1 INTRODUCTION

Intracellular pH (pHi) is a valuable marker for cancer as it is known to be increased in tumors.1-4 In principle, high-resolution imaging of pHi is possible by means of chemical exchange saturation transfer (CEST).5-7 CEST is an emerging MRI technique that enables non-invasive detection of organic compounds present in low concentration in living tissue (e.g. metabolites8-17 or proteins18-20) with a comparable resolution to conventional MRI. In addition, CEST signals are intrinsically sensitive to pH due to their indirect detection via the abundant water signal and the underlying phenomenon of chemical exchange.18, 21 The most prominent endogenous CEST signals in vivo are the amide and guanidyl signals resonating at a frequency offset of approximately 3.5 and 2.0 ppm, respectively. The amide and guanidyl signals are primarily intracellular, as both can mainly be associated with mobile proteins/peptides18-20, 22, 23 within cells, whereas additionally, the guanidyl signal is also related to creatine.12, 13, 24, 25

To separate the pH-dependency from concomitant effects, such as concentration changes, several CEST-MRI-based approaches have already been presented.26-55 Most applications aim at the investigation of pH changes in stroke,31-33, 40, 45, 48, 54 but also applications in tumors have been presented.30, 34, 36, 50, 52, 55 However, state-of-the-art methods do not allow for simultaneous compensation of (1) concentration changes, along with (2) superimposing CEST signals, (3) the magnetization transfer contrast (MTC), and (4) spillover dilution (i.e. dilution of CEST signals depending on the water relaxation properties and other superimposing magnetization transfer signals),56, 57 which can all strongly vary within tumors.58 Therefore, the aim of this study was to develop a method for mapping pHi in tumors based on endogenous CEST signals. We hypothesize that this can be realized at a magnetic field strength of B0 = 9.4 T by using (1) a concentration-independent ratiometric approach26-28, 59 (i.e. the ratio of one CEST signal at different saturation amplitudes B1, CESTratio) in combination with (2) the spillover-corrected inverse metric (i.e. the relaxation-compensated magnetization transfer ratio, MTRRex56, 57, 60) and (3) a separate first-order polynomial and Lorentzian-fit analysis14, 24, 61 of the amide and guanidyl signals. This would allow assessing the potential of pHi as a biomarker for cancer diagnosis or treatment monitoring.

The presented method is based on the finding that, for amide and guanidyl protons, the dependency of the exchange rate kex on pH at a fixed temperature T (e.g. 37°C in vivo) is fully characterized by only one rate constant kc. In addition, we present a stable and robust approach to precisely calibrate kc in vitro. The calibration relies on the symmetric dependency of CEST signals (i.e. the MTRRex contrast) on pH, which is a fundamental insight of this study. Calibration of the pH dependency was accomplished using in vivo-like (i.e., comparatively small CEST signals and a large MTC) model suspensions from porcine brain lysates. Finally, we demonstrate that with prior knowledge of kc and R2s (i.e. the transversal relaxation rate of the CEST signal s) absolute pHi mapping can be achieved by determining kex from the CESTratio at different B1. Due to the different ranges of pH sensitivity of the amide and guanidyl signals, the two individual pHi maps were combined by a weighted approach to enable a stable pHi mapping over a broad range of physiological pH. The functionality was again evaluated under in vivo-like conditions using porcine brain lysates at various pH and tissue concentrations. Applicability of the presented method for examinations in vivo was demonstrated in tumor-bearing mice (n = 19).

2 THEORY 2.1 pH mapping using the amide or guanidyl signal The exchange rates kex of amide and guanidyl protons are dominantly base-catalyzed within the physiologically relevant pH range62: (1)with the base-catalyzed exchange rate constant kb and the dissociation constant of water KW. For a fixed temperature T (e.g. 37°C in vivo), the dependency of kex on pH is fully characterized by only one rate constant kc. Thus, mapping of absolute pH values is possible by measurement of kex and prior knowledge of the exchange rate characterizing constant kc: (2) 2.2 Calibration of kc The exchange rate characterizing constant kc can be determined experimentally by analyzing the CEST signal as a function of pH. Calculation of the isolated relaxation-compensated (i.e. spillover-corrected) CEST signal can be achieved by the inverse magnetization transfer ratio, MTRRex.56, 57, 60, 63 In steady-state (i.e., saturation duration , the longitudinal relaxation time of water) and in the large shift limit (LS, i.e. frequency offset of the CEST pool , leading to an angle between the effective field and the z axis; thus, ) MTRRex is given by: (3) Zlab is the acquired Z-spectrum, defined as the normalized water signal after pre-saturation at the frequency offset , and Zref is the Z-spectrum without the respective CEST signal. Here, the CEST signal (i.e. MTRRex) is either the amide or the guanidyl signal. MTRRex is the ratio of the exchange-dependent relaxation rate Rex and the longitudinal relaxation rate of water R1w. Moreover, Rex is given by: (4)with the relative concentration of the CEST pool fs, the labeling efficiency and the transversal relaxation rate of the CEST pool R2s. MTRRex as a function of kex increases until it reaches a maximum at64-66: (5) Remarkably, kex,max only depends on the applied B1 while being independent of the concentration fs, R1w and R2s (Figure 1A). Using Equation 1, the dependency of MTRRex on kex can be translated into a direct dependency on pH (Figure 1B). MTRRex as a function of pH is a symmetric function with position and with a defined full width at half maximum (FWHM) (detailed description is presented in the Supporting Information Section S2). Consequently, the exchange rate characterizing constant kc can be determined experimentally by finding pHmax in the measurement of model suspensions at various pH values: (6) MTRRex simulated for various B1 and R2s as a function of kex (A) and pH (B). (A) The position of the maximum kex,max (dashed line, Equation 5) depends only on the applied B1 while being independent of the concentration fs and R2s (dotted line). (B) Remarkably, translation of the kex-axis into pH-values leads to a symmetric dependency of MTRRex with a defined FWHM. By measurement of the peak position pHmax (dashed line, Equation 6), the exchange rate characterizing constant kc (Equation 1) can be determined. For the simulations values of R1w = 1 Hz, fs = 0.018 and kc = 53.2 µHz were used

The symmetric shape of the function MTRRex(pH) and the nearly constant FWHM under variation of B1 and R2s allows for a precise, stable and robust fitting procedure. It was already shown by Woessner et al64 that the maximal CEST signal yields insight into the exchange rate. Here, we extend this approach for a full characterization of the exchange process by acquisition of only one multi-pH MTRRex image at one B1. This approach can be understood as a novel alternative quantification method analog to “quantification of exchange rate using varying saturation power” (QUESP)37, 38 by performing quantification of exchange rate by variation of pH.

2.3 Measurement of kex by a ratiometric approach The acquisition of Z-spectra at two different RF amplitudes, B1,high and B1,low (with B1,high > B1,low), allows the calculation of a relaxation-compensated and in particular concentration-independent ratio that only depends on kex and R2s26-28, 59: (7) Here, again MTRRex is either the amide or the guanidyl signal. The CESTratio has a distinct dependency on pH and approaches 1 and for low and high pH values, respectively (Figure 2A). For a fixed R2s, measurement of the CESTratio allows experimentally determining kex by: (8) CESTratio simulated for various combinations of B1,high and B1,low (A), and kc (B) as a function of pH. (A) The CESTratio has a distinct dependency on pH and approaches 1 and for low and high pH values, respectively. (B) The pH sensitivity of the CESTratio (dashed lines, normalized to 1) also strongly depends on kc. The pH range with at least a pH sensitivity of 30% for either the amide or the guanidyl signal (gray shaded region) has a smooth transition at around 7.3. For the simulations values of kc = 53.2 µHz, R2s = 100 Hz (A), and B1,high = 1.4 µT, B1,low = 0.6 µT (B) were used

In order to correct for B1 field inhomogeneities, the CESTratio needs to be adjusted according to the local B1 field in each pixel. This can be achieved by correcting MTRRex(B1,high) and MTRRex(B1,low) in Equation (7) using the contrast-B1-correction method.67 For the contrast-B1 correction, MTRRex is acquired at multiple B1, fitted at the local B1 values (e.g. determined by the WASABI68 method), and calculated at the desired nominal B1 values. In this study, a fit model based on the MTRRex theory (Equations 3 and 4) was implemented (Supporting Information Section S4.4). This allows for determination of the CESTratio with data acquired at more than two B1,26, 53 which was essential in this study to increase the accuracy of kex (i.e. in the case of noisy data in vivo). Alternatively, B1 field inhomogeneities can also be corrected via multiplication of the nominal values of B1,high and B1,low in Equation (8) by the respective local relative B1 value in each pixel.

2.4 Combined pH mapping using the amide and guanidyl signals Separate analysis of the amide and guanidyl signals according to Section 2.1 to 2.3, 2.1 to 2.3 leads to an individual pH map for the amide and the guanidyl signal (i.e. pHamide and pHgua). These two separate pH maps can be combined pixel-wise into a single pH map (i.e. pHcombi) using a weighting based on the pH sensitivity of each CEST signal: (9)

The individual pH sensitivities ws for each CEST signal (s = amide or guanidyl) can be defined by the derivative of the CESTratio as a function of pH (Figure 2B, dashed lines, normalized derivative of Equation 7 with Equation 1).

For simplicity, in this manuscript only the case of a continuous wave (cw) pre-saturation is presented. Nonetheless, the theory can also be extended to the more general case of a pulsed pre-saturation, which is presented in the Supporting Information Section S1. This, in principle, enables the translation of the presented method to whole-body MR scanners and, thus, the application in humans.

3 METHODS 3.1 Model systems 3.1.1 Porcine brain lysates

Porcine brain lysates (i.e. pig brain homogenates) were prepared from mixed white and gray matter according to a previously established procedure.69 Samples were prepared at various pH and tissue concentrations. The pH series (pH of approximately 5.5, 6.0, 6.2, 6.4, 6.5, 6.6, 6.7, 6.8, 7.0, 7.2, 7.3, 7.4, 7.5, 7.6, and 8.0) was prepared at a constant tissue concentration of ctissue = 33.3%(w/v) and the concentration series (ctissue = 16.7, 22.2, 27.8, and 33.3%(w/v)) with a constant pH of around 7.0. More details about the preparation and a list of the precise final titrated pH values, which marginally differ from the target pH, are provided in the Supporting Information. To ensure the most in vivo-like characteristics (i.e. mitigation of the denaturation of proteins and metabolites), the samples were chilled on ice during preparation. All samples were prepared directly before MRI examination.

3.1.2 DLD1 xenografted nude mice

DLD1 xenografted nude mice (n = 19, female NMRI Nude, Janvier Labs, Le Genest-Saint-Isle, France) were examined with approval by the local regulatory authorities under G284-15. 2∙106/100 µL phosphate buffered saline (PBS) of the human colon cancer cell line DLD1 were injected subcutaneously into the right flank under isoflurane anesthesia. MRI measurements were done under sevoflurane anesthesia. During the examination, animals were heated to 37°C to maintain a constant body temperature by use of warm air to ensure a homogeneous heating without temperature gradients. Animal respiration was monitored using a breathing surface pad. A respiratory small animal cradle (SA instruments, Inc., NY, USA) served to suppress breathing-induced motion artifacts in the tumor region during MRI. Eye cream was applied during anesthesia to avoid eye drying. Tumors were investigated post-mortem and checked for necrosis by routine hematoxilyn/eosin staining.

3.2 CEST-MRI

All measurements were performed on a 9.4 T small animal MR scanner (BioSpec 94/20 USR, Bruker BioSpin MRI GmbH, Ettlingen, Germany) using either a single resonant coil for in vitro examinations (Bruker, two-channel transmit/receive 1H volume resonator, diameter 40 mm) or a double resonant coil for in vivo examinations (Bruker, one-channel transmit/receive 31P/1H volume resonator, diameter 40 mm). For CEST-MRI, a custom-built pulse sequence with a 2D RARE70 image readout was used. All measurements were stabilized at a temperature of 37.0 ± 0.1°C using the internal heating device and a rectal temperature sensor.

3.2.1 Image acquisition protocol

The acquisition protocol consisted of three CEST scans at different B1 and one WASABI68 scan for mapping of B0- and B1-field inhomogeneities. For normalization of the Z-spectra, two M0 images were acquired at −300 ppm at the beginning and end of each scan. M0 images were acquired after a relaxation interval of 20 and 12 s to ensure full relaxation of the water magnetization (i.e. recovery toward the equilibrium magnetization) for in vitro and in vivo examinations, respectively. Details about the WASABI scans are provided in the Supporting Information.

For the in vitro and in vivo examinations, the CEST scans were realized by a cw pre-saturation pulse (i.e. rectangular) of RF amplitude B1 = 0.6, 1.0 and 1.4 µT and a duration of tsat = 10 and 6 s, respectively. In the following, the parameters of the in vitro and in vivo examinations are presented in the manner in vitro (in vivo). Z-spectra were sampled at 92 (42) frequency offsets in unequal steps between 100 and –100 (10 and 0.1) ppm (the complete lists of frequency offsets are provided in the Supporting Information), resulting in an acquisition time of 16:47 (4:56) minutes per B1. Images were acquired using a single-shot 2D RARE acquisition with centric encoding, FOV = 30 × 30 (30 × 30) mm2, matrix = 60 × 60 (40 × 40), resolution = 0.5 × 0.5 (0.75 × 0.75) mm2, slice thickness = 2 (2) mm, partial Fourier (PF) = 1 (1) (i.e., no PF), RARE factor = 60 (40), TE = 4.628 (4.628) ms, TR = 281.126 (188.57) ms, acquisition bandwidth = 50 (50) kHz (i.e., 833.3 (1250) Hz/pixel). The total acquisition time for one pH CEST examination including one WASABI and the three CEST scans was 52:24 and 16:33 minutes for in vitro and in vivo examinations, respectively.

In addition, anatomical high-resolution 3D images were acquired using a T2w-RARE sequence (FOV = 30 × 30 × 16 mm3, matrix = 240 × 240 × 16, resolution = 0.125 × 0.125 × 1 mm3) with an acquisition time of 6:06 minutes.

3.2.2 Data processing

Data processing was performed using in-house developed software in MATLAB® (The MathWorks Inc., Natick, USA). Normalization of Z-spectra was realized by interpolating between the two M0 images, which were acquired at the beginning and end of each scan, to obtain an individual M0 image for each Msat(Δω) image. In order to correct for B0 inhomogeneities, the acquired Z-spectra were shifted along the frequency axis (i.e. Δω dimension) according to the B0 map obtained from the WASABI measurement. For the B0 correction a smoothing spline fit was used with a smoothing parameter of 0.999.

In order to extract the CEST signals from the background, Z-spectra were fitted pixel-wise using a first order polynomial and Lorentzian-fit model. For extraction of each CEST signal s (i.e. either the amide or the guanidyl signal) an individual first order polynomial and Lorentzian-fit was performed: (10)

Here, c0,s and c1,s are the coefficients of the first-order polynomial function (i.e. intercept and slope) and is a Lorentzian function. The amide and guanidyl CEST signals were fitted by inclusion of spectral data points only close to their resonance frequencies (approximately ±0.5 ppm). A detailed description of the fit model and the optimized fit parameters (i.e. fitted offsets, start values and borders for each CEST pool and B1, respectively) is provided in the Supporting Information.

The fitted Z-spectra were used for calculation of the MTRRex contrast as defined in Equation (3), with and .71 The MTRRex contrasts were corrected pixel-wise for B1 inhomogeneities using the contrast-B1-correction method67 with the fit model defined in Supporting Information Equation (S9). Again, a detailed description of the optimized fit parameters (i.e. start values and borders) is provided in the Supporting Information. For the calibration of kc (Section 2.2) the B1-corrected MTRRex signal was fitted to the symmetric function specified in the Supporting Information Section S2 in order to determine the peak position pHmax (Equation 6). For measurement of kex (Section 2.3), the CESTratio was calculated by the B1-corrected MTRRex contrasts at B1,high = 1.4 and B1,low = 0.6 µT.

With the calibrated value for kc at hand and a fixed R2s, the measured values of kex were used for a pixel-wise calculation of the two separate pH maps (i.e. pHamide and pHgua) according to Equation (2). For R2s, 100 Hz was used in agreement with the broad range of values found in literature.28, 41, 56, 60, 72, 73 The two individual pH maps were combined using Equation (9). For calculation of the weights ws, values of B1,high = 1.4 and B1,low = 0.6 µT, as well as the calibrated values for kc (Section 4.2) and the R2s from literature were used. Weights were calculated numerically from the derivative of Equation (7) and normalized to 1.

4 RESULTS 4.1 Calibration of kc for the amide and guanidyl signals

To characterize the pH dependency of the amide and guanidyl signals (i.e. by determining kc, Equation 1), a novel approach for a stable and robust calibration in vitro was developed. In a first step, the isolated CEST signal MTRRex (Equations 3 and 4) was simulated at various B1 (Figure 1). MTRRex as a function of kex reaches a maximum at kex,max, which only depends on B1 (Equation 5, Figure 1A, dashed line) while being independent on the concentration fs and R2s (Figure 1A, dotted line). The same trends were also present for MTRRex as a function of pH, with a maximum at pHmax only dependent on B1 (Figure 1B, dashed line) and independent of fs and R2s (Figure 1B, dotted line). Remarkably, as expected from a single pool described by only one kc, the dependency on pH was a symmetric function of defined FWHM (Supporting Information Section S2) enabling a stable and robust fitting and, thus, precise determination of pHmax. As demonstrated in the Theory Section 2.2, measurement of pHmax, in turn, allows the direct calculation of kc (Equation 6), and thus, the calibration of the pH dependency (Section 4.2).

To ensure a calibration of kc under in vivo-like conditions, porcine brain lysates were investigated. In the Z-spectra, typical signals comparable to in vivo examinations were present, with pH-dependent resonances from amide and guanidyl protons at around Δω = 3.6 and 2.0 ppm, respectively (Figure 3A), as well as, a broad concentration-dependent offset of the MTC (Figure 3B). Extraction of the amide and guanidyl signals from the background of concomitant magnetization transfer signals, using the presented first-order polynomial and Lorentzian-fit model (Equation 10), was stable over a broad range of pH (Figure 3C) and B1 (data not shown). For the calibration of kc, Z-spectra between pH 5.5 to 8.0 and B1 = 0.6 to 1.4 µT were included (Figure 4). The pH dependency of the isolated MTRRex signals from amide and guanidyl protons showed the expected symmetric shape in coherence with theory (Figure 1A,B). Determination of pHmax by fitting the symmetric function specified in the Supporting Information Section S2, led to an average kc of 3.2 ± 0.5 and 53.2 ± 3.7 µHz for the amide and guanidyl signal, respectively. For the kc values a slight B1 dependency was present, which however, was insignificant for the observed variation within the regions of interest (ROIs) (Figure 4C,D).

Single-voxel Z-spectra (B1 = 1 µT) of porcine brain lysates at various pH (A) and tissue concentrations (B). Clear resonances of the amide and guanidyl signals are present at 3.6 and 2.0 ppm, respectively. (C) Z-spectra at various pH with separate first-order polynomial (i.e. Zref, solid black lines) and Lorentzian-fits (i.e. Zlab, dashed black lines) of the amide and guanidyl signals. The fits are only illustrated within their respective range of offsets that were used for the separate fitting procedures

In vitro calibration of kc for amide (A, C) and guanidyl (B, D) protons in porcine brain lysates at various pH. (A, B) Experimental MTRRex values are in good coherence with the theoretical symmetric shape as a function of pH (cf Figure 1B). (C, D) Experimental determination of pHmax (Equation 6) by an individual fit for each B1, led to an average kc of 3.2 ± 0.5 and 53.2 ± 3.7 µHz for the amide and guanidyl signals, respectively. (A, B) All values are mean ROI values ± SD. (C, D) Error bars for kc are obtained from the 95% confidence interval of the fit 4.2 pH mapping in porcine brain lysates

With the calibrated kc for the amide and guanidyl signals at hand, absolute pH mapping was realized by determining kex (Equation 8) from the CESTratio at different B1 (Equation 7). To optimize the used B1, in a first step, the CESTratio as a function of pH was simulated for various combinations of B1,high and B1,low (Figure 2). The CESTratio approached 1 and for low and high pH values, respectively, allowing to increase the pH sensitivity of the CESTratio by maximizing the difference between B1,high and B1,low. Thus, the complete range of practicable B1 to reliably isolate the individual CEST signals over a broad range of pH values by the fitting procedure (i.e. 0.6–1.4 µT, Section 4.1) was used. In principle, also the range of pH sensitivity could be shifted along the pH scale by varying B1,high and B1,low at a constant ratio (Figure 2A, green and red line), which, however, was in conflict with the maximization of the pH sensitivity (i.e. measurable range of CESTratio) and the applicable range of B1 values. In addition, the pH sensitivity of the CESTratio strongly depends on kc (Figure 2B). Thus, in this study, the two individual pH signals from the amide and guanidyl protons (i.e. pHamide and pHgua) were combined by a weighted approach (Section 2.4) to ensure a good pH sensitivity over a broad range of physiological pH (Figure 2B, gray shaded region). The normalized derivative of the respective CESTratio as a function of pH (Figure 2, dashed lines) was found to serve as a reliable measure of the pH sensitivity for the weighted combination approach (Equation 9).

In porcine brain lysates, the experimental CESTratio(amide) and CESTratio(gua) showed the distinct dependency on pH within their respective range of pH sensitivity, in coherence with theory (Figure 5A,B, gray shaded regions, cf Figure 2). To experimentally evaluate the pH sensitivity, the two individual pH signals were correlated to the titrated pH of the porcine brain lysates (Figure 5C,D). For the pHamide and pHgua signal a good correlation (r = 0.6912, p = .129 and r = 0.9250, p < .0001) was observed in the range of pHtitrated 7.3–8.0 (Figure 5C) and 6.2–7.3 (Figure 5D), respectively, in coherence with theory (Figure 2B). Although the pHamide signal seemed to have a good correlation also between 5.0 and 6.8, the deviations (i.e. error bars in Figure 5C) were too large for a reliable measurement of the pH. The found ranges verified the application of the weighted approach to combine the two individual pH signals for absolute pH mapping over a broad range of physiological pH. The final combined pH signal showed a very good correlation (r = 0.9430, p < .0001) in the range of pH 6.2–8.0 (Figure 6C), while being independent of the tissue concentration (Figure 6D). Deviations were strongest at low pH < 6.2 due to the low pH sensitivity of the CESTratio at such low pH values (i.e. low B1-dispersion of the amide and guanidyl CEST signal, cf Figure 4A,B), preventing a reliable measurement of the pH in this range by the presented method. The final pH maps showed a homogeneous appearance with an average variation of around ± 0.2.

Experimental CESTratio (A, B) and pH (C, D) from CEST-MRI in porcine brain lysates of the amide (A, C) and guanidyl (B, D) signals. (A, B) The CESTratio (B1,high = 1.4 µT, B1,low = 0.6 µT) for amide (A) and guanidyl (B) showed the distinct dependency on pH (i.e. titrated pH measured with a pH meter, pHtitrated) in coherence with theory (i.e. simulated CESTratio, black lines) within the range of their respective pH sensitivity (gray shaded regions, cf Figure 2). For the simulations values of kc = 3.2 µHz (A) and 53.2 µHz (B), R2s = 100 Hz, and B1,high = 1.4 µT, B1,low = 0.6 µT were used. (C, D) Correlation of experimental pHamide (C) and pHgua (D) values with pHtitrated. For the pHamide and pHgua signals a good correlation (r = 0.6912, p = .129 and r = 0.9250, ptitrated 7.3–8.0 and 6.2–7.3, respectively (gray shaded regions), in good coherence with theory (cf Figure 2B). Deviations were strongest outside of these ranges, because the pH sensitivity of the CESTratio depends on the change of the respective CEST signal with varying B1 (i.e. the B1-dispersion at each pH, cf Figure 4A,B), allowing a reliable measurement of pHamide and pHgua only in the aforementioned ranges. Calculation of the CESTratio and pH maps was performed pixel-wise. All values are mean ROI values ± SD Final pH maps from CEST-MRI (i.e., pHcombi) of porcine brain lysates at various pH values (A, C) and various tissue concentrations (B, D). A very good correlation (i.e., r = 0.9430, pCESTratio at such low pH values (i.e. low B1-dispersion of the amide and guanidyl CEST signal, cf Figure 4A,B). (C) The mean pH values shown in (A) are illustrated in black. (C, D) All values are mean ROI values ± SD 4.3 Application to tumor-bearing mice

To demonstrate the applicability of the presented method for in vivo measurements, tumor-bearing mice were investigated (n = 19). A tsat of 6 s was sufficient to reach the steady-state in tumorous tissue required for quantitative MTRRex analysis (Equations 3 and 4, Supporting Information Figure S1). The presented first-order polynomial and Lorentzian-fit model (Equation 10) also enabled in vivo a stable extraction of the amide and guanidyl signals from the background of concomitant magnetization transfer signals (Figure 7D). In the tumor lesions, physiologically meaningful pHi values of around 7.2 indep