the dilution effect of cell division while avoiding overwhelming signal loss that could prevent adequate visualization of tumor boundaries. The balanced steady-state free precession (bSSFP) pulse sequence was evaluated for its suitability for imaging glioma tumors and compared to T2-weighted two-dimensional fast spin echo (FSE) and T1-weighted spoiled gradient recalled echo (SPGR) at 3 T in terms of signal-to-noise ratio and contrast-to-noise ratio efficiencies. Ultimately, a three-dimensional bSSFP protocol consisting of a set of two images with complementary contrasts was developed, allowing excellent tumor visualization with minimal iron contrast when using pulse repetition time = 6 ms and α = 40°, and extremely high sensitivity to iron when using pulse repetition time = 22 ms and α = 20°. Quantitative histologic analysis validated that the strong signal loss seen in balanced steady state free precession pulse sequence images of iron-loaded tumors correlated well with the presence of iron. Magn Reson Med, 2010. © 2010 Wiley-Liss, Inc.

With the escalating development of normal and transgenic mouse models of disease comes the need for efficient, noninvasive methods for their characterization and study (1, 2). MRI is an excellent tool for longitudinal small animal studies as it is noninvasive and involves no ionizing radiation, allowing for long-term repeated imaging, and can achieve the high resolution necessary for small animal imaging, with recent advances using high-field-strength scanners and specialized hardware (3–5).

The active invasion of tumor cells into the surrounding brain tissue makes complete surgical resection of glioma tumors impossible and limits the success of radiation therapy, thereby contributing to the extremely poor prognosis faced by glioma patients (6). The application of cellular MRI techniques, using superparamagnetic iron oxide contrast agents, to the study of glioma has the potential to allow a dynamic assessment of glioma cell invasion and contribute to the development and evaluation of novel therapies targeting invading cells (7).

As small animal imaging demands high resolution and high signal-to-noise ratios (SNR), clinical protocols are not ideal. T1-weighted images can be acquired in reasonable scan times; however, they show low tumor-brain contrast without gadolinium. T2-weighted images show excellent tumor-brain contrast but suffer from artifact in three-dimensional (3D) images due to the long pulse repetition times (TR) required and are acquired relatively inefficiently. Recently developed sequences that extend the echo train acquisition of 3D fast spin echo (FSE) have the potential to improve on this but have yet to be used for small animal or brain tumor imaging (8).

The balanced steady-state free precession (bSSFP) pulse sequence is very SNR efficient and produces unique T2/T1 contrast (9). This sequence has been shown recently to produce images of a mouse glioma model, with contrast similar to T2-weighted RARE (rapid acquisition with relaxation enhancement) images (10). A challenge presented by bSSFP, however, is its high sensitivity to local field inhomogeneities. The result is a characteristic “banding artifact”, which worsens at higher field strengths and with longer TR (9). Multiple acquisition radiofrequency (RF) phase cycling techniques have been developed that ameliorate this problem and have allowed for bSSFP imaging at higher field strengths and with longer TR (10, 11).

Although the sensitivity of bSSFP to local field inhomogeneities can be problematic, it has also been what has enabled this sequence to be used for highly sensitive cellular imaging, which has allowed the detection of single cells both in vitro and in vivo at 1.5 T and 3 T (12–14). Imaging iron-loaded tumors presents unique challenges since the large blooming artifacts in SSFP images due to iron oxide may prevent visualization of the tumor boundaries. Additionally, since a tumor consists of a large mass of rapidly dividing cells, there is the need to balance between loading cells with sufficient iron to curb the effect of dilution due to cell division while avoiding overwhelmingly large areas of signal loss.

The purpose of the current work was to develop a protocol for imaging iron-loaded glioma tumors in mice that would allow good tumor visualization, as well as highly sensitive detection of iron-loaded cells. The approach taken was to develop a suitable iron-loaded tumor model and to optimize an MRI protocol for imaging these tumors. For optimal tumor visualization, maximum tumor-brain and white matter–gray matter contrast with minimal iron-induced contrast was desired, whereas for maximum sensitivity to iron-loaded cells, the opposite criteria were sought. The scanning parameters for the bSSFP pulse sequence were optimized to produce a set of two images with complementary contrasts, allowing for efficient high-resolution scanning of iron-loaded mouse glioma tumors.

MATERIALS AND METHODS

Cell Culture

GL261 mouse glioma cells (National Cancer Institute Division of Cancer Treatment and Diagnosis (DCTD) Tumor Repository, Frederick, MD) were maintained at 37°C and 5% CO2 in RPMI-1640 medium (Gibco, Invitrogen Corp., Carlsbad, CA) supplemented with 10% fetal bovine serum (Gibco, Invitrogen Corp., Carlsbad, CA) and passaged every 2-3 days.

Cell Labeling With Superparamagnetic Iron Oxide Nanoparticles and Micron-Sized, Inert Iron Oxide Beads

For iron labeling, cells were plated in 6-well plates at 5 × 105 cells per well and allowed to adhere overnight. The culture medium was then aspirated and replaced with fresh medium to which iron oxide agents were added. Two types of iron oxide agents were used: small (80-50 nm), dextran-coated superparamagnetic iron oxide nanoparticles (SPIO) and micron-sized, inert iron oxide beads (MPIO). In the case of SPIO labeling, Feridex IV® (AMAG Pharmaceuticals, Inc., Cambridge, MA) was first mixed with Lipofectamine™ 2000 (Invitrogen Corp., Carlsbad, CA) in serum-free medium for 1 h with intermittent vortexing to allow complexes to form, and then 0.5 mL was added to the cells with an equal volume of complete medium to produce the desired final iron concentration. The Fe:Lipofectamine concentrations used were 100:5 μg/mL for low loading (∼5 pg Fe/cell) and 200:5 μg/mL for high loading (∼15 pg Fe/cell). For labeling with MPIO, 1-μm SiMag beads (Chemicell, Berlin, Germany) were added directly to the plates in 1 mL complete medium. The iron concentrations used in this case were 22 μg Fe/mL and 66 μg Fe/mL for low (∼5 pg Fe/cell) and high (∼15 pg Fe/cell) loading levels, respectively. Cells were incubated at 37°C and 5% CO2 overnight, and then washed four times in the plates with Hanks' buffered salt solution (Gibco, Invitrogen Corp., Carlsbad, CA), detached using 0.25% trypsin-EDTA (ethylenediaminetetraacetic acid) (Gibco, Invitrogen Corp., Carlsbad, CA), and washed twice more with Hanks' buffered salt solution before being resuspended in the solution for injections.

Appropriate iron incubation concentrations were determined from prior uptake experiments, in which iron uptake was quantified using a colorimetric assay (15). Briefly, following thorough washing, cells suspended in Hanks' buffered salt solution at ∼2 × 106 cells/mL were plated in triplicate at 100 μL per well in a 96-well plate and dried uncovered overnight at 80°C. Calibration samples were also included each time; 100 μL of 5M hydrochloric acid was added to each well and the plate was sealed to prevent evaporation and incubated for 5 h at 60°C; 100 μL of freshly prepared 5% potassium ferrocyanide was then added to each well and the plate was incubated in the dark at room temperature for 15 min. Absorbance at 650 nm was then read on a plate reader and the amount of iron per cell was calculated using calibration data.

Animal Model

Male C57/Bl6 mice were anesthetized with isoflurane (2% in oxygen for induction; 1% in oxygen for maintenance). The hair was removed from the scalp using Nair® (Church & Dwight Canada Corp., Mississauga, ON, Canada) and the skin was disinfected using Betadine scrub (Purdue Products L.P., Stamford, CT) followed by alcohol and then povidone iodine. A midline incision was made and the animal was positioned in a stereotactic frame (Stoelting Co., Wood Dale, IL, USA) equipped with a mouse adaptor. After leveling the skull, a 2mm burr hole was drilled at 2.0mm lateral and 0.5mm anterior to bregma. The needle of a 10-μL Hamilton syringe containing the cell suspension was advanced to 4.2mm below the skull surface and allowed to rest for 2-3 min. The needle was then pulled back 0.7 to 3.5 mm below the skull surface and 5 × 104 GL261 cells in 1-2 μL Hanks' buffered salt solution were injected over 10 min into the caudate putamen. The needle was then left in place for an additional 5 min and then retracted over 2-3 min. The burr hole was plugged with gel foam and sealed with bone wax, and the incision was sutured. The animals were observed until alert and active and then returned to their cages. All animal experiments were approved by the University of Western Ontario Animal Use Subcommittee.

The injected cell population consisted of unlabeled cells only, low-level iron-loaded cells only, or a combination of iron-loaded and unlabeled cells (25% iron-loaded cells at either the low or high loading level, or 5% iron-loaded cells at the high loading level).

MRI

MRI was conducted on a 3 T scanner (GE Healthcare, Waukesha, WI), with the addition of an in-house-designed and -built high-performance insertable gradient coil (maximum gradient strength 500 mT/m and slew rate 3200 T/m/s) using a custom-built solenoidal RF coil (1-cm diameter × 1-cm length). For in vivo imaging, the mice were anesthetized using isoflurane, and body temperature was maintained by circulating warm water. Ex vivo scanning was performed on fixed heads bathed in Fluorinert™ (3M, St. Paul, MN) with the skull intact.

The suitability of the bSSFP pulse sequence for imaging mouse glioma tumors was evaluated in vivo using an unlabeled tumor, 14 days post–cell injection. 3D bSSFP images were obtained using two flip angles (20° and 30°) and three TR/bandwidth combinations (5.5 ms/31 kHz, 7 ms/21 kHz, and 12 ms/10 kHz, with scan times of 9, 12, and 9 min, respectively). The echo times were fixed at the symmetrical position of TR/2, and the spatial resolution was 100 × 100 × 200 μm. The resulting images were compared in terms of SNR efficiency and tumor-brain contrast-to-noise ratio (CNR) efficiency, defined as SNR and CNR normalized by the voxel size and square root of scan time, respectively. The best protocol from this comparison was then evaluated against sequences more commonly used for clinical glioma imaging, namely, T2-weighted two-dimensional FSE (echo time/TR = 80 ms/3000 ms, 100 × 100 × 400 μm voxels, echo train length = 15, bandwidth = 15 kHz, scan time 6 min) and T1-weighted 3D spoiled gradient recalled echo (SPGR) (echo time/TR = 1.5 ms/6.3 ms, flip angle = 20°, bandwidth = 15 kHz, 100 × 100 × 200 μm voxels, scan time 12 min), before and up to 30 min post–gadolinium injection (0.1 mmol/kg in 0.2 mL administered intraperitoneally). Again, the images were compared for SNR and CNR efficiency, as well as qualitatively for sensitivity to artifact.

The ability to separately control the bandwidth and TR, as well as the incorporation of a multiple acquisition RF phase cycling technique, which combined multiple images acquired with incremental RF phases by sum of squares (11), allowed the bSSFP protocol to be further optimized for even better tumor delineation and localization, as well as higher sensitivity to iron-loaded cells. The scanning parameters for the two images acquired with this protocol were (1) for tumor visualization: 100-μm isotropic resolution, 40° flip angle, echo time/TR = 3 ms/6 ms, 21 kHz bandwidth, and 16 phase cycles for a scan time of 30 min, and (2) for sensitive iron-loaded cell detection: 66 μm × 66 μm × 100 μm resolution, 20° flip angle, echo time/TR = 11 ms/22 ms, 15 kHz bandwidth, and 10 phase cycles for a scan time of 127 min. The two flip angles were chosen based on calculations of white matter–gray matter contrast at 3 T using T1 and T2 value estimates to calculate bSSFP signal intensity over a range of flip angles (16). With a flip angle of 20°, white matter–gray matter contrast is minimal, whereas with a flip angle of 40°, maximal white matter–gray matter contrast is obtained. This further optimization was conducted ex vivo using mouse heads, 18 days post–tumor cell injection and included tumors containing high- or low-level iron-loaded cells in the proportions mentioned above. The acquisition times were chosen to obtain an SNR of 50 for both scans.

Histologic Analysis

Mice were euthanized using Euthanyl (Bimeda-MTC Animal Health Inc., Cambridge, ON, Canada) and perfused with cold saline followed by cold 3.75% formalin. Following ex vivo MRI, the brains were removed and cryoprotected by passage through a sucrose gradient (10%, 20%, and then 30% for 24 h each) and then embedded in gum-sucrose in an orientation parallel to that of the MRI and frozen to −80°C. Cryostat sections (20 μm) were then collected and consecutive slides were stained for hematoxylin and eosin and Perls Prussian Blue (PPB) for iron with eosin counterstain.

Microscopy images were acquired of adjacent hematoxylin and eosin– and PPB-stained sections from throughout the tumor volumes at 50× magnification using a Zeiss Axio Imager (Germany) equipped with a Retiga EXi Digital CCD camera (Q Imaging, Vancouver, BC, Canada). The images were then aligned and stitched together using the Hugin photograph stitching software to allow visualization of entire intact sections. Using the “lasso” and “magic wand” tools in Adobe Photoshop (Adobe Systems Inc., San Jose, CA), the tumors were cropped from the PPB-stained sections and segmented into the following classes, which were identified based on color and morphology: iron, blood vessels, hemorrhage, microcyst/necrotic regions, and the remaining tumor parenchyma. Using ImageJ (NIH, Bethesda, MD), the fraction of pixels corresponding to each class was determined by using the “Multithresholder” plug-in to assign a value of zero to all pixels within a 10-grayscale-value window (to account for the slight gradient in pixel values surrounding the segmented areas) around the pixel value for each class and a value of 255 to the rest of the image. The histogram tool was then used to measure the number of zero value pixels. This was repeated for each class in turn, as well as for the entire tumor area, to allow the fractional composition of the tumor area to be calculated.

Analysis of MRI Data

Using OsiriX imaging software, the tumor areas were outlined on the images obtained using the 40° flip angle/6 ms TR protocol, since the tumor boundaries are more easily visualized on these images. The resulting regions of interest were then copied onto the corresponding 20° flip angle/22 ms TR protocol images, which were used for comparison to the histology data. DICOM (Digital Imaging and Communications in Medicine) series containing only the tumor regions were exported for analysis in ImageJ.

Histograms of the entire tumor volumes were generated using ImageJ and used to determine an appropriate signal intensity threshold to define strong signal loss in the MRI images. A threshold of 4000 was found to be appropriate for all of the images discussed here.

The MRI image corresponding to each PPB section was determined using the adjacent hematoxylin and eosin–stained sections as a guide. From histograms of individual slices generated using ImageJ, the fraction of pixels within the tumor area with signal intensity below 4000 was determined. The fractional area of strong signal loss was correlated to the area of each segmented class on the PPB sections.

Additionally, to validate that overall correlation between segmented areas on histology and strong signal loss on MRI images corresponded spatially, the segmented histology images were overlaid on the matched MRI images (warping was necessary due to the changes that occur during histologic processing and was performed using the warping tool in Adobe Photoshop) and evaluated qualitatively for spatial agreement.

Statistical Analysis

Statistical analyses were performed using GraphPad Prism software (GraphPad Software, Inc., La Jolla, CA). A total of 51 sections from 21 animals (one to six slices/animal) were used for linear regression analysis of histology segmentation area vs MRI strong signal loss area.

RESULTS

The suitability of the bSSFP sequence for imaging mouse glioma tumors was first evaluated in vivo on a mouse harboring a 14-day-old unlabeled intracranial tumor. The results from this evaluation are shown in Fig. 1. Minimal differences were observed between the two flip angles, with α = 20° producing slightly higher SNR and CNR efficiencies. For both flip angles, the narrower bandwidth and longer TR images had higher SNR efficiency but were more sensitive to motion than the other protocols. The midrange TR/bandwidth protocols produced a good compromise in SNR efficiency and excellent CNR efficiency; thus, the TR/bandwidth/flip angle = 7 ms/21 kHz/20° protocol was deemed the best option.

Figure 1.

Evaluation of bSSFP for the visualization of mouse glioma tumors. Images of an unlabeled mouse glioma tumor 14 days post–cell injection. a: bSSFP images acquired with two flip angles (20 & 30°) and three TR/bandwidth combinations (5.5 ms/31 kHz, 7 ms/21 kHz, and 12 ms/10 kHz). It is apparent that the longer TR/narrower bandwidth protocols are more sensitive to motion than the others. b: Comparison of SNR and CNR efficiencies for the various bSSFP protocols for the tumor shown in (a). In general, the narrower bandwidth/longer TR protocols have higher SNR efficiency but minimal or no advantage in terms of CNR efficiency. The TR/bandwidth/flip angle = 7 ms/21 kHz/20° produces a good compromise between SNR efficiency and motion sensitivity and has excellent tumor-brain CNR efficiency. c: Image showing region of interest locations for tumor signal, brain signal, and background noise measurements.

The comparison of this bSSFP protocol with T2-weighted two-dimensional FSE and pre- and postgadolinium T1-weighted SPGR sequences, again using a mouse harboring an unlabeled tumor, is shown in Fig. 2. T2-weighted FSE resulted in 55% higher tumor-brain CNR efficiency than the bSSFP protocol but was half as efficient in SNR. The T1-weighted SPGR images were very inefficient in both SNR and CNR and offered no benefit over the bSSFP protocol, which produced superior contrast enhancement even 20 min after gadolinium injection, when the SPGR contrast had already begun to decrease. These data show that the 3D bSSFP sequence is able to provide good tumor-brain contrast with higher efficiency than two-dimensional FSE and 3D SPGR sequences.

Figure 2.

Comparison of bSSFP to conventional sequences for imaging mouse glioma tumors. a: Images of a mouse glioma tumor 14 days post–cell injection acquired with bSSFP, T2-weighted FSE, and T1-weighted SPGR pre– and up to 30 min post–gadolinium injection. The tumor can be well visualized in the bSSFP and T2-weighted images. Although gadolinium enhancement is apparent, it does not contribute additional information in this case. b: Comparison of SNR and CNR efficiencies for the various imaging protocols for the tumor shown in (a). The T2-weighted protocol produces better CNR efficiency than the bSSFP protocol before gadolinium injection but has much lower SNR efficiency. The T1-weighted SPGR protocol has low SNR and CNR efficiencies, even with gadolinium enhancement. The bSSFP protocol shows increased CNR efficiency 20 min after gadolinium injection. c: Image showing region of interest locations for tumor signal, brain signal, and background noise measurements.

Examples of the complementary images achieved using the optimized two-acquisition bSSFP protocol are shown in Fig. 3. The extension of TR to 22 ms, combined with higher resolution (66 × 66 × 100 μm) and a flip angle of 20° to minimize white matter–gray matter contrast, resulted in much higher sensitivity to iron-loaded cells, which appear as discrete, punctate areas of signal loss on the MR images. A TR of 6 ms and a flip angle of 40° maximized tumor visualization while minimizing the effect of the iron.

Figure 3.

Two-acquisition bSSFP protocol for tumor visualization and high sensitivity to iron contrast. a,b: Ex vivo unlabeled tumor 18 days post–cell injection. c,d: Ex vivo tumor containing 25% cells labeled with SiMag beads at 5 pg Fe/cell 18 days post–cell injection. The images in (a) and (c) were acquired using TR = 6 ms and flip angle = 40°. The images in (b) and (d) were acquired using TR = 22 ms and flip angle = 20°. The short TR protocol allows excellent visualization of the tumor border and minimizes the effect of the iron. The longer TR protocol maximizes the effect of the iron and produces a flat background, allowing for highly sensitive detection of iron-labeled cells.

An example of the histology segmentation is shown in Fig. 4. This specific example was chosen because it contained all tissue classes segmented; however, it should be noted that hemorrhage was only observed in 20 of the 51 slices, corresponding to 10 of the 21 animals. Hemorrhage was likely due to stroke in most cases as it was mainly observed at the tumor borders, was quite extensive when it did occur, and was associated with poor animal health. Iron staining was observed on 49 of the 51 slices. From this segmentation data, the fractional area occupied by each tissue class was calculated for each histology slice and correlated to the fractional area of strong signal loss on the matching MRI images. The resulting correlations are shown in Fig. 5.

Figure 4.

Example showing histology segmentation. a: Hematoxylin and eosin–stained section of mouse brain 18 days postinjection with 5 × 104 Feridex-labeled (5 pg/cell) GL261 cells. Inset shows area of hemorrhage at 400× magnification where red blood cells are visible. b: Tumor area on adjacent PPB-stained section. c: Segmented tumor from (b) yellow = iron, pink = hemorrhage, gray = microcyst/necrosis, red = blood vessels, black = tumor parenchyma.

Figure 5.

Correlation of segmented histology class area and area of ‘strong signal loss’ on matched histologic and MRI slices. a: Iron staining area. b: Blood vessel area. c: Hemorrhagic region area. d: Microcystic/necrotic area. The iron staining on histology positively correlated (P < 0.001) with the area of strong signal loss on MRI, with an r2 of 0.3688. Cystic/necrotic area negatively correlated (P < 0.05) with the area of strong signal loss on MRI (r2 = 0.1066). Neither hemorrhage nor blood vessel areas correlated significantly with strong signal loss on MRI (r2 = 0.05276 and 0.009108, respectively, P > 0.05). Solid line indicates the line of best fit and the dashed lines represent the 95% confidence intervals.

The area of iron staining showed a strong significant positive correlation with the area of strong signal loss, with r2 = 0.3688 and P < 0.0001. The area occupied by microcyst/necrosis showed a negative correlation with the area of signal loss the MRI images, indicating that these regions are associated with brighter MR images (r2 = 0.1066; P < 0.05). Hemorrhage and blood vessel area did not correlate significantly with signal loss on the MR images (r2 = 0.05276 and 0.009108, respectively; P > 0.05). Areas of hemorrhage may appear dark on MRI images; however, when iron is present, its effect overwhelms the effect of the hemorrhage. The signal loss due to hemorrhage is also more diffuse and variable than that associated with iron.

The spatial correspondence of iron staining on histology and matched MRI slices is demonstrated in Fig. 6. Despite the fact that many histology slices are contained in any given MRI slice, the iron staining, represented by the red dots, corresponds well to the areas of signal loss on the MR images. The majority of red dots are located over areas of signal loss; however, there are some exceptions where red dots lie over areas of higher signal intensity. One reason for this imperfect alignment may be the approximate nature of the superposition technique used. Also, the red dots do not take into account differences in PPB staining intensity, but rather only its location. This may explain why some red dots are located over areas of higher signal intensity, as well as why large areas of signal loss do not necessarily contain a higher density of red dots. Red dots over higher signal intensity regions correspond to weaker PPB staining and those over large areas of signal loss correspond to stronger PPB staining. This figure also demonstrates that the TR = 22 ms/α = 20° protocol is much more sensitive to the presence of iron than the TR = 6 ms/α = 40° protocol, given that many more red dots align with areas of signal loss and that the signal loss areas are much stronger on the image acquired with the former protocol than that acquired with the latter.

Figure 6.

Overlay of iron staining on histology (red dots) and corresponding MR images for three slices from the same animal. Left column: TR = 22 ms/α = 20°; right column: TR = 6 ms/α = 40°. These images demonstrate that the correlation shown in Fig. 4a also corresponds to a spatial correlation between iron staining and areas of strong signal loss on MR images. Also, it is evident from these images that the long TR protocol is much more sensitive to the presence of iron than is the short TR protocol.

Figure 7 shows a comparison between SPIO- and MPIO-labeled tumors in which the iron loading per cell was 5 pg in all cases, but the fraction of labeled cells in the injected population was 25% or 100%. Little difference is evident between the two agents. It appears that, even with this low iron/cell loading, 100% labeling is excessive as it leads to large areas of signal loss that prevent structural details from being visualized. At this iron/cell loading level, 25% labeling results in a manageable amount of signal loss remaining after 18 days, while still showing strong evidence of iron.

Figure 7.

Comparison of SPIO- and MPIO-labeled tumors. SPIO-labeled tumors are shown in (a) and (c), and MPIO-labeled tumors are shown in (b) and (d). Cells were labeled with 5 pg iron per cell in all cases; the injected population consisted of either 25% labeled cells (a,b) or 100% labeled cells (c,d). Even at this low iron/cell level, 100% labeled cells is excessive for both SPIO and MPIO as it leads to large areas of signal loss that prevent structural details from being visualized. At this iron/cell level, 25% labeled cells in the injected population results in a manageable amount of signal loss remaining after 18 days.

Figure 8 illustrates the difference between the low iron loading (5 pg/cell) and high iron loading (15 pg/cell) in the case of MPIO labeling. The fraction of labeled cells injected was 25% in both cases. It appears that 15 pg/cell is excessive when 25% of the cells are labeled with MPIO, whereas 5 pg/cell is reasonable. Such a strong difference between loading levels was not seen in SPIO- labeled tumors. Also, it appeared that labeling only 5% of the injected population, even at a high loading level, was insufficient, especially in the case of SPIO.

Figure 8.

Comparison of low (5 pg/cell) and high (15 pg/cell) cell loading levels for two MPIO-labeled tumors. The fraction of labeled cells in the injected population was 25% in both cases. The 5 pg/cell level results in a manageable amount of signal loss remaining after 18 days, whereas the 15 pg/cell loading level appears to be excessive iron per cell, as the area of signal loss prevents any tumor structure from being visualized.

DISCUSSION

Through the incorporation of intracellular iron contrast agents, cellular imaging with MRI has the potential to allow for a dynamic assessment of glioma cell invasion, which may lead to more effective treatment of glioma patients. We therefore sought to develop a mouse glioma model that included the potential for cellular imaging of invasion.

While glioma tumors are clinically imaged using T2-weighted sequences and T1-weighted sequences with gadolinium enhancement, the high SNR efficiency of the bSSFP pulse sequence makes it an attractive alternative for 3D imaging in small animal models. It has recently been shown that this sequence can be used to produce contrast similar to T2-weighted RARE images of mouse glioma tumors at 4.7 T (10). The current work demonstrates that bSSFP is also suitable for imaging glioma tumors in mice at 3 T and produces images with 35% lower CNR efficiency than a T2-weighted two-dimensional FSE sequence, but with twice the SNR efficiency. bSSFP also showed enhanced CNR in the mouse brain upon intraperitoneal gadolinium injection that was superior to that seen using a T1-weighted SPGR sequence, even 20 min postinjection. This is an important observation as it demonstrates that an intraperitoneal injection of gadolinium is sufficient for excellent enhancement when bSSFP is used.

We have previously shown that bSSFP is also ideal for single cell imaging at 1.5 T and 3 T, using a high-performance insertable gradient coil, as it is highly sensitive to the presence of iron-loaded cells (5, 12, 14). MRI of iron-containing tumors poses additional challenges due to the fact that they are composed of a heterogeneous population of cancer cells, some of which divide quickly, some of which divide slowly, some of which may not divide, and some of which may begin to divide and then stop dividing for some period of time. The in vivo retention and dilution of iron labels in proliferating cells is complex and is not completely understood. In general, the length of time that a cancer cell retains its iron label will depend on its rate of division, as well as on the initial iron content of the cell.

The fact that cancer cells are proliferating also presents the need to balance between loading the cells with enough iron to curb the dilution effect of cell division and avoiding generating a large area of signal loss that would prevent adequate visualization of the tumor itself. The approach taken to address these issues included both optimization of the iron-loaded tumor model and the MRI protocol.

Advances in our bSSFP sequence enabled independent control of bandwidth and TR, as well as the incorporation of RF phase cycling, which allowed the TR to be greatly extended without introducing problematic banding artifacts into the images. We show that bSSFP can be used to generate complementary contrasts that allow for excellent tumor delineation and localization, as well as sensitive detection of iron-loaded cells. A flip angle of 40° resulted in excellent white matter–gray matter contrast and, when combined with a short TR of 6 ms, allowed for excellent tumor visualization with minimal iron effect. Decreasing the flip angle to 20° resulted in minimal white matter–gray matter contrast, allowing for easier detection of iron-loaded cells, and the extension of TR to 22 ms allowed for very sensitive detection of iron, which appears as discrete, punctate areas of signal loss in MR images.

A quantitative comparison of the MRI data with matched histologic sections validated that the strong signal loss seen in the bSSFP images was mainly due to the presence of iron. Also, despite the fact that several histologic slices contribute to any given MRI slice, it was possible to align segmented histologic slices with corresponding MRI slices and demonstrate the correspondence between iron staining and MRI signal loss. Although hemorrhage may also appear dark on bSSFP images in some cases, the area occupied by hemorrhage on the histology sections analyzed here did not correlate significantly with the area of strong signal loss on corresponding MR images. The appearance of hemorrhage on MR images depends on the stages of the hemorrhage present (17, 18). In general, the appearance of hemorrhage is more variable and diffuse than that of iron-labeled cells, which appear as punctate areas of signal loss. In the present work, it appeared that when iron was present, its strong effect on the MR signal overwhelmed the effect of hemorrhage. It is possible, however, that in some cases, hemorrhage appears similar to cellular iron in bSSFP images. Further investigation into techniques to better distinguish hemorrhage from iron-loaded cells would be beneficial.

Little difference was observed between tumors labeled with SPIO vs MPIO; however, in both cases, when comparing different iron-loading schemes, it appeared that labeling a fraction of the injected cells is beneficial and that labeling cells with a low to moderate amount of iron (5-10 pg/cell) would allow tumor structure to be visualized while still providing strong iron contrast for up to 18 days after tumor inoculation.

The high level of iron sensitivity achieved with this novel long TR bSSFP protocol may allow for longer-term tracking of iron-labeled cells. With the advent of iron-based MRI reporter genes (19–22), which involve much smaller amounts of iron than routinely achieved using iron contrast agents, such sensitive techniques are becoming increasingly important. Additionally, many disease pathologies are associated with endogenous iron deposits, for example, Alzheimer's plaques; thus, this technique has the potential to be applied to other disease models in small animals.

CONCLUSIONS

Initial investigations demonstrated that the bSSFP pulse sequence is suitable for imaging unlabeled mouse glioma tumors at 3 T, delivering tumor-brain CNR efficiency comparable to that of T2-weighted two-dimensional FSE, but much greater SNR efficiency. Furthermore, the bSSFP sequence displayed superior CNR enhancement post–gadolinium administration than T1-weighted SPGR, even 20 min after intraperitoneal injection. The implementation of multiacquisition RF phase cycling and independent control of bandwidth and TR allowed the development of a two-acquisition bSSFP protocol that allowed for excellent visualization of iron-loaded tumors with minimal iron contrast when using TR = 6 ms and α = 40°, and extremely high sensitivity to iron with minimal background contrast when using TR = 22 ms and α = 20°. Quantitative histologic analysis validated that the signal loss observed in bSSFP images of iron-loaded tumors correlated strongly with the presence of iron. While hemorrhage may sometimes appear dark in MR images, the signal loss is more diffuse than that caused by iron and it appears here that the effect of iron on the MR signal overwhelms the effect of hemorrhage.

Finally, minimal differences were observed between SPIO and MPIO; however, it appeared that loading a fraction of the injected cells was beneficial and that labeling cells with a moderate amount of iron allowed for good tumor visualization while still providing persistent iron contrast. The nonuniform distribution of signal loss within the tumor volume may indicate differences in cell division and/or metabolism throughout the tumor and warrants further investigation. The possibility that some of the signal loss is due to iron within other cell types, for example, macrophages, should not be ignored. The methods developed in this work have the potential to allow for the detection of cell invasion away from a bulk tumor and may also have application in other studies that require high sensitivity to iron.

Acknowledgements

Financial support for this project comes from the Ontario Consortium for Small Animal Imaging. B.K.R. receives salary support from the Barnett-Ivey Heart and Stroke Foundation of Ontario Endowed Chair award. L.M.B. is supported by the University of Western Ontario.

Ancillary