Imxage contrast in standard clinical magnetic resonance imaging (MRI) is based on the concentration of protons associated with water and the environment in which they exist. Conventional MRI acquisition schemes (e.g., T2- and T1-weighted imaging) reveal anatomical information and reflect tissue integrity but suffer from the inability to specifically detect the underlying microstructural tissue composition. A recently developed contrast mechanism known as chemical exchange saturation transfer [CEST (1, 2)] is sensitive to metabolite–water proton interactions. By exploiting the chemical exchange of protons associated with metabolite and those of water, CEST MRI can detect low-concentration metabolites without the use of exogenous contrast agents, potentially providing information about microstructural tissue composition not available via conventional MRI measures.
The goal of CEST MRI is to indirectly image exchangeable metabolite protons that resonate at frequencies different (Δω) from those protons associated with bulk water, Δω = 0. This is accomplished by selectively saturating metabolite protons using low-bandwidth radiofrequency (RF) irradiation. This saturation is then transferred to the bulk water protons via direct chemical exchange, resulting in an attenuation of the measured water proton signal (3). If these nuclear spins are of sufficient concentration and in the slow to intermediate exchange regime (i.e., their exchange rate is less than or equal to the irradiation offset frequency), then CEST contrast can be observed.
Amide proton transfer (APT) CEST detects the transfer of saturation from amide protons of protein/peptide backbones. It results from selective irradiation of the amide proton resonance at 3.5 ppm downfield from water and is one of the most common endogenous CEST experiments performed in vivo. The magnitude of the APT effect depends on both the concentration of amide protons and the exchange rate. Therefore, APT measurements have been used to study tissues where either the protein/peptide concentration or the pH may be altered (4–7). For example, APT imaging has been applied to brain tumors both in animals (6) and in humans (5, 8) where it has been shown to be sensitive to increases in endogenous cytosolic proteins and peptides (9). More specifically, APT imaging has been used to differentiate cellular protein content between tumor and healthy cells (10). In vivo brain tumor studies have shown APT can: (1) separate surrounding edema from tumor (11), (2) grade tumors (8), (3) detect nonenhancing high-grade tumors (12), and (4) separate treatment effects from tumor progression (13). However, the application of CEST for monitoring therapy response has been minimal and, in particular, there have been no examples of using it to assess the response of breast cancer to neoadjuvant chemotherapy.
In the case of neoadjuvant chemotherapy, no quantitative MRI “marker” is available to determine whether a treatment is eradicating micrometastatic disease; neoadjuvant administration allows the primary breast mass to function as this marker. If the primary breast tumor responds to neoadjuvant chemotherapy, systemic micrometastases are also presumably responding. If the primary tumor continues to grow, the treatment can be changed to a regimen, which could be more effective for both primary and metastatic disease. Existing clinical methods include assessing frank changes in tumor size as measured by physical exam, conventional MRI, and/or ultrasound, but these methods still lack specificity. The tumor environment is typically characterized by changes in the vasculature (14) and cellularity (15). Due to the complicated nature, there is a need for quantitative imaging techniques that are sensitive to unique tumor tissue properties. Dynamic contrast-enhanced MRI reports on tumor vasculature (16, 17), while diffusion-weighted MRI is sensitive to tumor cellularity (18), making these advanced MRI techniques highly relevant for evaluation of therapeutic efficacy. Furthermore, PET has also been used to monitor the response to therapy in treatment of patients with breast cancer through its sensitivity to glucose metabolism using fluorodeoxyglucose (FDG) (19–21) and thymadine using fluoro-L-thymidine (FLT) (22). A noninvasive imaging method that can assess changes in relevant breast tumor biology prior to gross morphological changes could greatly improve clinical breast cancer care by providing complementary biochemical information. APT is sensitive to cellular mobile protein content as well as tissue pH, and as these tissue properties are known to change in response to therapy, it is reasonable to hypothesize that changes in APT observed early in the course of therapy can be predictive of eventual response. Toward this end, we experimentally optimized an APT sensitive CEST acquisition strategy for application in the human breast and, as a first evaluation, sought to determine the repeatability of APT measurements in healthy controls. Additionally, we provide preliminary data on the ability of APT to assess the response of locally advanced breast cancer to neoadjuvant chemotherapy.
A phantom comprised of bovine serum albumin (BSA; Sigma–Aldrich Corp., St. Louis, MO) with 15% by mass concentration was prepared in distilled water. The BSA was cross-linked with 5% glutaraldehyde (Electron Microscopy Sciences, Ft. Washington, PA) and centrifuged to remove bubbles.
To assess reliability, 10 healthy women (ages 23–45 years; mean: 35 years) with no history of breast diseases were scanned twice within a 24-h period. To assess the sensitivity of APT CEST to treatment, three female patients (ages: 33, 39, and 57) with pathologically proven locally advanced breast cancer underwent an MRI examination before and after receiving neoadjuvant chemotherapy (days after therapy: nine, nine, and six). All three patients had HER2 negative tumors and received doxorubicin and cyclophosphamide every two weeks for four cycles followed by 12 weekly cycles of paclitaxel. After therapy, patients underwent either mastectomy or conservation therapy. Response to therapy was assessed based on the pathology report from surgery. The study was approved by our local institutional review board, and signed consent was obtained prior to all examinations.
All data were acquired on a 3.0T Achieva MR scanner equipped with a dual-channel body coil for RF transmission (Philips Healthcare, Best, The Netherlands). A MammoTrak table with a dedicated 16-channel receive double-breast coil was used for phantom and breast imaging.
CEST imaging data were acquired with a 3D single-shot turbo field echo (TFE) with echo spacing/echo time/α = 7.1 ms/3.5 ms/10°, SENSE parallel imaging (acceleration factor of 2 in both right-left and anterior-posterior directions), and a 1-3-3-1 binomial pulse for fat suppression. For phantom imaging, the parameters of the saturation train were varied to examine the effects of varying individual pulse amplitude, pulse duration, and total saturation time. Specifically, the peak pulse amplitude was swept between 0.25 and 1.25 μT in increments of 0.25 μT, whereas the number of pulse elements and duty cycle were held constant at 35 and 90%, respectively. In addition, the effects of duty cycle were examined at values of 50%, 60%, 70%, and 80% with the number of pulses per pulse repetition time (35) and pulse duration (25 ms) held constant. Finally, the total duration of saturation was examined at 440, 600, and 907 ms, both while holding the peak pulse amplitude (0.5 μT) and pulse duration (25 ms) constant. Each phantom scan was performed twice to ensure repeatability of the measures.
Human subject and patient scans included 10 unilateral sagittal slices acquired (field of view = 192 × 192 × 50 mm3) with a resolution of 2.5 × 2.5 × 5.0 mm3. For the CEST preparation, a 35-element pulse-train was applied with the following parameters: amplitude of (excitation) radiofrequency field = 0.5 μT, duration = 25 ms, and a 91% duty cycle (i.e., a 2.5-ms gap between pulses). The total pulse-train duration was 962.5 ms and, ignoring the T1 relaxation during the 2.5 ms gaps between pulses, the effective pulse-train flip angle was 184.87°. The offset frequency (Δω with respect to water) of irradiation was swept between ±6 ppm in 0.3 ppm increments with a total scan time of 6 min 42 s. CEST imaging performed on the BSA phantom was identical to that used on the human subjects.
For each subject, slice planning was based on anatomical images preformed at the beginning of each scan session. The anatomical images included four image stacks covering both breasts (sagittal right, sagittal left, bilateral axial, and coronal) with in-plane resolution of 1.37 × 1.37 mm2, slice thickness of 7 mm, a gap of 2 mm, resulting in a total scan time of 56 s. This survey was used for placement of the unilateral CEST images on both healthy controls and patients. For the healthy controls, the slice stack was centered on the left breast with an attempt to select the same field of view from both scan sessions. For the patient studies, the slice stack for the CEST sequence was determined by attempting to maximize breast coverage, while including the entire tumor.
The patient data were acquired as part of an ongoing clinical trial (23) and made use of dynamic contrast-enhanced data to segment the lesion volume prior to CEST analysis. The DCE acquisition employed a 3D spoiled gradient echo sequence with pulse repetition time = 8.1 ms, echo time = 4.6 ms, and flip angle = 20° using SENSE parallel imaging (acceleration factor = 2). The acquisition matrix was 192 × 192 × 20 over a sagittal 256 × 256 × 100 mm3 field of view. Each 20-slice set was collected in 16 s at 25 time points and 0.1 mmol/kg of gadopentetate dimeglumine (Gd-DTPA; Magnevist, Wayne, NJ) was injected at 2 mL/s after the third dynamic scan.
A semiautomated process was employed to define fibroglandular (FG) tissue and tumor masks for both healthy controls and patient data sets. The FG mask was defined by accounting for three separate aspects of the images. First, a threshold-binning procedure eliminated background noise and voxels affected by partial-volume averaging. Next, the skin and chest wall were manually excluded from the region of interest of FG tissue. Finally, the six slices that were best matched for each subject were manually chosen to ensure slice consistency between repeated scans.
Patient data sets were masked for tumors as well as FG tissue by incorporating an additional mask based on the contrast agent enhancement. This enhancement mask was defined as voxels with at least 50% enhancement following the administration of the contrast agent. Individual FG regions of interest were combined with the threshold-based enhancement maps to yield a mask of the primary tumor.
Each slice as a function of offset frequency, S(Δω), was nonrigidly coregistered using diffeomorphic demons with regularization criteria as implemented in Medical Image Processing, Analysis, and Visualization. The registered data were then normalized by acquisitions at 10 ppm up and downfield from water. The normalized data were fit to a single Lorentzian and the minima of the fit were used as the center water frequency. Data were shifted to align the minima from each voxel to Δω = 0 to account for amplitude of static (polarizing) field inhomogeneities (24). The shifted CEST spectra were reported on a voxel-by-voxel basis, so-called CEST z-spectra (25) and the magnitude of the CEST effect at the amide proton resonance (at 3.5 ppm) was calculated as follows: (1) difference spectra were generated as the difference between the fitted Lorentzian and the acquired CEST z-spectra, and 2) the difference spectra were integrated around the 3.5 ppm resonance from 3.1 to 4.1 ppm for each voxel and slice and termed APTresidual. Using the Lorentzian as a benchmark for comparison of the CEST effect accounts for spurious asymmetric magnetization transfer effects, lipid contamination, and baseline noise, all of which can confound the conventional asymmetry calculation in anatomies outside the brain (14). The FG tissue was masked as previously described for quantitative comparisons and reproducibility.
To assess reproducibility, we followed the outline provided by Galbraith et al. (26). While details can be found in their publication, we briefly summarize the relevant procedures. For each healthy control, the difference between repeated measurements for APTresidual, d, was calculated and tested for normality using the Shapiro–Wilk test. For all normally distributed data, a Student's t-test was used to test for significant differences between repeated measurements. A Kendall's tau test was performed to ensure that the measurement error was independent of the mean. The following statistical measurements of reproducibility were then computed:
1The mean squared difference, dsd:
The dsd is then be used to calculate the 95% confidence interval:
The confidence interval value indicates the change in a group of n patients greater than this value would be significant at the 5% level, for which 2.26 is the critical value.
2The within-subject standard deviation (wSD) was calculated as:
3The within-subject coefficient of variation (wCV) was calculated by dividing the wSD by the overall parameter mean. For transformed data, the wCV is calculated as:
4The repeatability, r, for each parameter was calculated via 2.77·wSD. The repeatability value indicates that the difference between the two measurements will be less than this figure for 95% of observation pairs.
Statistical analyses were performed using Excel (Microsoft, Redmond, Washington) and the statistical toolbox in MATLAB 2010b (The MathWorks, Natick, MA).
For the patient data, we merely assessed the change in the mean and standard deviation of the APTresidual distribution measured at each time point. These values were then separated into either the responding patients and or the nonresponding patients based on whether or not there was residual disease found at the time of surgery.
The optimal RF irradiation parameters were determined experimentally by examining the CEST z-spectra of a BSA phantom with a T1 equal to 1.28 s (27) and T2 equal to 55 ms, similar to that of breast FG tissue (28). Figure 1 displays the CEST z-spectra (solid lines) as well as the residuals of the Lorentzian fit (dashed lines) as a function of saturation pulse offset with errorbars indicating the standard deviation of the repeated scans. Of particular interest was the amplitude of the residuals around the amide resonance (3–4 ppm). To assess the effect of saturation pulse amplitude on the APT, it was varied from 0.25 to 1.25 μT, with results shown in panel a. Based on these results, a saturation power of 0.50 μT was chosen due to the exceptional APT effect with limited contamination from direct saturation. Panel b displays results of varying the duty cycle, demonstrating the small impact of this parameter on the observed APT effect. Panel c demonstrates the results of changing the total saturation time among 440, 600, and 907 ms while maintaining a constant pulse amplitude and duration. While the lineshape and amplitude of the residual spectra are strikingly similar for the three pulse-train lengths, it is known that the spectral resolution suffers with shorter pulse-train lengths, although this is not apparent in Figure 1c. Based on these results, we chose to use the following pulse-train for the breast studies: total duration of 962 ms, amplitude of (excitation) radiofrequency field = 0.5 μT, and pulse duration = 25 ms.
Representative 3T CEST MRI results from a healthy control are shown in Figure 2. Panel a displays CEST data as a function of offset frequency including 40 equally spaced saturation offsets ranging from +6 to −6 ppm. It is important to note that these CEST images demonstrate adequate fat suppression. Representative test–retest CEST z-spectra and difference spectra (between the Lorentzian fit and the acquired CEST data) for masked FG tissue are shown in panel b. The gray and black circles indicate the normalized CEST z-spectra from repeated scans of the same individual and correspond to the right y-axis. The difference spectra between the Lorentzian fit and the CEST z-spectra for each region of interest are also plotted as dashed lines and correspond to the left y-axis. The corresponding APTresidual maps are overlaid on the anatomical scan of the same healthy control for scan 1 (panel c) and scan 2 (panel d), visually demonstrating the reproducibility of this technique.
To qualitatively depict the results of our reproducibility study, Figure 3 shows the APTresidual maps for the subject with the highest correspondence, |d| = 0.08%, in panels a and b. Examples of average correspondence, |d| = 0.60%, are found in panels c and d, and the APTresidual maps for the subject with the worst reproducibility results, |d| = 1.71%, are shown in panels e and f. Comprehensive results from the reproducibility study can be found in Figure 4 in which the APTresidual values for each voxel in masked regions of interest are plotted for each data set along with the mean and standard deviation values, denoted by solid and dashed lines, respectively.
As the results of the Shapiro–Wilk test ensured normality of the data, a Student's t-test was used to test for significant differences between repeated measurements. The mean difference for all subjects (0.16%) was not significantly different from zero, and based on the results of the Kendall's tau test (P = 1.00), the individual difference values were not dependent on the average APTresidual value. The 95% confidence interval limits were ±0.70% (α = 0.05) and the repeatability of the mean values was 1.91. A summary of these data is found in Figure 5 with individual differences (APTresidual,1 − APTresidual,2) plotted against the mean value for each subject.
Patient 1 had a complete response (i.e., no residual tumor), Patient 2 had a partial response (i.e., 76% reduction in tumor size), and Patient 3 had progressive disease that included metastases to the brain and, therefore, did not go to surgery. Patient data are summarized in Table 1. These patients underwent MRI examination at 3 T before and after a single cycle of neoadjuvant chemotherapy, and the results are displayed in Figure 6. The top row shows data acquired prior to one cycle of neoadjuvant chemotherapy (NAC) therapy, and the bottom row shows that after one cycle of NAC therapy. Figure 6a shows data for a responder with anatomical T1-weighted images shown on the left and APTresidual maps on the right. Similar results are shown for a nonresponder in Figure 6b. The APTresidual maps are overlaid on the CEST image without saturation. The APTresidual,1 values were (mean ± standard deviation) 4.86 ± 0.15%, 2.92 ± 1.74%, and 3.32 ± 1.3% for Patients 1–3, respectively. After one cycle of treatment, the APTresidual,2 values were 3.5 ± 1.59%, 4.63 ± 0.19%, 5.93 ± 0.22% for Patients 1–3, respectively. These quantitative MRI results are summarized in Table 1.
Table 1. Patient Information and Quantitative MRI Results
Tumor stage nodal stage
Estrogen receptor status
Days between 1st treatment and 2nd MRI
APTresidual,1 (%)/ pretreatment size (cm)
APTresidual,2 (%)/ excised size (cm)
% Change in APTresidual
Serial biopsies can reveal biochemical properties but are invasive and subject to sampling error, producing misleading results. A need exists for a noninvasive, sensitive, and specific imaging biomarker for detecting, characterizing, and monitoring tumor progression and therapeutic response. The presented assessment of the variability of breast CEST MR imaging at 3 T in healthy volunteers suggests confidence for further application of this technique to breast pathologies. This study presents CEST MRI of three patients with pathologically proven locally advanced breast cancer before and after a single cycle of neoadjuvant chemotherapy. The measured APTresidual is hypothesized to correspond to amide protons associated with the backbone and flexible side chains of the mobile proteins resonating between 3.1 and 4.1 ppm. This measure is assumed to reflect the cellular protein and peptide content. An increase in this APTresidual metric was found in the patient with progressive disease as indicated in Figure 6b, potentially reflecting the continued cellular proliferation. The patient exhibiting complete response to the neoadjuvant regimen, Figure 6a, demonstrated a decrease in the measured APTresidual. It is important to note that the size of the tumors did not appear to change substantially following one cycle of neoadjuvant chemotherapy based on the underlying anatomical image. These results indicate that CEST MRI of the amide protons is potentially sensitive to microstructural molecular changes that occur prior to macroscopic changes in gross morphology.
To the best of our knowledge, only a recent Rapid Communication has reported application of CEST MRI to the breast (29). In this previous feasibility study, six patients with invasive ductal carcinomas underwent dynamic contrast-enhanced and CEST MRI at 3 T. The CEST effects observed between 1.2 and 1.8 ppm down field from water revealed the most significant differences between FG tissue and tumors. Although the authors did not offer interpretation, it is reasonable to hypothesize that these detected differences were most likely due to protons associated with exchangeable hydroxyl (OH) protons of glycosaminoglycans, known to be elevated in breast tumor tissue (30, 31). Although they report higher spatial resolution than our presented study, the spectral resolution of their CEST data was >0.6 ppm, potentially complicating the data analysis and subsequent interpretation. The presented research builds on this early effort, both of which point to the possible utility of CEST MRI in the context of breast cancer.
The observed CEST z-spectrum is the result of a complex interaction of imaging parameters as well as the underlying tissue microenvironment. For example, the CEST z-spectrum is known to be dependent on the chosen experimental parameters, particularly the RF saturation characteristics (32, 33). Herein, we experimentally optimized the CEST preparation parameters of our pulsed train saturation including pulse amplitude, pulse duration, and total saturation time on a tissue-like phantom composed of 15% cross-linked BSA. CEST z-spectra, shown in Figure 1a, indicate that a pulse amplitude of 0.25 μT does not elicit an APT effect, but when increased to 0.5 μT, examination of the APTresidual (dashed lines) effect becomes appreciably visible. A further increase in the saturation amplitude does slightly increase the magnitude of the residual spectrum. However, with increasing amplitude comes the penalty of long scan times to mitigate the impact of increased power deposition. Consequently, the specific absorption rate (SAR) was maintained at less than 2% with a saturation pulse amplitude of 0.5 μT while it was increased to 3% at 0.75 μT. As motion is a concern in breast imaging, we chose to use the pulse amplitude producing sufficient APT effects but minimized scan time. In the limit that the pulse duration is long, our results demonstrate a negligible dependence on the individual pulse duration. While not an exhaustive study, these phantom experiments served to establish optimal RF saturation parameters with the consideration of hardware and timing limitations on the human scanner. We concluded for this study that a low pulse amplitude played for a sufficiently long duration results in a favorable amide proton CEST sensitivity.
CEST imaging is potentially sensitive to microstructural molecular changes that occur prior to macroscopic changes in gross morphology and traditional contrast mechanisms. In particular, the magnitude of the CEST effect depends on the interaction between relatively mobile macromolecules, such as those associated with the amides, and bulk water. The CEST saturation spectrum is sensitive to solute/water proton concentration in exchange at specific resonance frequencies, such as that for amide protons, and the modulation of this exchange by the hydrogen ion concentration (pH). CEST-derived metrics of APT have been applied to tumors in animals (9, 10) demonstrating an increase between healthy tissue and that of tumor. Salhotra et al. (9) attributed this to an increase in cellular proliferation and subsequent accumulation of defective proteins. Application in human brain cancer has further indicated the sensitivity of CEST MRI, particularly APT, for the delineation of neoplasia in vivo (5, 10, 34, 35). Treatment effects have been explored by Zhou et al. using a radiation necrosis model in rats (10), demonstrating the ability to detect viable malignancy from radiation necrosis and predicting tumor response to targeted radiation therapy. This study expanded the application of CEST MRI to examine cancer treatment effects, indicating the molecular sensitivity of this technique to the increased cytosolic concentration of proteins and peptides.
In the breast cancer patients, the observed changes in APTresidual could be attributed to changes in the concentration of proteins and peptides or in the amide proton exchange rates, influenced by the local pH, and perhaps even T1 (6). Existing literature indicates a negligible increase in tissue pH in tumors (36), leaving the concentration of amide protons to dominate the APTresidual measurement. A practical drawback in CEST MRI is that the observed spectral asymmetry is exquisitely sensitive to the underlying MT asymmetry (37) as well as magnetic field inhomogeneities. As presented in this study, it is possible to fit the direct saturation contribution of the CEST spectra to a single Lorentzian lineshape (38) to correct for the deviation of the point of maximum saturation from the water on-resonance frequency. Furthermore, integration of the deviation of the data from this Lorentzian lineshape to examine saturation effects can eliminate the assumption of conventional magnetization transfer symmetry. Finally, acquisition techniques are known influence the resulting spectra (39, 40). The measured CEST effect relies on a complex interaction of many variables, and it is apparent that standardization of pulse sequences, acquisition parameters, and data processing techniques are essential for interpretation of results moving forward.
In the presented study, reliable APTresidual maps of healthy FG tissue were produced with adequate fat suppression. The metric of APTresidual, similar to that suggested by Jones et al. (41) is used instead of the conventional asymmetry measure [APTasymmetry; Ref.6] to account for spurious asymmetric MT effects, lipid contamination, and low signal-to-noise. Repeatability results indicate that a change in APTresidual larger than ±1.91% for an individual, or ±0.70% for a group of 10 patients would be statistically significant (α = 0.05). Despite this promising reproducibility, this study also had limitations. It is known that the menstrual cycle status affects the composition of the FG tissue (42), potentially confounding APTresidual measures. Additionally, despite the observed decrease in mean APTresidual in tumors that either partially or completely responded to the entire neoadjuvant chemotherapy regimen, it is important to note that these measures were obtained after one cycle of treatment. Therefore, it is possible that the observed trends in APTresidual represent a transient effect occurring only early in treatment. A larger patient population is required for a more comprehensive analysis of APTresidual measures in relation to neoadjuvant chemotherapy. As the imaging protocol implemented in this study was optimized for detection and quantification of the APT effect, a detailed optimization of strategies for detection of other metabolites known to exist in breast tumors, particularly the protons associated with the hydroxyl groups of glycosaminoglycans (30, 31) is ongoing.
In summary, we were able to establish the reproducibility of APT imaging of the breast and presented initial data on the ability of the technique to assess the early response of breast cancer to neoadjuvant chemotherapy. Future efforts will include studying the effects of age and stage of menstrual cycle (as potential sources of changes in breast biochemistry) on APT values, and incorporating the technique in an ongoing longitudinal, multiparametric study of treatment assessment in breast cancer. In addition, the CEST z-spectra will be evaluated for hydroxyl groups associated with glycosaminoglycans. The preliminary patient response data presented in this contribution indicates that these latter goals are reasonable.
The authors offer our sincerest thanks to the women who participated in our study. Seth Smith is partially supported by NIBIB. VICTR supports Adrienne Dula. The authors thank Ms. Donna Butler, Ms. Leslie McIntosh, and Mr. David Pennell for expert technical assistance. They thank our referring physicians: Dr. Vandana G. Abramson, Dr. Ana M. Grau, Dr. Mark C. Kelley, Dr. Ingrid A. Mayer, Dr. Julie A. Means-Powell, and Dr. Ingrid M. Meszoely. Dr. Rick Abramson, M.D., provided many informative insights.