Contradiction between amide‐CEST signal and pH in breast cancer explained with metabolic MRI

Purpose Metabolic MRI is a noninvasive technique that can give new insights into understanding cancer metabolism and finding biomarkers to evaluate or monitor treatment plans. Using this technique, a previous study has shown an increase in pH during neoadjuvant chemotherapy (NAC) treatment, while recent observation in a different study showed a reduced amide proton transfer (APT) signal during NAC treatment (negative relation). These findings are counterintuitive, given the known intrinsic positive relation of APT signal to pH. Methods In this study we combined APT MRI and 31P‐MRSI measurements to unravel the relation between the APT signal and pH in breast cancer. Twenty‐two breast cancer patients were scanned with a 7 T MRI before and after the first cycle of NAC treatment. pH was determined by the chemical shift of inorganic phosphate (Pi). Results While APT signals have a positive relation to pH and amide content, we observed a direct negative linear correlation between APT signals and pH in breast tumors in vivo. Conclusions As differentiation of cancer stages was confirmed by observation of a linear correlation between cell proliferation marker PE/Pi (phosphoethanolamine over inorganic phosphate) and pH in the tumor, our data demonstrates that the concentration of mobile proteins likely supersedes the contribution of the exchange rate to the APT signal.


| BACKGROUND
Metabolism in cancer is widely investigated by the use of magnetic resonance imaging (MRI). Metabolic MRI is a noninvasive technique that can give new insights into understanding cancer metabolism and potentially provide biomarkers to evaluate or monitor treatment plans. Among the different techniques, a method based on chemical exchange saturation transfer (CEST) 1 has attracted great interest recently. [2][3][4][5][6][7][8] This MRI method is a powerful and sensitive technique in which low concentration solutes can be visualized through the water signal. The contrast depends on the exchange rate of saturated mobile protons to the bulk water resonance. This enables indirect imaging of endogenous molecules containing these mobile protons, such as amides (proteins and peptides), a process which is called amide proton transfer (APT). Several studies have explored the use of different CEST approaches for treatment monitoring in cancer patients such as chemotherapy, [9][10][11][12][13][14][15][16] radiation therapy, 17 oncolytic virus therapy, 18 radiosurgery 19 and antibiotic treatment. 20 In theory, 21 the measured amide signal is primarily related to the concentration of mobile amide protons, the exchange rate (dependent on pH), duration of saturation pulse, and T 1 relaxation of water. The chemical exchange rate of amide protons is base-catalyzed. 6 Therefore, an increase of pH results in an increase of the exchange rate and consequently causes an increase in APT signal.
All healthy cellular functioning highly depends on a strict acid-base balance (pH homeostasis). This delicate balance is influenced by many metabolic processes such as proton production, proton transportations, chemical buffering, and vascular removal of waste products. Malignant cells show a pronounced increase in metabolic processes resulting in excessive production of protons. 22 To adapt to this intracellular acidity, the number and functions of proton-exporting mechanisms are increased. 22 This adaptation keeps intracellular pH (pHi) at normal or slightly alkaline levels, which favors protein synthesis and mitosis, while the extracellular pH (pHe) decreases due to the acidity from the exported protons. 23 There are many potential pH regulators involved in this process including: Na + /HCO 3 − co-transporters, Na + /H + exchangers, monocarboxylate transporters, the vacuolar ATPase, carbonic anhydrase, anion exchangers, the Cl − /HCO 3 − exchangers, and ATP synthase. 24,25 Phosphorous magnetic resonance spectroscopy ( 31 P-MRSI) provides a noninvasive technique to measure the pHi of tumor cells. The resonant frequency of inorganic phosphate (Pi) is pH-dependent 26 and, in many tissues, the majority of the Pi resonance is intracellular (~85%). 27 As such, pHi can be measured by 31 P-MRSI by calculating the chemical shift difference between Pi and a pH-independent reference peak.
Moreover, with 31 P-MRSI, cell proliferation biomarkers, the phoshomonoesters (PME) phosphoethanolamine (PE) and phosphocholine (PC) can be detected, providing a direct indication of cytotoxicity. 28,29 Also, the phosphodiesters (PDE) glycerophosphocholine (GPC) and glycerophosphoethanolamine (GPE) can be measured. The PME/PDE ratio is related to mitotic count and therefore indirectly to the tumor grade. 30 This technique has been used in a previous study 31 to show the feasibility of monitoring membrane metabolism during NAC treatment.
The authors also showed that the pH was increased by 0.19 units after the completion of NAC treatment. In another study, 32 reduced APT signal during NAC treatment was observed. These findings are counterintuitive, if one interprets the data solely in terms of the pH dependence of the chemical exchange rate of amide protons, which is base-catalyzed. Therefore, an increase in pH results in an increased exchange rate, causing an increase in APT signal.
In this study, we combined both APT-MRI and 31 P-MRSI in breast cancer patients to better comprehend the relation between APT signal and pH. APT-MRI and 31 P-MRSI were acquired in breast cancer patients receiving NAC treatment before and after the first cycle of NAC using 7 T MRI.

| Subjects
This MRI study was performed in accordance with the guidelines of the University Medical Center Utrecht ethics committee (trialregister.nl: NTR4980). Twenty-two breast cancer patients (aged 36-64 years, mean 47 years) gave informed consent to participate in this study. The patients were selected for being treated with NAC, and were examined before and after the first cycle of NAC (at approximately three week intervals). Table 1 summarizes the demographics and tumor characteristics of these patients.

| Acquisition
All patients were scanned in a prone position on a 7 T MR system (Philips, Best, The Netherlands). CEST was acquired using a 26-channel bilateral breast 1 H transceiver coil (MR Coils, Zaltbommel, The Netherlands) and the 31 P-MRSI with a home-built two-channel unilateral 1 H/ 31 P dual-tuned transceiver coil. Therefore, the patient was repositioned between the two measurements. Third order image-based B 0 shimming was performed with least square error optimization using a 3D B 0 map followed by manual segmentation of the breasts. 33

| Apt-MRI
For APT-MRI, 33 frequency offsets were acquired unevenly distributed over the frequencies from −1 ppm to 33 ppm relative to the water resonance; more offsets were obtained around the frequency of the amide peak (3.5 ppm) and the water peak (0.0 ppm) for better fitting of these resonances. The frequency offsets associated with the nuclear Overhauser effect (NOE) were not included due to signal distortions by unsuppressed lipid resonances. A saturation train of four seconds (20 sinc-Gauss RF pulses, pulse duration = 100 ms, inter-pulse delay = 100 ms, peak amplitude B 1 ≈ 2 μT, duty cycle of 50%, average nominal B 1 ≈ 0.9 μT) was followed by a gradient-echo readout, 34

| Data analysis and pH measurement
Image processing and data analysis of the CEST data were performed with MATLAB 2014b (MathWorks, Natick, MA, USA). B 0 correction was applied using the WASSR method. 36 APT maps were calculated using a three-pool Lorentzian model 37 (free water pool, APT and MT) with the AC, adriamycin and cyclophosphamide; ER, estrogen receptor; FEC, 5-fluorouracil, epirubicin and cyclophosphamide; HER2/neu, human epidermal growth factor receptor 2; PR, progesterone receptor; TNM stage, classification of malignant tumors (tumor, nodes, metastasis).
Levenberg-Marquardt algorithm (fitting parameters can be found in Table 2) using the amplitude of the fit. To determine the mean APT signal in the entire tumor volume, a region of interest (ROI) was drawn on a CEST image acquired at 5.4 ppm downfield from the water resonance.
The hypothesis, assuming the chemical exchange is dominantly base-catalyzed and that it is directly related to the APT signal, was calculated 6 , and pK w = 11.2. 6,21 This relationship was used to compare it with the measured data.
All MRSI data were analyzed using IDL 6. aligned to α-ATP at 7.56 ppm as α-ATP is the most insensitive to pH of all peaks with high signal-to-noise ratio. pH values were calculated using the following form of the Henderson-Hasselbalch equation 39 : where pK A = 6.75 is the dissociation constant of Pi, δ HA = 3.27 is the chemical shift of the protonated form of Pi, δ A = 5.69 is the chemical shift of the nonprotonated form of Pi, and δ is the difference in chemical shift frequency between the Pi peak and the reference α-ATP peak, measured in parts per million (ppm). 31 P-MRSI is known to be used to image metabolites involved in the disturbed anabolism and catabolism of the cell membrane in breast cancer.
Metabolite ratios such as PME/PDE ratio and phosphoethanolamine/inorganic phosphate (PE/Pi) ratio were related to pH and APT signal. For calculating the metabolic signal ratios for PME/PDE and PE/Pi, all spectra were frequency-aligned to PE at 6.83 ppm and quantified using a nonlinear least-squares algorithm (AMARES). 40

| Statistical analysis
Statistical analysis was performed in GraphPad Prism (GraphPad Software, San Diego, CA, USA). A linear regression was used to determine the relation between the measured APT signal, metabolic ratios and the pH before and after the first cycle of NAC treatment. The relation was considered statistically significant if P < 0.05. Relative changes of these parameters, as measured prior to and after the first NAC cycle, were quantified. Values less than the first quartile -1.5 x interquartile range, or greater than the third quartile +1.5 interquartile range, were determined as outliers (<[Q1-1.5 x IQR] or > [Q3 + 1.5 x IQR]).

| RESULTS
The mean age of the 22 patients was 47 years (33-61 years) and each patient completed all cycles of NAC treatment (four different regimes were used; Table 1). Assessment of hormone receptor status was performed on the pretreatment core biopsy; six patients had human epidermal growth factor receptor 2 (HER2) positive tumors and 16 patients had HER2 negative tumors. Ten tumors were estrogen receptor (ER) positive, of which seven were progesterone receptor (PR) positive. Eight tumors were triple negative.
An example of the analysis of the acquired data of a patient (patient 13 fromTable 1 before the start of NAC treatment) is shown in Figure 1. For analysis of 31 P-MRSI, one voxel containing the tumor was selected ( Figure 1A) and nine metabolites were fitted ( Figure 1B). The metabolic ratios of metabolites involved in membrane metabolism (PME/PDE and PE/Pi) and the pH (based on the chemical shift between Pi and α-ATP) were calculated. From the APT-MRI, the mean APT signal was determined inside the tumor volume ( Figure 1C) based on the three-pool Lorentzian fit.
We found a statistically significant correlation between APT-CEST and pH when combining all data from all patients ( Figure 2). This correlation was, however, in the opposite direction to that which the hypothesis proposed (dashed gray line); an increase of pH showed a decrease in APT-CEST signal. When splitting the group into before and after the first cycle of NAC treatment, the linear regressions were still in the opposite direction compared with the hypothesis, yet no longer statistically significant (Figure 3).
The PME/PDE ratio, known to be involved with membrane anabolism and catabolism, showed no significant correlation with pH ( Figure 4). A significant negative correlation between PE/Pi ratio and pH was found prior to NAC treatment. After the first cycle of NAC treatment, the correlation was no longer statistically significant.
, with k 0 = 26.8, k b = 3.4×10 6 , and pK w = 11.2 and normalized to the measured data. The linear regression is shown by the solid black line which is already statistically significant without correcting for known (Sun and Sorensen) pH effects (P < 0.05). Note that the measured data is contradicting the hypothesis; a decreasing APT-CEST with increasing pH opposed to the hypothesis; *, Statistically significant In this study we acquired a unique dataset at 7 T consisting of the APT-MRI and 31 P-MRSI of breast cancer patients receiving NAC treatment. This data enabled the investigation of the relation between APT and pH with a direct measure of pH through 31 P-MRSI, which has not been presented before. Based on the chemical shift between Pi and α-ATP from the 31 P-MRSI spectra, we calculated the pH of all patients before and after the first cycle of NAC treatment. We showed that there was a statistically significant correlation between the APT signal and pH and between PE/Pi and pHi before the start of NAC treatment. After the first cycle of NAC treatment, these correlations appeared to be no longer statistically significant.
The pHi (the Pi signal mostly originates from the cytosolic compartment 41 ) remained neutral to slightly alkaline, which tumors are known to be.
We showed that proliferation measured with PE/Pi is higher with lower pHi, which suggests increased cellularity at lower pHi. This would also suggest that there will be an increased concentration of proteins, resulting in higher CEST effects at lower pH, mainly due to the concentration of proteins and peptides. This was confirmed by the APT measurements during this study. However, this is in contradiction to the naïve interpretation that APT must be smaller at lower pH due to the lower base-catalyzed amide proton exchange rate, but hints at the finding that, for the APT signal, concentration effects outweigh direct pH effects. The measured APT signal depends on physiological parameters and sequence parameters.
First, the physiological parameters are discussed.
As mentioned before, APT signal is primarily related to the concentration of mobile amide protons, the amide proton exchange rate, and T 1 relaxation of water. A change in T 1 relaxation time due to therapy effects could influence the APT signal, as T 1 relaxation is positively related with APT signal. 42,43 However, the APT signal varies by a factor of two between tumors, and this cannot solely be assigned to changes in T 1 relaxation time, as this would mean differences in tumor T 1 of 50%, 44 which is unrealistic. Therefore, we assumed that the expected effect size of T 1 relaxation on the APT signal is much less in comparison with the concentration of mobile amide protons and the exchange rate.
In normal tissue, it is expected that the extracellular compartment contains a low concentration of mobile amides. Therefore, APT signal in normal tissue most likely originates from proteins and peptides in the intracellular space. In tumor tissue, however, the extracellular space could contain increased concentrations of mobile proteins and peptides as a result of accumulation of blood-borne proteins, such as albumin, due to perforated blood vessels. 45 Therefore, both intracellular and extracellular compartments are likely to contribute to APT signal originating from tumor tissue. Intracellular pH in tumor tissue did not change much (slightly alkaline, mean pH of 7.4) and so APT signal change, originating from the intracellular compartment, assuming that T 1 relaxation will have the least effect (see above), reflects a change in protein content. Extracellular pH in tumors is acidic and APT-MRI, originating from the extracellular compartment, reflects a change in protein content and pH. Only a small part of the pH (~15%) that is determined by 31 P-MRS originates from the extracellular compartment and therefore it is difficult to rule out parameters affecting the APT signal originating from the extracellular compartment. However, this data does suggest that the concentration of the mobile amide protons is the main contributor to the observed APT signal.
Considering the main influence of the concentration of the mobile amide protons, the dependency of the APT signal on the mobile amide protons is much stronger than its contra-related dependency to the exchange rate. A recent study in a rat model of brain metastasis determined that the proportion of APT signal originating from changes in protein concentration was approximately 66%, with the remaining 34% originating from changes in tumor pH. 46 It has been shown that NOE signals from aromatic protons, in a range of +1 to +5 ppm from water, affect the quantification of APT effects. 47 These aromatic NOE signals also originate from proteins, yet are insensitive to pH. If these signals decrease due to therapy effects, they could surpass the effect of the actual amides, resulting in this inverse pH dependency. It could also be that overlapping CEST signals from different exchanging sites influence the APT signal. It has been shown that fast-exchanging amines can have an inverse pH dependency in animals 48 and homogenates. 49 However, the nominal B 1 in this study was 2 μT (sinc-Gaussian pulses) with a duty cycle of 50%. This means that the average nominal B 1 is less than 0.9 μT. Also, the relative B 1 in the tumor area is approximately 50-60% (see below), resulting in approximately 0.5 μT average effective B 1 in the tumor. Considering this B 1 , the main contributor to the measured signal is APT (approximately 90%), 50 which makes the contribution of fast-exchanging protons highly unlikely. A more likely explanation for the decrease of APT with increasing pH would be the decreasing cellularity (PE/Pi) with pH. A low cellularity hints at a lowered concentration of proteins, which is a more probable explanation for the decreased APT signal.
A sequence-dependent parameter affecting APT signals is B 1 inhomogeneity. In this study, the CEST was optimized for signal generated by the slow-exchanging amide protons, and the optimal B 1 for detecting this exchange is approximately 1 μT. 51 The peak B 1 amplitude in this study was set to 2 μT to account for B 1 loss in the hardware setup. The bilateral breast coil setup consisted of two quadrature RF coils placed in front of the breasts, which led to a decrease in B 1 , varying from 60% in the front of the breast to 50% towards the pectoral muscle. This means that the APT signal coming from tumors located near the nipple could be higher than the APT signal coming from tumors located near the pectoral muscle. Also, the level of variance in B 1 and its effect on the APT signal cannot explain the level of observed APT signal changes, particularly when considering that the tumor location with respect to the RF coil is not expected to change substantially over one cycle of chemotherapy.
The overall calculated APT contrast could also be influenced by the degree of fat suppression. We used RF and gradient spoiling to reduce lipid artefacts. However, insufficient fat suppression in the tumor may have resulted in an underestimation of the CEST amplitude, 52 possibly affecting the change in APT signal.
Also, the zero-order approach used in this study is not exact, and different power levels will have different contributions of MT and water saturation, probably leading to different sized APT effects. However, all of the experiments were performed with the same parameters for the RF pulses and, therefore, this will not affect our conclusion.
We were able to perform 31 P-MRSI analysis for every patient. A linear relationship was found in this study between PE/Pi before NAC treatment and the pHi. PE is a key metabolite involved in the Kennedy pathway that produces phosphatidylserine in the major building block of cell membranes. 29 Pi is involved in many metabolic pathways, including energy transfer, protein activation, and carbon and amino acid metabolic processes. 53 Therefore, PE/Pi could be an indicator for cellularity. After the first cycle of NAC this correlation is not significant. NAC treatment has been shown to be most effective if a combination of anthracyclines and taxanes is used, causing damage to DNA and disrupting the pathways necessary to facilitate mitosis. 54 Therefore, it is most likely that PE, Pi and pH in the tumor are all affected by the chemotherapy. However, further research is required to understand the linear relationship between PE/Pi and pH.
In conclusion, in this study we have shown that APT-MRI and 31 P-MRSI provide complementary information about tumor metabolism in breast cancer patients. A linear correlation between APT signal and pH, and a linear correlation between PE/Pi and pH in the tumor were found in breast cancer patients before the start of NAC treatment. This correlation was opposite to the intrinsic relation between APT signal and pH, demonstrating that the concentration of mobile amide protons is the main contributor to the observed APT signal. In fact, when correcting for the known intrinsic relation of APT with pH, the observed concentration range of mobile amides will be even higher.