Animal experiments were conducted in compliance with the French guidelines for the care and use of research animals (authorization 67-48220). Sixteen adult female Swiss nu/nu mice with a mean weight of 20 g were anesthetized (100 mg/kg ketamine and 5 mg/kg xylazine) and 5 × 10⁵ human glioblastoma cells (U-87 MG cell line obtained from the American Type Culture Collection [Rockville, MD]) suspended in 5 μl of PBS were inoculated with a Hamilton syringe into the striatum of the right hemisphere via stereotactic injection (1.8 mm lateral and 0.1 mm anterior to bregma, at a 3 mm depth from the skull surface). Inoculation time was 10 min to limit backflow along the injection tract.

The MVA-FCU1 suicide gene protocol used in this study is described in details elsewhere [5]. Briefly, tumors were allowed to grow for 5 days before stereotactic injection (day 5) of 5 μl of control vehicle (PBS) or 1 × 10⁶ PFU of MVA-FCU1 vector using the same coordinates on the stereotactic frame, as for tumor cell implantation. One day after injection (day 6), half of the animals were given 5-fluorocytosine (5-FC) by oral gavage at 100 mg.kg⁻¹ (0.5 ml 5-FC 0.5% in water) twice a day during 14 days (until day 20). As a result, four groups of four mice were obtained: sham group, sham group treated with 5-FC only, sham group with injection of MVA-FCU1 vector only, therapy group administered with MVAp11K7.5-FCU1 vector and 5-FC. To estimate the evolution of the intracerebral tumor volume, animals were examined with magnetic resonance imaging (MRI) every 7 days starting on day 7 after the tumor inoculation. At the onset of hemiplegia, moribund state, or inability to feed or drink, animals were euthanized by an overdose of anesthetic.

MRI studies were performed using a preclinical dedicated 0.1T resistive magnet (Bouhnik SAS, Velizy-Villacoublay, France) as previously described [8,9]. During imaging procedures, mice were maintained under gaseous anesthesia (isoflurane 1–1.5% in air) in a warmed-up MR compatible technical cell dedicated to small animal imaging (Minerve, Esternay, France) that aimed at maintaining the animal homeostasis and that standardized animal positioning. After intraperitoneal administration of 1 mL of a Gd-DOTA contrast agent (1/4-dilution in physiological saline solution; Dotarem, Guerbet, France) at a dose of 5–6 mmol kg⁻¹, both T1- and T2-weighted images were acquired covering the entire mouse brain. An in-house built radiofrequency solenoid copper coil (2 cm in diameter, five turns, Q factor of 200) was placed around the animal head's before its snout was driven into the anesthesia mask with a bite bar. The air inside the imaging cell was heated to 37 °C and an inductive coil was slid along the outer tube of the technical cell for coupling at 4.26 MHz corresponding to the ¹H resonance frequency at 0.1T as shown in Fig. 1. The technical imaging cell was placed inside the vertical magnetic field of the MR imager and the animal positioning was controlled with a scout sequence (50 sec) before use of 3D gradient echo imaging sequences to produce T1- and T2-weighted images. The complete imaging procedure including positioning of the animal in the MR imager lasted about 1 h 45 using the following imaging sequence parameters: FLASH T1-weighted sequence [10], isotropic voxel 0.5 mm × 0.5 mm × 0.5 mm, TR/TE 100/7 ms, flip angle 80°, cubic FOV = 32 mm, acquisition matrix 64 × 64, 64 slices, NEX = 8, total acquisition time 56 min; CE-FAST T2-weighted sequence [11], reconstructed voxels 0.5 mm × 0.5 mm × 1 mm, TR/TE 14/21 ms, flip angle 70°, cubic FOV = 32 mm, acquisition matrix 64 × 48, 32 slices, NEX = 60, total acquisition time 32 min.

The 3D reconstructed MRI data were processed using the Image J software (Rasband WS, Image J, US National Institute of Health, Bethesda, Maryland, USA, http://rsb.info.nih.gov/ij/, 1997–2009). The brain tumor segmentation was obtained in each axial slice by selecting high intensity pixels related to the tumor using a threshold, the value of which being determined manually both by visual data analysis and by considering the data grey level histogram. The thresholding step was followed by manual erasing of selected non brain related areas, e.g. scalp or muscles in T1, and fat or eyes in T2-weighted data. The number of selected voxels was then multiplied by the individual voxel volume (125 × 10⁻³ mm³ and 250 × 10⁻³ mm³ respectively in T1- and T2-weighted images) in order to calculate the volume of the tumor. Data were processed twice by the same observer to test for intra-observer variability and by a second observer to test for inter-observer variability. To assess efficacy of the suicide gene therapy, the tumor doubling times (t_D) were calculated for each animal using every volume measurements by fitting the experimental T1 and T2 curves under assumption of geometric increase [12]: V_t = 2(V₀)^t/t_D, where V₀ is the initial tumor volume (day 7) and V_t is tumor volume at time t.

Animals were sacrificed at different times post-injection by administration of a lethal dose of Pentobarbital^®. They were exsanguinated by transcardial perfusion first with 0.1 M Phosphate Buffer (pH7.4, ice-cold, 20 ml/mouse), then with a 4% formaldehyde solution in PBS (ice-cold, 5 ml/mouse). Brains were dissected out of the skull and rinsed in PBS before Tissue Tek^® OCT (Sakura Finetek, France) inclusion and rapid freezing on dry-ice. Twenty-micrometer thick transversal sections were cut in a cryomicrotome and placed on poly-L-Lysine-coated slides. Sections were washed in distilled water and mounted in Mowiol^® (Calbiochem, Germany).

Statistical analysis for comparison of tumor doubling times between groups was performed using the non-parametric Mann-Whitney U test. For inter- and intra-observer tumor volume measurements, a t test was used. A value of p < 0.05 was considered statistically significant.

Previous investigations have demonstrated a good correlation between MRI results and histological measures of brain tumor volume in small animal models, [13–15]. Based on these results, no comparison between histological brain volumes and MRI results was undertaken in this study. Up to now, MRI follow-up of brain tumor growth in mice relies on the measurement of the tumor's largest radius, under the hypothesis that this radius reflects the actual tumor volume, or using ellipsoidal modeling of the volume based on measurements of largest tumor dimensions, or slice by slice manual outlining of the tumor [13,14,16–23]. Reconstructed data in all these studies using magnetic fields ranging from 1.5 T to 11.7 T were anisotropic, with a slice thickness four to 10 times larger than the in-plane resolution, leading to reconstructed voxel volumes ranging from 6 × 10⁻³ mm³ [20,21] to 40 × 10⁻³ mm³ [14] (whenever the voxel volume was given or could be derived from the data given). In addition, only some of these studies performed several sequential tumor volume measurements (two measurements [20,14], four measurements [16], five measurements [18], between four and nine measurements [20]) reflecting the tumor growth kinetics. In addition to tumor volume growth measured by T1- and T2-weighted images, diffusion MRI has been successfully evaluated as a biomarker of early treatment response because the apparent diffusion coefficient (ADC) of water is sensitive to changes in tissue cellularity [23]. ADC increases as water mobility, induced by cell kill, increases in extracellular spaces [24]. However, we decided not to use diffusion MRI sequences even knowing that the ADC is independent of the magnetic field strength [25]. Our principal reasons in this study were first to demonstrate that low field MRI is able to accurately follow tumor growth with the advantages of lower installation and running costs [9] and second, to avoid lengthening of in vivo exams as T1- and T2-weighted images acquisition time was 1 h 45 per mouse at 0.1T.

In our T1 protocol, we chose to acquire an isotropic 3D-volume with a voxel volume of 125 × 10⁻³ mm³. A lower signal in T2-weighted data acquisition does not allowed to keep the same imaging strategy leading to voxel volumes of 250 × 10⁻³ mm³. Tumor volumes were directly measured in every data set, without making any assumption on the tumor shape [15]. There are obviously differences when comparing voxel resolutions and signal to noise ratio at 0.1T versus those obtained at high and very high magnetic fields reported above (to that respect see the extensive discussion in [9]). However, to test the relevance of any therapy aiming at reducing a tumor volume or able to delay its growth, only significant tumor volume differences are expected during follow-up when compared to the voxel size. For that reason, we think that low field MRI was adapted to test in vivo the efficacy of the proposed suicide gene therapy protocol directed against brain tumor growth in an orthotopic human glioblastoma mice model.

The relatively long imaging procedure (1 h 45 per mice), which was repeated weekly, did not impact on animal survival because we used gas anesthesia and used a warmed non-magnetic imaging cell. For comparison, the images acquisition times obtained at higher field were 45 min for a set of T2-weighted images at 4.7T [4] and 42 min at 8.45T for T1- and T2-weighted images [14].

Manual determination of a tumor volume based on ellipsoidal modeling using bi-dimensional measurements of corresponding larger radius in selected MRI slices is an approximate but classical method [26]. However, the use of a slice by slice manual thresholding method allowing defining highlighted pixels in the T1 and T2 images for segmentation prior to tumor volume determination is a time-consuming task. Moreover, and even if inter- and intra-observer differences in tumor volumes measured in this study were not statistically significant, reproducibility of manual methods for MR volume determination requires experienced observers. To overcome these difficulties, a vectorial multichannel-multiphase brain tumor segmentation technique aiming at reducing the number of operator dependent actions to the single choice of a volume of interest (VOI), which may include the whole brain, has been proposed recently by our group [27]. Application of such semi-automatic methods should greatly facilitate image data analysis in future works.

The T1-weighted volume growth curves were similar among the three sham groups showing that both the tumor model and MRI method were reproducible. The T2-weighted volumes also followed similar growth patterns but showed a greater heterogeneity, which reflected the individual variability in edema volume, but also a larger image slice thickness and a lower signal to noise ratio compared to T1 acquisitions. The tumor volumes calculated among the sham groups were similar and comparable to the ones calculated in the treated group after the end of the treatment administration. The significant differences observed in tumor doubling times between the sham groups and the treated group confirmed that the intratumoral injection of MVA-FCU1 vector in combination with 5-FC oral administration neutralized the growth of the intracerebral glioblastoma tumor during the 2 weeks of 5-FC administration in agreement with previous in vivo results obtained in subcutaneous colon carcinoma implanted tumors in nude mice [9]. A separate study (J. Kintz et al., manuscript in preparation) was conducted in order to determine the potential toxicity to normal brain of the MVAp11K7.5-FCU1/5-FC protocol. Brains from mice were injected with MVA-FCU1 and mice were treated with 5-FC per os for 14 days. No noticeable side effects were observed during the time of the experiment and mice were still alive 3 months after virus inoculation.