Introduction

Chemical exchange saturation transfer (CEST) MRI enables to obtain images of endogenous cellular components or exogenous agents that contain exchangeable protons. CEST effects are detected indirectly through the reduction of the water signal following selective saturation of exchangeable protons on the CEST agent [1] (for reviews, see, e.g. [2, 3]). With respect to imaging of cancer, CEST has been shown to provide information of high clinical relevance spanning the diagnosis of tumors [3, 4], tumor grading [4, 5], and the assessment of therapy response after radiation treatment [6]. CEST became an emerging interest in the field of breast cancer diagnosis, revealing endogenous CEST contrasts in breast tumors [7,8,9,10].

The use of glucose has been proposed as a new molecular imaging approach for diagnosing tumors given its high sensitivity at the molecular level and the known enhanced glucose uptake by tumors [11]. However, its fast conversion to lactic acid prevents its efficient imaging. Several excipients and glucose analogs were tested as tumor-detecting agents in murine tumor models to assess their potential for clinical translation of CEST MRI [12]. Glucosamine (2-amino-2-deoxy-d-glucose, GlcN) is a glucose analog commonly used food supplement and has excellent safety profile. It is taken up by tumor cells through the glucose transporter, then undergoes phosphorylation and can accumulate in the cells [13, 14]. Recently, it was demonstrated as an exogenous contrast agent for imaging of breast tumors and their metastases in mice [15, 16]. Here, the translation of the GlcN CEST MRI method to a clinical MRI setup was examined, in order to evaluate the feasibility of obtaining a new class of contrast agent for breast cancer MRI that enables the tumor to be distinguished from the surrounding tissue, and to compare it to the conventional MRI method.

Material and methods

Subjects

This study was reviewed and approved by the institutional review board (Tel Aviv and University and Sheba Medical Center); all patients signed informed consent. Twelve female patients with newly diagnosed breast lesions suspected as malignant cancer were included in this study, between August 2019 and March 2020. The histopathological markers of all enrolled patients are shown in Table 1. The patients were fasted for at least 6 h before the study and were scanned twice with the CEST MRI protocol: before and at 2 h after drinking a solution of 7.5 g of GlcN sulfate (Arthryl, Rafa) in 150 ml water.

Table 1 Histological characteristics of the patients
Full size table

MRI protocol

All MRI measurements were performed on a 3-T clinical MRI scanner (PRISMA, Siemens) using a bilateral diagnostic breast coil (16 channels, Siemens) with the parameters listed in Supplementary Material (Table 1S). All acquisition parameters were optimized according to phantoms results of GlcN solutions at 3T [17]. The routine breast MRI protocol also included T1W-DCE with the injection of gadoterate dimeglumine at a dose of 0.1 mmol/kg (Dotarem, Guerbet) contrast agent following the CEST scans.

Optimal slice positioning of the CEST sequence was supported by high-resolution T2W images acquired at the beginning of the examination and unilateral shimming was performed to improve B0 homogeneity. The CEST protocol included a series of saturation frequencies offsets (17–24) in the range of ± 6 ppm, using a train of 5 gauss saturation pulses with 100 ms long, interpulse delay of 61 ms and 2 s pause between measurements (total saturation pulse duration of 1.5–2 s), saturation attenuations of 2.5 μT, and recovery time of 5 s. One fully relaxed image was acquired for normalization. CEST sequence was applied with fat suppression and followed by a spiral-reordered 3D GRE read-out, resulting in a scan time of ~ 4–4.5 min.

Post processing

Data processing was performed with Matlab 2017a (The Mathworks). Conventional Z-spectra were corrected for B0 inhomogeneity (based on the B0 field-maps [18]) and normalized. ROIs of the tumors and surrounding tissues were carefully selected by the radiologist who had originally read the case clinically and based on complementary data obtained by Gd-T1W examinations. The calculation of the Z-spectra was performed on a voxel-by-voxel basis using spline interpolation of offsets acquired in the experiments [19]. Reported CEST values were averaged (mean ± standard deviation) within the defined ROIs based on magnetization transfer asymmetry (MTRasym) calculations:

$$ CEST\left(\Omega \right)=\left[{M}_{SAT}\left(\Omega \right)-{M}_{SAT}\left(-\Omega \right)\right]/{M}_0 $$
(1)

where MSAT(Ω) and MSAT(−Ω) are the signal intensity with RF saturation at Ω and −Ω, respectively, and M0 is the signal intensity without RF saturation.

To resolve individual contributions to the GlcN CEST effect, the Z-spectrum was analyzed using a five-pool Gaussian fitting model [20, 21] for the separation of the water, MT, hydroxyl, amine/amide, and NOE signals. Fitting the results by Gaussian functions was found to give better coefficient of determination (R2) than the conventional Lorentzian fitting. The reason may be that field inhomogeneity had non-negligible contribution to the line shape. The fitting of the Z-spectrum to the sum of multiple Gaussian functions was performed using the following equation:

$$ y={\sum}_i{A}_i\exp \left[-41n2{\left(x-{\omega}_i\right)}^2/{\sigma_i}^2\right] $$
(2)

where ω is the frequency offset from the water resonance, and Ai, ωi, and σi are the amplitude, frequency offset, and linewidth of the CEST peak for the ith proton pool, respectively. The integrals of the different pools were generated following their multi-Gaussian fitting and are given as integral = √(π/2)Aiσi.

Contrast to noise ratio (CNR) was calculated as (SB−SA)/√(σ2B+σ2A), where SA and SB are the mean values for the regions-of-interest (ROIs) of the tumors and the surrounding areas of normal breast tissue respectively and σA,B are their standard deviations [22, 23]. In the same way, CNR was calculated for the T1W imaging.

Statistical analysis

The nonparametric Wilcoxon signed-rank test was used to calculate statistical significance as obtained before and after GlcN oral administration. p-values < 0.05 were considered significant. The small size of the study group does not allow the evaluation of the statistical significance of this research results vs. the clinical results.

Results

CEST signal was found to be higher in malignant breast lesions than in surrounding breast tissue in all patients (median age, 53 years, N = 12, p = 2.2 E−03) (Table 2 and Figs. 1, 2). The CEST signal was calculated based on MTRasym for chemical shifts at 2 ppm from the water signal. The choice of this chemical shift was to allow isolation from the broad water signal. The CEST signal location was determined based on complementary data obtained by Gd-T1W examinations and according to its visual location as identified by an expert radiologist. Before the GlcN oral administration, the average CEST value for the tumors was 4.71 ± 3.31% while that of the normal tissue was 0.43 ± 0.36%. The average CEST values 2 h after the GlcN oral administration were 8.12 ± 4.09% and 0.48 ± 0.52% in the tumor and the normal tissues respectively (N = 12, p = 2.2 E−03). Thus, the average CEST value in the tumor increased by 3.41 ± 0.79% following the GlcN oral administration while that of the normal tissue remained about the same.

Table 2 Summary of all patients results as obtained from MRI data
Full size table
Fig. 1
figure 1

a–c A 34-year-old female patient with grade II–III invasive ductal carcinoma in the right breast and (d–f) a 63-year-old female patient with grade III invasive ductal carcinoma in the right breast. a, d Gadolinium (Gd) T1W images. b, e and c, f are CEST MTRasym maps calculated at frequency offset of 2 ppm before and after oral administration of GlcN, respectively

Full size image
Fig. 2
figure 2

Plots of the MTRasym at frequency of 2 ppm as obtained for tumor tissue measured in the patients (n = 12) (a) before and after oral administration of GlcN and (b) the average results compared with normal breast tissue (error bars represent the standard deviation of the distributions of MTRasym values) and (c) the averaged CNR results as obtained by the T1W, and CEST imaging methods (error bars represent the standard deviation of the distributions of averaged CNR)

Full size image

Figure 3 outlines the spectral results of the breast tumor as displayed with Z-spectra for patient no. 2. A multi-pool Gaussian fitting approach was applied to account for the contribution of each metabolite. Figure 3 points to the increase in the hydroxyl (OH) and amine/amide CEST signals as well as in the NOE signal of breast cancer tissue as observed by the Z-spectra following GlcN oral administration, indicating GlcN uptake in the tumors.

Fig. 3
figure 3

Z-spectra of breast tumor obtained for a 50-year-old female patient with grade III IDC, DCIS in the left breast (a) before and (b) after oral administration of GlcN and the corresponding multi-pool Gaussian fitting, shows the contribution from different pools (error bars represent the standard deviation of the distributions of normalized intensities values). c and d are the fitted different pools, separated from the raw Z-spectra, respectively (R2 > 0.99 and residual errors are less than 2%). The MT pool is not easily apparent in the figure since it is much smaller compared to the other pools

Full size image

Both the Gaussian fitting analysis and MTRasym analysis show GlcN CEST effects in patients with breast tumors; however, the Gaussian approach yields information of the different pool contributions to the Z-spectra as demonstrated in Fig. 4 and Table 2. Both CEST amplitudes and linewidths were increased after GlcN treatment (Table 2S, Supplementary Martial). The integrals for the hydroxyl, amine/amide CEST, and NOE pools were increased with averaged factors (N = 12) of 2.2 ± 1.2 (p = 3.38 E − 03), 1.4 ± 0.4 (p = 9.88 E − 03), and 1.6 ± 0.6 (p = 2.09 E − 02), respectively.

Fig. 4
figure 4

CEST peak integral change for each examined patient at the frequency ranges of (a) OH, (b) amine/amide, and (c) NOE as calculated by the multi-pool Gaussian fitting before and after oral administration of GlcN and (d) the corresponding averaged CEST peak integral for the signals of the OH, amine/amide, and NOE (for all patients; patient #8 was excluded from the analysis due to technical difficulty in Z-spectra fitting)

Full size image

Discussion

To our knowledge, this is a first demonstration for the detection of human breast cancer with GlcN CEST MRI method at a clinical setup. This study reveals the feasibility of the technique to differentiate between breast cancer tissue and normal-appearing breast tissue. This observation suggests that the method could be used to track changes in metabolic activity in response to GlcN uptake in the tumor, a key marker for cancer diagnosis. Therefore, the method may be used as a complementary molecular imaging method. At the same time, determining the method’s threshold conditions and thus identifying a typical GlcN CEST scale range for breast cancer will require a larger sample size. The high contrast to noise ratio (CNR) in the GlcN CEST images points to a potential in detecting small tumors at earlier stages. The MTRasym in the surrounding breast tissue was low, which is in agreement with recently published data [9].

This study showed that both the MTRasym analysis and the multi-pool Gaussian fitting may assist in detecting dynamic changes of the CEST signal in breast cancer following oral administration of GlcN. Since MTRasym calculations are impacted by the presence of other CEST species in the contralateral side of the direct effect, they were limited to the hydroxyl peak. Z-spectra were fitted with five Gaussian line shapes to separate the contributions of each peak contribution to the CEST effect. When compared with the results reported at higher preclinical field strength [16], lower absolute differences between the resonant frequencies of water and the other examined peaks at a clinical field strength give rise to considerably larger water direct saturation and spillover effects. In this study, we found that fitting to a Gaussian line shape [21] yielded better results than fitting to a Lorentzian line shape [20]. This could be related to the fact that the 3-T clinical scanner contributes more to field inhomogeneity than the 7T results. The multi-Gaussian fitting approach allowed to measure tumor contrast referring to 3 main exchangeable pools: the hydroxyl, amine/amide, and NOE pools. The integral increase of the 3 peaks, which is attributed to GlcN administration, is due to an increase in both linewidth and amplitude of the examined pools. The contrast may arise from GlcN and its metabolic product accumulation in the tumor, as was previously shown for a murine model [15, 16]. Since there is limited published data regarding the metabolic activity of breast carcinomas, it is difficult to anticipate which type of exchangeable protons was detectable [7]. A CEST peak around 1.5–2 ppm may originate from –OH protons belonging to GlcN metabolites or from glycosaminoglycans (GAGs). The origin of the CEST contrast detected at 2.5–4 ppm may be a weighted sum of signals that originate from different amine/amide groups arising from GlcN and its phosphorylated products. As was shown previously [15], the CEST effect of GlcN increases at lower pH values. Hence, the enhanced CEST effect following the GlcN administration can also be explained by tissue acidosis associated with lactate buildup [16]. Potential contributions from metabolites responsible for the NOE (~−3 ppm) could be proteins and peptides as well as N-acetyl groups. The NOE effect indicates that the CEST arises from intracellular metabolites of GlcN. Free GlcN molecules do not give forth to NOE effects [15]. Thus, the observation of NOE proves that GlcN enters into the cancer cells and that its metabolites contribute to the CEST effect. This result concurs with previous 13C NMR spectroscopy studies of extracts of implanted tumors in mice [16], which indicated the accumulation of GlcN and its metabolites in the tumors. One may note that the fat suppression has only a minimal effect on the CEST results since it is too short (few ms) to build up any additional CEST effects, and that this pulse is always applied even in the M0 image; thus, a potential perturbation of the water will be offset independent.

The GlcN CEST contrast varies from patient to patient due to a variety of factors such as tumor viability and malignancy, pH change of the surrounding tissue, and individual reaction to GlcN administration, which might all contribute to the wide range of results. The small number of patients is a limitation of this study that may impede the ability to decide whether there is a significant correlation between the observed CEST results and pathological markers. Thus, further clinical trials are required to assess the detectability and the sensitivity of GlcN CEST signal with regard to the tumor pathological characteristics.

In the present study, each patient was scanned twice: before and 2 h after the administration of the agent. The choice of this time interval was done since it takes 2 h to get high concentration of GlcN in the blood [24]. This might be time constraints encountered in clinical practice. However, since GlcN CEST has highly significant contrast for breast tumors, it is reasonable to suggest that in the future a single scan (at a range of frequencies that are most sensitive for the detection of GlcN) 2 h after oral administration of GlcN will be enough; thus, only an additional few minutes will be required over the routine clinical scan time.

In future, complementary studies should be performed in a larger scale of patients for further evaluation and optimization of the technique. Additionally, the GlcN CEST MRI technique should be examined for its ability to differentiate between malignant and non-malignant processes. In recent years, concerns have been raised regarding the repeated use of Gd [25]. GlcN CEST has potential clinical application that allows a non-Gd MRI of breast cancer. The similarity of the biochemical basis of the proposed method to that of the positron emission tomography (PET) points to the possibility that both morphological and metabolic information will be obtained in a single MRI session. It can serve as a guide for tumor diagnosis as well as its treatment monitoring.