Magnetic resonance spectroscopy (MRS) is a noninvasive technique, which is increasingly used in clinical research of breast cancer. With MRS, it is possible to reveal several metabolites of the biosynthetic pathway of the two major biological membrane phospholipids, phosphatidylcholine (PtdCho) and phosphatidylethanolamine (PtdEth) (1). The anabolic and catabolic products of these phospholipids have repeatedly and consistently been linked to malignancy (1–5). Phosphocholine (PC) and phosphoethanolamine (PE) are both anabolites as well as catabolic products of PtdCho and PtdEth, respectively. Glycerophosphocholine (GPC) and glycerophosphoethanolamine (GPE) are catabolic products of these phospholipids. PtdCho is synthesized in mammalian cells via several pathways; the major ones are the (Kennedy) cytidine diphosphate choline (CDP-choline) pathway and the phosphatidylethanolamine methylation pathway (6). The CDP-ethanolamine pathway and phosphatidylserine decarboxylation are the two dominant mechanisms by which cells are provided with phosphatidylethanolamine (1). Biosynthesis and breakdown of phospholipids are important processes for signal transduction in cell growth and proliferation, which are altered in cancer cells (7). Therefore, the metabolic products of phospholipids are the most favorable metabolites for studying breast cancer and evaluating anticancer therapies. Signals of these metabolites, such as PC, GPC, PE, and GPE, can be measured using either 1H MRS or 31P MRS.
When using 1H MRS to study breast cancer in patients only one signal related to phospholipid metabolism is generally observed. This signal is referred to as total choline containing compounds (tCho), i.e., choline + PC + GPC, but it also contains resonances of PE and GPE. Even resonances of other chemical compounds such as taurine, myo-inositol, and glucose may contribute to the observed tCho peak in 1H MRS (8). Lipid signals from the adipose tissue surrounding the glandular tissue also contaminate the tCho detection with 1H MRS, as coherent resonances of diallylic components of these lipids inside the measurement volume may overlap with tCho. Moreover, the strong methylene and methyl resonances of lipids can cause spurious sidebands, which may overlap with tCho (9). This effect is also present at higher magnetic field strength such as 4 T and beyond, thereby impeding the advantageous effect of intrinsic higher signal-to-noise ratio (SNR) of the tCho signal at higher field strength (10). Furthermore, at higher magnetic field strength B0 inhomogeneity increases due to substantial susceptibility differences between water and lipids. Therefore, even at 7 T, the signal of tCho is only detectable under ideal conditions that include superior B0 shimming, good water and fat suppression, and selection of a volume that excludes lipid tissue. As a consequence, tCho present at the edge of the tumor and the diffuse as well as infiltrative parts, which extend into the lipid tissue may not be detected with 1H MRS. Nevertheless, 1H MRS of breast cancer has been used to distinguish benign from malignant lesions based on a cut-off value of the quantified tCho level (11). Despite the fact that the reported tCho level in breast lesions varies over a wide range (0–10 mmol/kg), the overall results are consistent by the fact that malignant lesions have a higher tCho level (11), and this may be used to improve diagnostic accuracy of breast MR imaging (12).
However, as the underlying components, such as PC and GPC, of the observed tCho signal in 1H MRS change in breast cancer (3), the tCho signal—generally considered as the sum of choline, PC, and GPC—might confound metabolic changes of PC, GPC, PE, GPE, and choline individually and may become unsuitable as a biomarker for cancer. Therefore, it seems valuable to invest in the detection of the individual signals of phospholipid metabolism, such as PC, GPC, PE, and GPE, to establish a biomarker for cancer. With 31P MRS, these phosphorylated metabolites in breast cancer can be observed. Previous 31P MRS studies applied at low field on patients with breast cancer suffered from low SNR and overlapping resonances of phosphomonoesters (PME) and phosphodiesters (PDE) due to low chemical shift dispersion and broad line width (13). At 7 T, the sensitivity is significantly increased, which even allows for the selection of smaller volumes. Furthermore, the chemical shift dispersion is increased providing a better separation between the individual resonances of PME (PC and PE) and between PDE (GPC and GPE). This potentially makes 31P MRS at 7 T a unique tool to investigate the phospholipid metabolism of breast cancer in vivo.
The phospholipid signals in the 31P MR spectrum are less prone to contamination by resonances of other highly concentrated compounds compared with their signal in the 1H MR spectrum. Therefore, localization accuracy is less demanding in 31P MRS, which can be used to gain SNR. For instance, in 1H MRS, the cubic voxel has to remain within the tumor to exclude lipid signal contaminations, while presumably no 31P MR signals are observable from these lipids. In fact, the signal of all glandular breast tissue may be observed without contamination of other signals from the breast. However, phosphocreatine (PCr) and adenosinetriphosphate in the chest and intercostal muscles have relatively high concentrations that may result in contamination of 31P MRS of the breast. Nonetheless, these artifacts can be strongly reduced by applying spatial selective saturation bands to suppress these signals. Furthermore, the 31P MR spectrum can be quantified relatively easy into tissue levels of phosphorylated metabolites, which enables the comparison between subjects and within subjects in longitudinal studies.
In this study, we show a simple method for in vivo 31P MRS of the breast at 7 T. Using outer volume suppression (OVS), we are able to obtain 31P MR spectra of the entire breast in which contamination of the chest muscle is strongly reduced. As SNR is maximized, distinct resonances of phospholipid metabolites can be detected even at low concentrations as generally observed in healthy glandular tissue. The apparent T1 relaxation times of the 31P spins of PC and PE of only glandular tissue are obtained in vivo, and image-based segmentation with B field corrections is applied to enable quantification of concentrations of PE and PC in glandular tissue of healthy volunteers and patients with breast cancer in vivo.
MATERIALS AND METHODS
MR measurements were performed on a 7 T whole body MR system (Philips, Cleveland, USA). A unilateral, double-tuned, dual-element radiofrequency (RF) coil with resonance frequencies of 298 and 121 MHz for 1H and 31P, respectively, was designed with focused MR sensitivity for the detection of 31P and 1H spins in the breast. The two elements were positioned symmetrically around the breast in an orthogonal alignment. As a consequence, the uniformity is degraded resulting in a two times higher B1 at the nipple and a two times lower B1 at the chest wall compared with the B1 at the center of the breast (14, 15).
All measurements were carried out in compliance with the local institutional medical ethics committee, and written informed consent was obtained before measurements.
Eleven female volunteers participated in this study, age 25 ± 3 years. Furthermore, to illustrate the steps toward quantification of metabolite levels in breast cancer, 31P MRS data from three patients with invasive breast carcinoma were obtained.
The first patient (pat 1) was 67-year-old and diagnosed with ductal adenocarcinoma in the right breast; stage T2NitcMx (itc: isolated tumor cells). According to the American College of Radiology Breast Imaging Reporting and Data System classification, the breast density of this patient was categorized as extremely dense breast tissue; 75–100% (category 4). In this patient (pat 1), the largest diameter of the tumor/glandular tissue mass was 4.3 cm. The second patient (pat 2) was a 52-year-old woman diagnosed with diffuse ductulolobular cancer in her left breast; stage T3N2Mx and categorized according to Breast Imaging Reporting and Data System as consisting of scattered fibroglandular densities; 25–50% (category 2). The largest diameter of the tumor/glandular tissue mass was 5.5 cm. The third patient (pat 3) was a 52-year-old woman with ductal adenocarcinoma in the right breast; stage T2N0Mx, with a breast density of 25–50% (Breast Imaging Reporting and Data System category 2). The maximum diameter of the tumor was 3 cm at the time of pathological analysis. A radiologist established the percentage of tumor tissue versus total glandular tissue from clinical 3 T data based on postcontrast fat suppressed T1-weighted MRI. The total amount of glandular tissue of the all patients consisted approximately of 25–50% tumor tissue at the time of MR imaging.
B0 field homogenization was performed with an image-based shimming algorithm (16) where a measured B0 field map was used to calculate second-order calibrated shim settings on the region of interest (whole breast) that was manually selected on the B0 map. These calculated shim fields were applied during the remainder of the examination. After shimming, the 1H carrier frequency was set to zero, from this frequency the 31P carrier frequency was determined. We used the signal of PE to normalize the frequency of the 31P MR spectra; the PE resonance was set to 6.8 ppm as starting value before peak fitting.
Before MRS measurements, a fat-suppressed 3D fast-field-gradient-echo sequence was used to image the breast tissue (T1-weighted, selective water excitation, fat-suppressed, flip angle 10°, TR = 8.8 ms, TE = 2.3 ms, 1 × 1 × 2 mm3, acquisition time = 1:53 min).
31P MR spectra were obtained from the entire breast using pulse-acquired FID detection with an adiabatic excitation (90° BIR-4 pulse, duration = 8 ms, limited bandwidth = 1800 Hz). The acquisition delay was 100 μs allowing for the detection of signals from metabolites with short T2 relaxation time. To eliminate contributions of 31P MR signal from chest muscle tissue (recognized by the presence of PCr, which concentrations are undetectable in breast tissue (13)), one saturation slab was used before excitation (Fig. 1). The OVS pulse was directly followed by a crusher gradient of 2 ms with an amplitude of 33 mT/m in both y and z directions. This crusher gradient was followed by a 0.5 ms delay before the start of the excitation pulse. The OVS pulse was driven with an adiabatic full passage pulse [hyperbolic secant, 4.2 ms, 4 kHz bandwidth (Fig. 2a)] applied in the regime of obtaining a nominal flip of 90° with a dispersed phase (17). This pulse was selected because its slice profile is insensitive to the effective B1 field (Fig. 2). Although spins inside the slice are sensitive to the B1 field, spins outside the excitation bandwidth of the pulse are not affected [less than 1% (Fig. 2a)], even in a two or three times higher local B1 field (Fig. 2b and c, respectively). Furthermore, due to its broad bandwidth, the chemical shift displacement of this pulse is small. Therefore, this OVS pulse will not affect the metabolite signals inside the breast. To reach optimal suppression of the signals from the muscle, the RF power of the OVS pulse was manually adjusted for every volunteer. The extent of suppression was evaluated on the spot with a 1:36 min pulse-acquired FID measurement (Fig. 1). After adjusting the necessary power for the OVS pulse, its power was fixed for the remainder of the experiments. Three measurements of 8:32 min each with different TR were obtained. The TR was set to 4, 8, and 16 s, and the number of averages was adapted to maintain equal acquisition time of 8:32 min (Fig. 3). In pat 1, three MR spectra with TR of 4, 8, and 16 s were also acquired; in pat 2, only one 31P MRS measurement was obtained with a TR of 4 s. As the tumor of pat 3 was much smaller than the total amount of glandular tissue but was focused in a 3 cm slice behind the nipple, a 1D chemical shift imaging (CSI) measurement could completely localize the tumor. Therefore, in this patient, a 1D CSI was obtained with a 90° adiabatic excitation, a TR of 2.5 s, and total acquisition time of 12 min. A hamming-weighted k-space acquisition was used with a nominal slice thickness of 20 mm (real slice thickness was 32 mm), no OVS pulse was used in this measurement.
Two phantoms were measured, one for calibration of the concentration and one to obtain a sensitivity map of the 31P coil. The phantom that was used for peak calibration was a sphere with 4 cm diameter located on the bottom of a half sphere with the same diameter as the coil that was filled with 0.9% NaCl solution to match the loading of breast. The small sphere contained 15 mM PE and 12 mM PC. From this phantom, a reference spectrum was obtained using the same 90° BIR-4 pulse used for in vivo measurements. The TR was set to 20 s, allowing to obtain an MR spectrum without T1 weighting.
A different phantom was used to obtain a sensitivity map of the 31P coil. This phantom was a half sphere with the same diameter as the coil that was extended with a cylinder with the same diameter to cover the complete FOV of the coil. This phantom was filled with 1M inorganic phosphate (Pi). A gradient echo 31P image was created, using an adiabatic half passage pulse (6 ms, BW 1000 Hz) for excitation and a repetition time of 20 s. The resolution was 4 × 4 × 8 mm, interpolated to 1 × 1 × 2 mm, to match to the FEE images of the breast. In contrast to the 1H B1 field, the 31P B1 field in this phantom at 7 T is in the so-called near field regime. In this regime, the B1 field is linearly related to the coil load, and the load does not affect the spatial distribution of the B1 field.
The effect of different coil loadings on the signal intensity was determined on six volunteers using a network analyzer and a pick-up probe next to the coil. For both the coil elements, the average signal variation between the volunteers was assessed. In the same way, we determined the signal difference between loading with the phantoms and the average load of the volunteers.
Processing and Segmentation
MR spectra were manually phased and a Lorentzian apodization of 20 Hz was applied in JMRUI 3.0 software (18). As small motion or susceptibility differences may affect the resonance frequency, small adjustments of the frequency were done by setting the PE resonance to 6.8 ppm, allowing a shift of maximally 0.2 ppm. Single Gaussian lines were fitted to the data with the AMARES algorithm (19). The line widths of PC and Pi were constrained to the line width of PE. The resonances of GPE and GPC were also fitted but not further analyzed, because these signals had very low SNR and showed often baseline distortions due to the partially suppressed PCr resonance. We did not include fitting of possible macromolecular resonances in the PME region. The amount of glandular tissue was determined from the fast-field-gradient-echo images using FSL software (FMRIB's Software Library (20)).
The progressive saturation series were used to determine the apparent T1 relaxation time of 31P spins of PC and PE. To obtain an accurate fit of T1 relaxation time, only data with a spectral resolution that allowed fitting of PE and PC individually were used. This means that we excluded volunteer data for T1 fitting in which the standard deviation on the peak areas found in the three progressive saturation measurements were 15% or larger. These peak areas were corrected for the different TR of the measurement with the T1 value found at 3 T (21). This 15% criterion and a TR of 4, 8, and 16 s, predefines the T1 value between 2.1 and 8.4 s. As the T1 values at 7 T are expected to be lower compared with 3 T (22), this may be a valid fitting range. We choose the 15% criterion to come to an objective measure to exclude data in which PE and PC could not be fitted accurately. This criterion led to an inclusion of six from the 11 volunteers for T1 fitting. The data were fitted to the function M = M0(1 − e−TR/T1) with a nonlinear least square fit (Matlab R2009a, The MathWorks Inc.).
Quantification Based on MR Imaging and Phantom Calibration
The area under the fitted in vivo peaks of PE and PC were scaled to the area under the curve of PE and PC in the calibration phantom with known concentration. We corrected the obtained concentrations for differences in glandular tissue volume, for B1 inhomogeneity of the 31P coil, and for T1 relaxation of 31P spins. As we efficiently suppressed signals from the muscle, we assumed that all 31P signal originated from glandular tissue. The 31P-sensitivity image was matched and multiplied with a mask of the glandular tissue (Fig. 4). The same was done for the 31P sensitivity image and the segmented fast-field-gradient-echo image of the calibration phantom. The ratio between the total intensity of the new 31P image (Fig. 4d) of the glandular tissue and the calibration phantom was used to correct for volume and B inhomogeneity. Finally, data from the pickup-probe were used to correct for loading differences between phantom and in vivo experiments.
All measurements were carried out successfully according to the protocol. Because of the superb lipid suppression in the fast-field-gradient-echo MR image, simple threshold segmentation could be applied in most cases to obtain a mask of glandular tissue. In some cases, additional region drawing was necessary to exclude signal from the muscle and parts of the skin.
Because of the relatively large distance of the muscle to the coil, 31P MR signals from muscle tissue were intrinsically suppressed. In addition, an up to 6-fold suppression of these signals was obtained using the OVS. Thus, the origin of the remaining 31P MR signals of PME and PDE can be attributed to glandular tissue (Figs. 1 and 3). The signals from PE, PC, and inorganic phosphate were detected with sufficient SNR and spectral resolution for fitting with JMRUI.
Differences in coil loading between subjects and the calibration phantom resulted in 5% signal reduction. The values in Fig. 5 are corrected for this signal loss due to loading. The variation of coil loading of different volunteers was ∼0.36 dB, which leads to a variation in signal of 4% between volunteers due to loading differences.
The apparent T1 relaxation time and quantified concentration of PE and PC are summarized in Table 1 and Fig. 5. The error bars on the values in Fig. 5 represent the standard deviation of the concentrations determined in each of the three MR spectra per volunteer. Additionally, quantified data from three patients were also shown (Fig. 5). The 31P MR spectrum of pat 1 (ductal adenocarcinoma) shows some PCr residue and distorted PDE signals due to the partial suppression by the saturation slab and shifted B0 in the chest muscle (Fig. 6). The signals of PC and PE could be separately quantified (Fig. 6c). A radiologist quantified the amount of tumor tissue with respect to the total glandular tissue. In pat 1 and pat 2, ∼50% of the glandular tissue was tumor tissue. Thus, the partial volume effect in the metabolite levels shown in Fig. 5 of those patients is 50%. This means that the metabolite level of tumor tissue only is likely to be even higher. Pat 2 had a larger tumor/glandular tissue mass (5.5 cm, largest diameter) than pat 1 (4.3 cm, largest diameter); therefore, the metabolite levels of pat 1 are probably more weighted toward normal glandular tissue. Despite the partial volume of normal glandular tissue, the levels of PC and PE in the whole breast were higher in both the patients compared with the healthy females. The tumor of pat 3 had a maximum diameter of 3 cm at the time of pathological examination; at the time of the MR experiment, the tumor size could have been different. However, localized 31P MRS was obtained of the tumor in pat 3, and we used one slice that covered the tumor for the quantification (Fig. 7 slice 3). Therefore, the partial volume effect is less, but still present since the breast density was 25–50% and 50% of all glandular tissue was tumor tissue. This leads to quantified concentrations for PE and PC in pat 3 that are lower than in pat 1 and pat 2. The 1D CSI shows the effect of the different B0 in the breast and the chest muscle; the strong PCr signal (in antiphase) from the muscle is leaking into the other slices (Fig. 7).
Table 1. Metabolite Concentration and T1 Relaxation Time in Glandular Tissue of Healthy Volunteers
5.6 ± 0.8
Average concentration [mM]
1.18 ± 0.41
0.84 ± 0.21
In this study, we determined PC and PE concentrations in glandular tissue of young healthy volunteers and determined the in vivo apparent T1 relaxation time of 31P spins of these metabolites with 31P MRS. The quantification method was also applied to patients to breast cancer to evaluate the feasibility of the method for clinical research. 31P MRS enables a more specific analysis of the biochemical status of breast tissue compared to 1H MRS, because it allows pinpointing the composition of the phosphoester signals rather than the detection of tCho alone.
Although absolute concentrations of PC and PE were higher in the breast tissue of the three patients with breast cancer when compared to healthy breast tissue, the ratio of PE/PC was somewhat higher (1.96 ± 0.34 in the three patients versus 1.41 ± 0.64 in the volunteers). This is in contrast to observations of decreased PE/PC ratios in breast cancer cells and experimental tumors compared to normal cells or tissue (1, 23). However, in these studies, it appeared that the concentration of PE was influenced by the ethanolamine level in cell culture medium, and it was found that PE was not as evidently related to tumor growth as the PC concentration (24). Therefore, it was thought that PE might be less disposed to cellular control as PC. Furthermore, the level of GPC was related to the cell s-phase fraction in an experimental mammary tumor, whereas no relation with GPE was found (24). As a consequence, the scope of biochemical research was mainly focused on the role of PC and GPC in breast cancer. Many studies have for example focussed on the role of choline kinase and other enzymes in the PtdCho cycle in malignant breast cancers, but those studies do not report on PE and GPE levels (25).
Ex vivo HR MAS of human biopsies also reveals PC and PE levels; however, in the 1H HR NMR spectrum, the PE resonance is overlapping with the PC resonance at neutral pH. Therefore, also in these types of studies, researches have focussed on PC, GPC, and free choline levels in the tissue biopsies (26). The levels of PC and GPC that are reported in HR MAS studies are variable, standard deviations are larger than 20% of the average level (27). Only few studies performed ex-vivo 31P HR NMR of human breast cancer biopsies in which PE and GPE can also be studied. The work of Smits et al. showed a larger amount of PE than PC in human breast cancer biopsies, which is consistent with our observations; however, the ratio PE/PC was quite variable (28). In this study, we provide an in vivo 31P MRS method that enables to study the PE cycle as well as the PC cycle directly in human breast cancer. This could open up possibilities to verify preclinical and ex vivo observations and to study phospholipid metabolism in human breast cancer in vivo.
Despite the lower intrinsic sensitivity of the 31P spins, we were able to detect resonances of PE and PC in glandular tissue of healthy volunteers who usually have very low levels of these metabolites (∼1 mM) within 8 min of acquisition time. This was realized by using high magnetic field strength, dedicated RF coils and simple FID acquisition that avoids signal loss due to T2 relaxation. This is a significant step towards the application of 31P MRS in clinic. We provide a simple strategy for 31P MRS of the breast where an MR spectrum with sufficient SNR to evaluate the PE and PC levels can be obtained in only 8:32 min, while previously at 1.5 T measurements times of 30 min were necessary to obtain similar SNR of 31P MRS of the breast (29). Although proton decoupling and nuclear overhauser enhancement are known to increase the SNR, particularly at 1.5 T, these methods were not used in this study. At 7 T, limited gain in SNR was expected based on the broad line shape that is dominated by susceptibility effects in the breast. Furthermore, at 7 T, these methods extensively raise RF power deposition and SAR (specific absorption rates) limits are easily exceeded.
To assess, e.g., very small tumor lesions, 31P MRS can be used without problems arising from partial volume of lipids. We assumed that the lipid tissue inside the breast does not contain water soluble phosphorylated metabolites such as PME and PDE (10), or that its metabolite levels are negligible compared with their levels in glandular tissue. Therefore, the PC and PE levels will not depend on the presence of lipids in the voxel as could be the case with the tCho levels detected by 1H MRS. Particularly, in voxels with high partial volume of lipids, the 3.2 ppm resonance addressed as tCho increases with increasing lipid fraction as lipids contribute to this resonance either by baseline artifacts or by a true coherent resonance at 3.25 ppm (10).
In some volunteers, the chest muscle was in close proximity to the coil and, therefore, caused large signals of PCr and other phosphorylated compounds of the muscle in the MR spectrum. To reduce these artifacts, we used OVS. However, the OVS slab was positioned in an area of substantially lower B1 than at the center of the breast. To compensate for that the amplitude of the pulse needed to be increased 3-fold (on average) compared to the amplitude that is needed for a similar effect in the center of the breast. As a consequence of the B1 nonuniformities, the OVS does not completely saturate all signals from muscular tissue. The adiabatic BIR-4 excitation pulse was driven at twice the nominal amplitude needed to satisfy the adiabatic condition in the breast. Therefore, all 31P spins in the breast have been excited properly, but residual magnetization of spins in muscle tissue may have not reached adiabatic condition, resulting in residual phase-distorted signals of predominantly PCr in the MR spectrum. Furthermore, the PCr signal in the muscle experiences a quadratic phase evolution during the OVS pulse, which leads to a completely distorted phase. Because of the delay between the OVS pulse and the acquisition window, this residual PCr signal causes a baseline distortion that may extent beyond the original resonance of PCr, thereby degrading the PDE signal of the breast. In addition due to different B0 conditions in the breast and the muscle tissue, the PCr signal from the muscle was shifted with respect to the other resonances originating from the breast tissue. The PME have a chemical shift that is further apart from the dominant artifact of PCr, and their concentration in muscle is much lower than the PDE concentration, also observed at 7 T (30). Therefore, particularly the signals of PE and PC can be attributed to the glandular tissue and the signals from GPC and GPE were excluded from quantification. Nonetheless, combined techniques such as multiple OVS pulses (i.e., CHESS (31)) or additional localization techniques (i.e., ISIS (32)) may be used to further improve reliable detection of PDE in the breast hence opening up the possibility to use the PC/GPC biomarker (2). However, in these combined techniques, extra care must be taken that RF power deposition remains within SAR guidelines. Furthermore, subtraction techniques such as ISIS should incorporate the substantial B0 fluctuations caused by breathing and heartbeat. Our method was focused on the detection of metabolites in healthy glandular tissue, requiring a large measurement volume, because the PME and PDE metabolite levels are low, whereas in cancerous tissue, the metabolite levels may be significantly elevated allowing for additional localization techniques such as CSI (15), which reduces contamination by signals from muscle tissue substantially (also shown in Fig. 7). Therefore, in patient studies, PDE may be robustly quantifiable as well.
The used acquisition method is sensitive to macromolecular signals because of the very short acquisition delay. Therefore, the values we have found for PE and PC might have contributions from macromolecules, because their broad signals overlap with the PE and PC resonance. However, as we were not able to fit the macromolecules separately, we retain the assignment PE and PC, which are in fact apparent PE and apparent PC.
B0 shimming over the entire breast is not straight forward since breathing causes B0 frequency shifts due to changing susceptibility of the air in the lungs. In this study, we used a relatively simple second-order shimming procedure without addressing dynamic field alterations. Because of the large variation in line width obtained in the volunteers, no criteria for line width were used to exclude volunteers or patients. Although the line width in the 31P MR spectrum was narrow enough to separately detect PC and PE, quantification accuracy may be improved if gating or dynamic shimming (33) would be applied.
We used the line width of PE to constrain the fits of the line width of the other metabolites, because the line width of PE is not affected by pH of the tissue, as could be the case for Pi particularly in tumor tissue with lowered pH value. As a matter of fact, we observed a broadened line width of Pi in the MR spectra of the patients not only with respect to PE but also with respect to the healthy subjects, which probably reflects differences in pH.
The quantification method we presented is based on MR imaging, requiring only two simple calibration measurements with a phantom and one MR image of the subject to be used for assessing the volume of the glandular tissue. As a fat-suppressed image of the breast is usually acquired in any clinical breast MRI protocol, quantification of 31P MRS can be performed without the need for extra measurements during the exam. A drawback of using this fat-suppressed image for glandular tissue segmentation is that it can generate an error if the voxels contain partial volume of lipid and water. Ducts in the breast can have smaller diameters than the resolution of our MR images (1 mm isotropically). However, in relation to the limited SNR of the 31P MRS data and the standard deviation on the metabolite levels observed in this study, we expect that the bias is small, because the percentage of volume that is responsible for the missing water signal is small compared with the total water volume in the breast. To eliminate these errors, quantitative water and lipid imaging could be used, e.g., by a multiple point Dixon method (34).
To calculate metabolite concentrations, the B and B fields of the 31P coil were assumed to be equal. We corrected the metabolite levels for the difference between loading of the coil with the calibration phantom and loading of the coil with the volunteers. As the difference in coil load between volunteers leads to only a 4% difference in signal and this is an order of magnitude lower than the measured variation of metabolite levels, we used the same value to correct for the coil load in every volunteer. Finally, we corrected the metabolite levels for saturation effects due to long T1 relaxation times of 31P spins, using the average T1 relaxation time as determined in six of the 11 volunteers. No correction for T2 relaxation was necessary, because the signal was acquired directly after excitation. As we did not correct for the phase of the menstruation cycle, we assumed this to be random in our data. However, as shown by Payne et al. (29), the PME level is affected by the menstruation cycle. By correcting for these effects, the differences in our quantified levels of PE and PC between volunteers could be further reduced.
Our quantification method relied on a uniform concentration of phosphorylated choline and ethanolamine compounds in all glandular tissue of the breast. This may not be the case, particularly in breast cancer, as phospholipid levels might change in cancer cells (2, 3). Although the three breast cancer cases presented in this study had a relatively high ratio of tumor tissue versus glandular tissue, therefore, limited partial volume effects, more advanced localization strategies for MRS are needed when implementing this method in studying patients with breast cancer to improve specificity in the distinction of changes in metabolite levels in tumor tissue and in healthy tissue. Despite all possible errors made during the quantification process, the variation in the values found for the concentration of PE and PC in normal volunteers is rather low. Next to this, the metabolite levels are within physiological range and agree with literature that report on tCho levels in the breast (10). Furthermore, we found higher levels of PE and PC in the patients with breast cancer, despite the partial volume of normal glandular tissue. Therefore, we have confidence that 31P MRS of breast cancer has a large potential to robustly establish levels of PME and PDE in the tumor, which could contribute to the in vivo study of breast cancer. More detailed study of breast cancer metabolism using 31P MRS may improve breast cancer diagnosis and treatment monitoring.
We were able to detect and quantify PC and PE in the breast of healthy volunteers. Using a dedicated 31P breast coil at 7 T combined with OVS, the PME signals had good SNR and were artifact free. With routine proton imaging and phantom calibration measurements, we have shown that the PME content could be quantified to absolute concentrations such that it could be used to compare between healthy females and patients with breast cancer. In this study, patients with breast cancer had higher levels of PE and PC than healthy volunteers. Therefore, quantitative 31P MRS of the breast is feasible and enables detailed studies of breast cancer that may contribute to improved diagnosis and monitoring of anticancer therapy.
We gratefully acknowledge financial support from Marina van Damme foundation (the Netherlands) and The Netherlands Organisation for Health Research and Development, ZonMw VENI-DK2009 (916.10.163).