|Home | About | Journals | Submit | Contact Us | Français|
Poor drug delivery to brain tumors caused by aberrant tumor vasculature and a partly intact blood-brain barrier (BBB) and blood-brain tumor barrier (BTB) can significantly impair the efficacy of chemotherapy. Determining drug delivery to brain tumors is a challenging problem, and the noninvasive detection of drug directly in the tumor can be critically important for accessing, predicting, and eventually improving effectiveness of therapy. In this study, in vivo magnetic resonance spectroscopy (MRS) was used to detect an anticancer agent, temozolomide (TMZ), in vivo in murine xenotransplants of U87MG human brain cancer. Dynamic magnetic resonance imaging (MRI) with the low-molecular-weight contrast agent, gadolinium diethylenetriaminepentaacetic acid (GdDTPA), was used to evaluate tumor vascular parameters. Carbon-13-labeled TMZ ([13C]TMZ, 99%) was intraperitoneally administered at a dose of ~140 mg/kg (450 mg/m2, well within the maximal clinical dose of 1000 mg/m2 used in humans) during the course of in vivo MRS experiments. Heteronuclear multiple-quantum coherence (HMQC) MRS of brain tumors was performed before and after i.p. administration of [13C]TMZ. Dynamic MRI experiments demonstrated slower recovery of MRI signal following an intravenous bolus injection of GdDTPA, higher vascular flow and volume obtained by T*2-weighted MRI, as well as enhanced uptake of the contrast agent in the brain tumor compared with normal brain detected by T1-weighted MRI. These data demonstrate partial breakdown of the BBB/BTB and good vascularization in U87MG xenografts. A [13C]TMZ peak was detected at 3.9 ppm by HMQC from a selected volume of about 0.15 cm3 within the brain tumor with HMQC pulse sequences. This study clearly demonstrates the noninvasive detection of [13C]TMZ in xenografted U87MG brain tumors with MRS. Noninvasive tracking of antineoplastic agents using MRS can have a significant impact on brain tumor chemotherapy.
Malignant brain tumors generally have a poor prognosis. Despite surgical resection and radiotherapy, glioblastoma multiforme (GBM) is usually rapidly fatal with a median survival of 12 months. This is due, in part, to poor drug delivery and the correspondingly limited therapeutic response caused by a partly intact blood-brain barrier (BBB), blood-brain tumor barrier (BTB), and reduced tumor blood flow.1 Temozolomide (TMZ) is an alkylating agent that has been successfully used for chemotherapy of GBM and anaplastic astrocytoma. Adjuvant TMZ therapy given concomitantly with and following radiotherapy demonstrated a significantly increased median survival up to 14.6 months from 12.1 months achieved with radiotherapy alone.2,3 The ability to monitor concentrations of the therapeutic agent within the tumor noninvasively over the course of chemotherapy is important to improve therapeutic strategies and to predict and evaluate therapeutic response.
Magnetic resonance imaging (MRI) and magnetic resonance spectroscopic imaging (MRSI) are highly informative imaging modalities that provide excellent morphological and functional mapping combined with high spatial resolution and no radiation exposure. Here, we demonstrate the significance of MRI/MRS for direct detection of the drug molecules as well as for functional monitoring of the anticancer effects of therapy. TMZ is a good candidate for noninvasive monitoring of brain cancer chemotherapy by MRS because this compound is given at relatively high doses, and noninvasive detection of the drug labeled with a 13C isotope at the methyl position was demonstrated in vivo with 13C MRS at clinically relevant concentrations.4 We previously demonstrated the intratumoral distribution of carbon-13-labeled TMZ ([13C]TMZ) in xenografted human breast carcinomas in mice with MRSI using a heteronuclear multiple-quantum coherence (HMQC) pulse sequence5 with gradient selection of coherence.6 Since TMZ is currently approved for use in brain tumors, preclinical studies with a brain tumor model is a necessary step for clinical translation of the method. One major problem is that concentrations of the drug in brain tumors are generally lower than those in breast tumor models because of the reduced efficiency of drug delivery. Combined with a less efficient MR setup for brain MRI/MRS due to the reduced filling factor of the RF resonator, this results in significantly reduced sensitivity of MRS to detect drug delivery in brain tumors. Temozolomide was successfully detected by HMQC in vitro at a concentration of about 100 µg/mL (0.5 mM [13C]) with a signal-to-noise ratio (SNR) of about 11 for a nominal uniform spatial resolution of 2.5 mm (15.6 mm3). A TMZ peak was also detected in in vivo HMQC spectra of a xenografted breast tumor post-intraperitoneal injection of [13C]TMZ at a total dose of 200 mg/kg with the same nominal voxel size. In this study, to provide the necessary gain in detection sensitivity required for in vivo detection of [13C]TMZ in brain tumor models, we performed single-voxel spatially localized HMQC 1H/13C spectroscopy using outer-volume suppression,7 for spatial localization. To our knowledge, this is the first report of the direct detection of drug molecules in xenografted brain tumors using an MR technique.
To evaluate vascular parameters of brain tumor xenografts, dynamic contrast-enhanced MRI studies with a conventional, clinically approved low-molecular-weight contrast agent, gadolinium diethylenetriaminepentaacetic acid (GdDTPA), were performed. Quantitative measurements of tumor vascular parameters such as tumor vascular volume and vascular permeability-surface area product in preclinical models can also be performed using dynamic MRI with high-molecular-weight contrast agents, such as a conjugate of GdDTPA with albumin (albumin-GdDTPA).8 However, preliminary results obtained for brain tumor xenografts demonstrated a negligibly low uptake of albumin-GdDTPA in brain tumors as reported previously.9 Also, high-molecular-weight MR contrast agents are currently not approved for clinical use.
In this study, a combination of contrast-enhanced T1 dynamic MRI and susceptibility-weighted bolus tracking following two consecutive i.v. injections of low-molecular-weight contrast agent, GdDTPA, was used to extract vascular parameters of the tumor, such as vascular volume, blood flow, and an effective volume transfer constant for the contrast agent.10 Noninvasive detection of TMZ in brain tumor-bearing mice following intraperitoneal administration of [13C]TMZ was performed using an inverse-detection HMQC pulse sequence with outer-volume suppression for volume selection.
Carbon-13-labeled TMZ (99% 13C at the methyl position) was obtained from Cambridge Isotope Laboratories, Inc. (Andover, MA). 3-(Trimethylsilyl)propionic-2,2,3,3-d4 acid sodium salt (TSP) was obtained from Sigma-Aldrich Co. (St Louis, MO). Magnevist (GdDTPA) was obtained from Bayer Healthcare Pharmaceuticals, AG (Leverkusen). All other chemicals were of reagent grade and were obtained commercially. Double-tuned volume coil (1H/13C) for 9.4T Bruker Biospec horizontal bore animal MR scanner was developed and built by Resonant Research LLC (Baltimore, MD).
Human malignant glioma U87MG cell lines were grown in Eagle's minimum essential medium with 1% penicillin, streptomycin, and 10% fetal bovine serum at 37°C with 5% CO2. Approval from the institutional animal care and use committee preceded all animal experiments in the present study. U87MG cells were inoculated intracranially in immune suppressed male severe combined immunodeficiency (SCID) mice (body weight of ~22 g) that were purchased from NCI (Bethesda, MD). Briefly, 1 × 105 cells in 2 µL of Hanks' solution were implanted by intracranial injection using stereotactic guidance. The skull was exposed, and a burr hole was drilled through the skull 2 mm lateral and 3 mm anterior to the bregma. Tumor cells were injected over 5 minutes into the brain parenchyma at a depth of 2.5 mm. After inoculation, the scalp was sutured. Three weeks following cell inoculation, animals with tumor sizes of over 3 mm (the shortest axis) were used for MR experiments. At least three mice were used for the experiments.
Magnetic resonance imaging and MRS studies were performed with a horizontal bore Bruker Biospec 9.4T MR scanner equipped with 121-mm shielded gradient systems. The Paravision 3.0.2 program (Bruker Biospin GmbH) was used as acquisition software. A TMZ phantom was prepared with 100 µM [13C]TMZ in saline in a 15-mL plastic tube. Carbon-13-labeled TMZ signals were measured using an inverse-detection HMQC pulse sequence with echo time (TE) = 20 milliseconds; repetition time (TR) = 1500 milliseconds; number of acquisition (NA) = 2048; dummy scan = 8. Three-dimensional orthogonal volume selection was performed with 3 pairs of slice-selective semi-adiabatic secant (sech) pulse (pulse width = 1 millisecond, excitation slice thickness of 16 mm) used for outer-volume suppression. Each pair was used to excite and saturate the proton magnetization in parallel bands in x, y, and z planes, respectively. Only magnetization from the unsaturated rectangular volume selected by these bands contributed to the measured MR signal. A composite-pulse decoupling sequence, WALTZ-16 (γB2 = 1 kHz) applied through the 13C channel and set at the resonance frequency of [13C]TMZ of 49 ppm, was used for broad-band decoupling of 13C during acquisition of the proton signal.
Mice were initially anesthetized with ketamine/acepromazine mixture (50 and 5 mg/kg, respectively, in saline) via intraperitoneal injection and immobilized in a plastic cradle positioned within the double-tuned 1H/13C radiofrequency (RF) coil. The tail vein was catheterized for GdDTPA injection before placing the probe in the magnet. A second catheter was placed i.p. for injection of [13C]TMZ solution. A cylindrical phantom containing 100 mM TSP solution was placed on the head of the mouse, and used as an external reference. For the duration of MR experiment, the animal was kept under gas anesthesia with 1% isoflurane in an air flow of 1 mL/min. Body temperature was maintained at 37°C by heat generated from a pad circulating with warm water. Mouse respiration was monitored with a dedicated small animal physiology monitoring system attached to the MR scanner.
The first i.v. injection of the GdDTPA contrast agent (167 mM solution in saline, total volume of 30 µL over 3 seconds) was performed during T1-weighted MRI scanning of the mouse brain. Briefly, a saturation recovery snapshot-FLASH (fast low-angle shot) pulse sequence with an excitation pulse flip angle of 10°, a TE of 1.245 milliseconds, and three T1 saturation recovery delays (250, 500, and 1000 milliseconds) was used to generate quantitative T1 maps of the mouse brain. An M0 map was acquired once with a recovery delay of 10 seconds prior to the contrast administration.8 Either 2 or 3 slices were selected with a slice thickness of 2 mm. Multislice T1 maps were acquired over a time period of 7 minutes with a temporal resolution of 6.5 s/slice, and GdDTPA was injected after 4 precontrast acquisitions to provide baseline T1 maps. The second bolus injection of 40 µL of GdDTPA was performed after completing T1 dynamic MRI studies. T*2-weighted FLASH susceptibility-weighted MRI scan was used to characterize vascular volume and perfusion in xenografted brain tumor and mouse normal brain. Acquisition parameters were as follows: TE = 3.5 milliseconds; TR = 333 milliseconds for 3 slices, and 500 milliseconds for 2 slices for temporal resolution of 1 second per multislice image. The same slices were selected as for T1-weighted acquisition, and acquisition sequence was repeated for 60 seconds before and 240 seconds after the bolus over a time period of 300 seconds. For both experiments, in-plane spatial resolution of 0.125 mm (128 × 64 matrix zerofilled to 128 × 128, field of view = 16 × 16 mm) was used.
After completing dynamic MRI experiments, the probe tuning to 1H and 13C was reconfirmed, and tri-planar scout images were acquired to determine the position of the selected volume. Radiofrequency power in both channels was adjusted to optimize HMQC natural abundance 13C spectroscopy of the TSP phantom. Outer-volume suppression with 6-slice–selective sech pulse was implemented to select a cubic region of interest (ROI) using graphical prescription routine. Automatic and manual shimming routines using a point-resolved spectroscopy method were consecutively used to shim the region of the TSP phantom and the tumor region of the mouse brain. The HMQC MRS of brain tumor was performed before and after i.p. administration of [13C]TMZ. Acquisition parameters were as follows: TE = 20 milliseconds; TR = 1500 milliseconds; NA = 256; sweep width (sw) = 4960 Hz; dummy scan = 8. Similar to the in vitro study, the HMQC pulse sequence was used for indirect 1H/13C MRS, and adiabatic pulse (sech, 1 millisecond) and a WALTZ-16 composite-pulse decoupling sequence were used for voxel selection by outer-volume suppression and broad-band 13C decoupling, respectively. Carbon-13-labeled TMZ (15.5 mM) was injected via i.p. catheter every 10 minutes up to a total of 1.0 mL and a total dose of 3 mg in a 22-g animal (450 mg/m2). The acquisition was continued for 60 minutes with a temporal resolution of approximately 6.5 minutes. After the acquisition was complete, animals were sacrificed, and extracted brain tissues were individually immersed in cold formalin for 2 hours, then 30% sucrose overnight for hematoxylin–eosin staining (see Supplemental Material, Fig. S1).
A MatLab (The MathWorks Inc.) script was used for the analysis of vascular parameters, and custom-written software in the IDL programming environment (Research Systems Inc.) and CSX3 software developed by Dr Peter B. Barker (The Johns Hopkins University) were used for the analysis of MRS data.
T*2-weighted signal changes are susceptible to T1 enhancement effect of GdDTPA because of extravascular leakage. In this case, ΔR*2[t] recovers to values above baseline, which may cause an underestimation of relative cerebral blood volume (rCBV) unless adequate methods are used to diminish or correct for this effect.11 To prevent this problem, in this study T*2-weighted FLASH MRI was always performed after T1-weighted saturation-recovery MRI. The first injection of GdDTPA for T1-weighted MRI served as a preload of a contrast agent for bolus-tracking experiments.11 The concentration of contrast agent is proportional to ΔR*2, but the proportionality constant is specific for tissue, magnetic field, and contrast agent. As the proportionality constant is unknown, only relative cerebral and tumor blood flow (rCBF and rTBF) and rCBV and tumor blood volume (rTBV) values are reported in this study.
Calculation of vascular parameters was performed using block-circulant deconvolution with manual localization of a suitable arterial input function (AIF) (criteria for choosing the voxel containing an artery were high peak and low time to peak).12 The result of the deconvolution is a so-called residue function multiplied by BF. The residue function describes how much contrast is left in the vasculature. Ideally, it starts at 1 and decreases to 0; however, because of a delay between AIF and tissue curve, it may not start at 1. For this reason, BF was calculated as the maximum value of the BF times residue function. Blood volume was calculated as the integral of this function. Calculating BV as the integral of the BF times residue function eliminates errors introduced by recirculation but may not compensate for susceptibility induced baseline changes after injection of contrast. On the basis of the T1 difference between precontrast and postcontrast, a ROI was drawn encompassing the tumor. Furthermore, an equally sized ROI was drawn in the contra lateral hemisphere. Blood flow and BV were calculated in each ROI and the ratio of these is reported.
A single voxel spatially localized spectrum of a [13C]TMZ phantom prepared in a plastic 10-mm diameter tube is shown in Fig. 1. Selection of the ROI with a volume of approximately 1.2 cm3 was performed by a combination of 6 slice-selective semi-adiabatic hyperbolic sech pulses.7 As shown in Fig. 1, we could reliably detect 100 µM of [13C]TMZ in 1.2 cm3 voxel in vitro in the phantom. The measured SNR obtained with the acquisition time of approximately 50 minutes was 26.8. A complete suppression of the bulk water signal at 4.7 ppm was achieved in in vitro experiments. For typical in vivo acquisition parameters with a relaxation delay of 1.5 seconds and total acquisition time of 30 minutes, the SNR was proportionately reduced to about 12.5 (data not shown). In vitro studies with a [13C]TMZ phantom demonstrated the feasibility of this technique to detect [13C]TMZ peak at 3.9 ppm in vivo in a xenografted U87MG brain tumor model.
Experimental protocols for MRI/MRS studies are summarized in Fig. 2. Experiments started with the first GdDTPA injection and the acquisition of quantitative T1 maps from the tumor region of the brain (Fig. 2A). T1 measurement studies were followed by the bolus-tracking experiment (Fig. 2B), which demonstrated that the initial drop in T*2-weighted MR signal quickly recovered in the normal brain, whereas slow recovery was detected in brain tumors (Fig. 3). This is a common phenomenon typically observed in normal brain and brain tumors.13,14 Complete or partial breakdown of BTB, and/or BBB in the tumor, and impaired tumor vascularization are two major factors contributing to this effect.14 Typical reconstructed images of ΔR1 differences, CBF and CBV, are presented in Fig. 4. The rTBF (rTBF = TBF/CBF) and rTBV (rTBV = TBV/CBV) values reconstructed from these maps are shown in Fig. 5. Both TBF and TBV were relatively higher than those in normal brain tissue, which indicates that U87MG tumors are well vascularized, have sufficient blood flow, and permeability of the BBB and/or BTB within the tumor was significantly compromised.
After completing dynamic MRI studies with GdDTPA enhancement, inverse detection HMQC spatially localized single-volume spectroscopy was performed as outlined in Fig. 2C. Initially, all acquisition parameters were adjusted for a cylindrical phantom filled with 100 mM TSP solution that was positioned next to the animal skull as shown in Fig. 6A. An indirect 1H/13C spectrum of the methyl resonance of TSP at 0 ppm is shown in Fig. 6B. WALTZ-16 (γB2 = 1 kHz) broad-band decoupling sequence applied to the carbon-13 RF channel centered at the 13C methyl resonance frequency of TSP (0 ppm) provided complete decoupling of the proton signal.
Three-dimensional volume selection of the brain tumor was performed as shown in Fig. 6C with 6 slice-selective sech pulses. A [13C]TMZ peak was detected in the proton spectrum of the tumor at around 3.9–4.0 ppm, and a natural abundance lipid peak was registered at 1.3 ppm. A [13C]TMZ peak was consistently detected in five tumor-bearing mice following i.p. injection of [13C]TMZ (SNR = 3.04) and reached its maximum amplitude at 30–50 minutes post-initial injection of [13C]TMZ. As for in vitro studies, an excellent suppression of the bulk water peak at 4.7 ppm was achieved in in vivo experiments using the HMQC acquisition technique with gradient selection of coherences.5 The detection of [13C]TMZ peak in U87MG xenografted tumors was attributable to (i) the partial breakdown of BBB and (ii) well-vascularized character of U87MG xenografts, as demonstrated by dynamic MRI. This experiment corroborated the feasibility of noninvasive detection of delivery of 13C-labeled drug molecules to brain tumor using spatially selective inverse detection HMQC-based MRS.
Noninvasive detection of [13C]TMZ in brain tumor by MRS allows repeated monitoring of TMZ during the course of therapy, which can provide a better opportunity to optimize the treatment plan by elucidation of the relationship between TMZ delivery and antitumor activity when combined with MRI. Although O6-methylguanine-DNA methyltransferase (MGMT) is known to be associated with TMZ resistance, our preclinical animal studies with human breast cancer xenografts showed practically unchanged MGMT expression levels in tumors treated with TMZ.15 In addition, minimal involvement of MGMT in TMZ resistance in glioma was reported by Bocangel et al.16 Similarly, there is no involvement of traditional multidrug resistance drug-efflux pump mechanisms in TMZ resistance.17 It was conceivable that changes in vascular parameters in tumor resulted in poor response to chemotherapy,15 which might be directly linked to delivery issue. The dynamic MRI results demonstrate that the U87MG tumors are well vascularized, which is also supported by an early report.18 Furthermore, the R1 difference maps show at least partial breakdown of the BBB. The rTBF/rTBV values (1.26/1.55) obtained are in good agreement with experimental results obtained for the high-grade gliomas in a recent arterial spin labeling (ASL)/dynamic susceptibility contrast (DSC) study.19 Another study using only ASL showed decreased TBF and TBV in U87MG tumors in mice.20 The inconsistency might be explained by differences in magnetic susceptibility between the tumor and normal brain tissue, which can be caused by differences in angiogenic state, vascular architecture, hematocrit and other factors.21
The T1 effect of GdDTPA leaking into the tissue via partly broken BBB or TBB can be a problem in DSC studies. In this study, the R1 measurements were performed prior to DSC, and the gadolinium injected for the R1 maps acted as a preloading for the DSC measurement, which should minimize T1 effects.11,22,23 On the other hand, in this study, the signal in the brain tumor does not return to baseline during the time in which we sampled, due to the residual susceptibility and/or dipolar T2 leakage effect, which means that TBV may be overestimated.11 Partial breakdown of the BBB is also responsible for incomplete recovery of a T*2-weighted signal in the brain tumor region, which means that leakage of a contrast agent, GdDTPA, into the brain tumor occurs.11 A recent article by Paulson et al.11 has discussed different analysis and acquisition methods for DSC. They demonstrated that preloading minimized T1 effects and that the subsequent choice of postprocessing method had little impact on the rCBV results obtained. The main goal of measuring perfusion and blood volume in this the study was to render delivery of [13C]TMZ probable and not to obtain absolute values. Therefore, we used a preloading dose of GdDTPA to minimize T1 effects and did not correct for the residual susceptibility. This can cause slight overestimation of rTBV in this study.
Brain tumor region selected for MR spectroscopy was chosen from the enhancing area in T1-weighted images. The same area is also characterized by incomplete recovery of the T*2-weighted signal loss, suggesting efficient extravasation of the contrast agent in this area. In a similar fashion, a small molecular weight agent, [13C]TMZ, also can leak and distribute into this brain tumor region. The efficient delivery of [13C]TMZ to enhancing areas of the brain tumor enabled the noninvasive detection of the carbon-13-labeled drug, in this experimental model. It is also known that the effect of residual GdDTPA on both 1H and 13C spectroscopy is minimal.24,25
In this study, we were able to detect [13C]TMZ by HMQC in mouse brain tumor with a voxel size of 0.15 cm3 at 9.4T. Translatability of this technology to the clinic is a key issue. High-dose administration of TMZ was tested in humans with a dose ranging from 750 to 1000 mg/m2.26,27 Clinically, patients with large brain tumors with volume of 15 cm3 and above would be the best candidates for such examination. A simplified comparison based on the assumption that the drug is uniformly distributed in the body allows us to estimate the total amount of TMZ in the imaging volume as approximately 0.02 mg in a mouse (0.15 cm3 voxel) and approximately 0.3 mg in a human tumor (15 cm3 voxel). Although the MR detection efficiency could be higher in preclinical experiments due to the higher magnetic field and arguably better RF efficiency, high-field clinical MR scanners with a magnetic field strength of 7.0T are available at several medical centers. To extrapolate our results to human studies with high-field MR scanners, we compared SNR of the N-acetyl aspartate (NAA) peak in normal mouse brain (voxel size of 0.064 cm3) with normal human brain at 3.0T (voxel size of 16 cm3) (see Supplemental Material; Fig. S3). A comparison of the SNR values of the NAA peak in mouse/human normal brain and SNR of [13C]TMZ peak in mouse brain tumor and phantoms confirms that the method should provide sufficient sensitivity to detect [13C]TMZ peak in human brain tumors using high-field MR scanners at 7.0T and high dose of TMZ.
The technique described here is designed to monitor drug delivery in the whole tumor, and not for acquisition of spatial distribution of the drug as in our previous report.6 The limitations of spatial distribution of drugs in brain tumors are attributed to insufficient drug delivery and a relatively small tumor size compared with other solid tumors in preclinical animal models. This study has clearly demonstrated noninvasive detection of [13C]TMZ in xenografted U87MG brain tumors with MRS. The relative volumes of xenografted U87MG brain tumors used in this study are comparable to those observed in patients with low-grade astrocytoma, which often forms large tumors. Although it is also possible to use the pharmacokinetics of an appropriate surrogate marker of drug delivery, such as GdDTPA, with MRI to follow and predict the delivery pattern of the drug to the tumor, this method has limited feasibility for brain tumors with a partly functional TBB. Unlike MR contrast agents, many anticancer agents used for brain cancer therapy, including TMZ, can penetrate the BBB, and their pharmacokinetics in the tumor can be vastly different from surrogate molecules such as GdDTPA. The major advantage of MRS is the direct monitoring of drug molecules, in this case TMZ. Although we still need to overcome many barriers for clinical translation, this study is an important first step toward noninvasive detection of drug delivery in brain tumors. We envision that noninvasive monitoring of [13C]TMZ in brain tumors by spatially selective inverse detection HMQC-based MRS will lead to more effective strategies for brain tumor therapy.
This study was funded by the National Institutes of Health (R01 and R56 CA097310 to D.A.).
The authors are very grateful to Dr Bachchu Lal for his help in optimizing brain tumor cell inoculation. The authors thank Dr Zaver M. Bhujwalla and Dr Arvind P. Pathak for their helpful discussions and critical reading of the manuscript.
Conflict of interest statement. None declared.