Patients
This single-centre feasibility study was performed at the University Medical Center Utrecht, The Netherlands. The study protocol was considered as a protocol development study, and as such it was approved by the medical ethics committee of the University Medical Center of Utrecht. All patients gave written informed consent prior to study participation.
Patients were recruited from the outpatient clinic of the Department of Medical Oncology. Patients were eligible for participation if they were aged ≥ 18 years, diagnosed with mCRC with ≥ 1 liver metastases ≥ 2 cm in diameter and registered to receive TAS-102 in the compassionate use program in The Netherlands. The eligibility criteria for this program were previously described [12]. Patients were scheduled to receive 35 mg/m2 TAS-102 twice a day, following a 28-day cycle according to routine clinical practice [2]. Patients with a contraindication to undergo MRS examinations were excluded. For this feasibility study we aimed to enrol five patients. Given that, in a previous 7-T 19F- MRS study on the metabolism of capecitabine [10], capecitabine and/or its metabolites were detected in all examinations in the two patients enrolled, we assumed that the inclusion of five patients was enough for this first exploration of TAS-102 metabolite detection.
Experimental setup
Experiments were performed on a 7-T whole body MR system (Philips, Best, The Netherlands). We used the setup described by Van Gorp et al. [10], which was previously used for MRSI of the fluorinated agent capecitabine and metabolites in the liver. In this setup, eight fractionated dipole antennas (MR Coils, Drunen, The Netherlands) were connected to the system for resonance frequency excitation and reception (Fig. 1). The coil sensitivity loss for 19F as compared to 1H is on average 7% for B1− and 5% for B1+ [10].
Phantom measurements were performed on a 25-mL tube containing TAS-102 (corresponding to 5.4 mM TFT) positioned in a body phantom together with a 25-ml tube containing 1 M NaF for chemical shift reference, at a room temperature of 18–20 °C.
Patient measurements were performed in the morning, during cycles 1–2 of TAS-102, between days 8–12 of the cycle and 30–180 min after the oral intake of TAS-102. These time points were chosen based on previously published data on the maximum concentrations of TAS-102 and metabolites in human plasma and because previous animal studies detected TAS-102 metabolites in different tissues (including the liver and tumour) 30–120 min after TAS-102 administration [13,14,15,16]. Since food intake influences the absorption and bioavailability of TAS-102, patients were instructed to take a light breakfast in the morning prior to the MRS exam to increase the maximum concentrations of TAS-102 in the body [17]. All patients wore MR-compatible clothing. The performance of the setup was regularly tested on a TFT phantom, while during each patient measurement a small NaF tube was included for chemical shift referencing, thus ensuring proper functioning of the setup.
A similar scan protocol was used for the phantom and patient measurements. B0 and B1 shimming were performed at the 1H frequency before switching to the 19F frequency. Spoiled gradient-echo images were acquired for every single transmit channel for B1 calibration (radiofrequency (RF) shimming) [18]. A Matlab-based minimisation procedure (Fminsearch, Matlab, Mathworks, Natick, MA, USA) was used to determine the offset phases for the separate transmit channels, in order to maximise the average signal in a region of interest delineating the tumour. A reference scan on 1H was performed to enable sensitivity-weighted combination of the eight receive channels. Three-dimensional (3D) chemical shift imaging (CSI) on 1H was done to enable the B1 correction for quantification. The region of interest was positioned on the largest in diameter, non-necrotic liver metastasis, aiming at encompassing as much of the substituted parenchyma as possible. In patients with more than one metastatic lesion, optimal B0 shimming was performed on the largest metastasis. Phase encoding was applied along three axes for the spatial localisation of a 3D volume surrounding the liver metastases. RF shimming was used to maximise the average B1+ signal in this region. Two subsequent 19F 3D CSI scans were performed (one with a repetition time of 25 ms and 44 excitations, another with a repetition time of 50 ms and 22 excitations) with the following technical parameters: bandwidth of 64,000 Hz; 1024 samples; number of voxels 10 × 10 × 10; voxel size 37.6 × 37.6 × 37.6 mm3; a block pulse with a flip angle of 15° with a B1 of 12 μT on the resonance for TFT. The acquisition delay due to the excitation RF pulse and phase encoding gradients was 0.44 ms. The total scanning time for each CSI examination was 14.24 min.
Post-processing
The data were apodised with a 50-Hz Lorentzian filter and a 50-Hz Gaussian filter in the time domain; Hamming filtering was applied in the spatial domain. We used a 1.5-mL phantom filled with 1 M NaF solution for frequency referencing at −45.9 ppm; thus, the TAS-102 metabolites of interest ranged between −1 to +13 ppm [14]. The signal-to-noise ratio (SNR) was automatically reported with custom-built interactive data language (IDL) software (Research Systems, Boulder, CO, USA). Considering the large number of data points that were obtained from the patients (three spatial dimensions and one spectral dimension), an SNR threshold ≥ 5.5 was used for the minimum detection limit of TFT and metabolites. This SNR was chosen at the 99.5% confidence interval, based purely on noise statistics, and an extended chemical shift range of possible metabolites between −8 and +20 ppm [10, 19]. An SNR of 5.5 implies a chance of 1 in \( \frac{1}{1-\operatorname{erf}\left(\frac{5.5}{\sqrt{2}}\right)} \) = 26330254 of a false positive signal. A complete 19F CSI dataset of one patient has 1024 frequency data points per voxel, of which 12% are located within the range from −8 to +20 ppm at the acquisition bandwidth used. With a total number of voxels of 10 × 10 ×10, this amounts to 122,880 signals, implying a 0.5% chance of a false positive signal per patient. We searched for any significant SNR in all patient examinations and all liver metastases depicted within the examinations.
Additionally we performed a second, more in-depth analysis, in which we focused on a smaller chemical shift range from 5 to 13 ppm. This chemical shift range was chosen based on previously reported chemical shifts for TFT and metabolites in the liver of rats [16]. We searched for spectra with signals above the SNR threshold of ≥ 4.8 (chosen at the 95% confidence interval).
Statistics
Continuous normally distributed variables were reported as mean ± standard deviation, while not normally distributed variables were reported as median and range.