Novel proton exchange rate MRI presents unique contrast in brains of ischemic stroke patients

Abstract

Background:

To map and quantify the proton exchange rate (k_ex) of brain tissues using improved omega plots in ischemic stroke patients and to investigate whether k_ex can serve as a potential endogenous surrogate imaging biomarker for detecting the metabolic state and the pathologic changes due to ischemic stroke.

New Method:

Three sets of Z-spectra were acquired from seventeen ischemic stroke patients using a spin echo-echo planar imaging sequence with pre-saturation chemical exchange saturation transfer (CEST) pulse at B₁ of 1.5, 2.5, and 3.5 μT, respectively. Pixel-wise k_ex was calculated from improved omega plot of water direct saturation (DS)-removed Z-spectral signals.

Results:

The derived k_ex maps can differentiate infarcts from contralateral normal brain tissues with significantly increased signal (893 ± 52 s⁻¹ vs. 739 ± 34 s⁻¹, P < 001).

Comparison with Existing Method(s):

The k_ex maps were found to be different from conventional contrasts from diffusion-weighted imaging (DWI), CEST, and semi-solid magnetization transfer (MT) MRI. In brief, k_ex MRI showed larger lesion areas than DWI with different degrees and different lesion contrast compared to CEST and MT.

Conclusions:

In this preliminary translational research, the k_ex MRI based on DS-removed omega plots has been demonstrated for in vivo imaging of clinical ischemic stroke patients. As a noninvasive and unique MRI contrast, k_ex MRI at 3T may serve as a potential surrogate imaging biomarker for the metabolic changes of stroke and help for monitoring the evolution and the treatment of stroke.

Graphical abstract

1. Introduction

The search for novel and robust noninvasive molecular imaging techniques to accurately detect the metabolic disturbances of ischemic brain tissues is currently ongoing. Unlike the stroke during the hyperacute or acute phase, the subacute and chronic phases provide an adequate window for evaluating novel imaging techniques. In addition, through assessing microstructural and metabolic changes due to ischemia, MRI techniques can not only help to investigate the secondary clinical symptoms but also monitor the curative effects and predict the prognosis of the patients under subacute and chronic phases in clinics^1-4.

As an emerging endogenous metabolic MRI technique, chemical exchange saturation transfer (CEST) has shown advantages in the metabolic assessment of stroke^5-7. Proton exchange is the fundamental contrast mechanism for CEST MRI and the in vivo imaging of proton exchange rate (k_ex) has not been achieved until recently^8,9. Particularly, in vivo k_ex imaging of healthy brains has been demonstrated based on water direct saturation (DS) removed omega plots^8,10.

The omega plot was initially evaluated in a phantom and has shown the capacity of assessing proton exchange rate from paramagnetic CEST agents¹⁰. The resonance frequencies of labile protons in paramagnetic agents are far from the water resonance and hence their signals are negligibly affected by the water DS effect. However, the saturation offsets of exchangeable protons in endogenous diamagnetic CEST metabolites in biological tissues are close to the water proton resonance^11,12. Hence, the water DS effect is a dominant confounding factor contributing to the Z-spectral signal during in vivo imaging^13,14. Mathematical removal of the water DS signal has been used to eliminate the predominant contamination by the DS effect for the quantification of k_ex in vivo⁸. Our previous study using DS-removed omega plots has shown that the exchange rate maps can linearly reflect the pH changes in the physiological range (pH 6.2 to 7.4) for ex vivo protein solutions⁸.

The purpose of this study was to investigate the changes of k_ex in ischemic brain tissues of subacute and chronic stroke patients based on the DS-removed omega plots and to evaluate the potential value of k_ex imaging in the detection of physiopathologic changes in ischemic stroke under clinical settings.

2. Methods

2.1. Patients

The study was approved by the institutional ethics committee and written informed consent was obtained from each participant prior to the examination. A total of 23 patients with ischemic stroke (16 men, 7 women, mean age 56 years, age range 32–81 years, duration of stroke onset from 2 to 18 days) were recruited from December 2018 to April 2019 in this prospective study. The inclusion criteria included suspected symptoms of ischemic stroke and explicit infarcts observed on CT or MRI. Lesions were consistent with symptoms and arterial distribution, showing low density in CT, or high signal in diffusion-weighted imaging (DWI) and T2-weighted fluid-attenuated inversion recovery (T2-FLAIR), low signal in T1-weighted fluid-attenuated inversion recovery (T1-FLAIR). Exclusion criteria included hemorrhagic transformation detected by CT or T1-FLAIR, bilateral hemispheric infarcts, insufficient image quality, other nervous disorders, and all contraindications to MRI. Of these patients, 6 excluded because of hemorrhagic transformation (n = 1), bilateral hemispheric infarcts (n = 2), and motion artifacts (n = 3). Thus, 17 cases were included for the final analysis.

2.2. Image acquisition

All MRI examinations were performed on a 3T MRI system (GE Medical Systems, Discovery MR750, Waukesha, WI, USA) with a 32-channel head coil. Several routine MRI protocols were used for localizing the infarct and excluding any other lesions, including transverse T1-FLAIR, T2-FLAIR, and DWI based on spin-echo echo-planar imaging (SE-EPI) sequence.

After the anatomical slice selection, single-slice Z-spectral data were acquired on a transverse section crossing the largest infarct region in reference to DWI using a SE-EPI sequence. The detailed parameters were as follows: repetition time (TR) = 3000 ms, echo time (TE) = 22.6 ms, matrix size = 128×128, field of view = 240 mm × 240 mm, slice thickness = 5 mm, CEST saturation power (B₁) was set to 1.5, 2.5 and 3.5 μT, saturation duration = 1500 ms, a total 33 saturation offsets including 0 to ±5 ppm at an increment of 0.25 ppm, plus ±6, +15.6, and +39.1 ppm (as reference image), and acquisition time of 3 min 18 s for each of the three Z-spectra.

2.3. Data analysis

All raw data processing and analysis were performed using home-built programs developed in MATLAB (2018a, MathWorks, Natick, MA, USA). Prior to analysis, motion artifacts in Z-spectral images were corrected based on image registration routine "imregister". All Z-spectra were B₀-corrected based on the B₀ map estimated from the Z-spectrum collected with the minimal saturation power (1.5 μT). Then the Z-spectra were flipped to be 100 × (1 − M_z/M₀), in which M₀ is the reference for Z-spectrum signal (M_z). The flipped Z-spectrum was then fitted to a linear combination of two Lorentzian functions corresponding to the bulk water and the residual signal, centered around 0 and −1.5 ppm respectively, chosen according to previous study (Cai et al., 2015). . The residual signal contains all exchanging components, including CEST, magnetization transfer (MT) from semi-solid components, and relayed nuclear overhauser enhancement (NOE)¹³. After the fitting, the bulk water peak was subtracted from the raw Z-spectra and the residual signal was used for further omega plot analysis. The omega plot was constructed as the plot of M_z/(M₀−M_z) versus $1∕{\omega }_1^2$, where M_z & M₀ are now taken from the signals of the DS-removed spectra, and ω₁ = γB₁ is the amplitude of the saturation radiofrequency (RF) pulse. Linear fit to the omega plot was used to determine the X-axis intercept (typically negative). Next, the intercept value was used to calculate the average k_ex of all contributing metabolites according to $\frac{-1}{x_{intercept}}=k_{ex}$ . In our previous research, a more generalized equation as follows was derived in case of more than one exchanging metabolite species contributing to the saturation transfer signal⁸.

$$\frac{M_z}{M_0-M_z}=\frac{1}{T_1^w}\left(\frac{{\sum }_{i=1}^n\frac{k_{s_i}}{f_{s_i}}}{{\omega }_1^2}+{\sum }_{i=1}^n\frac{1}{f_{s_i}{k}_{s_i}}\right)$$ (1)

in which $T_1^w$ is the longitudinal relaxation time of the water, k_si are the exchange rates of all metabolites’ pool and the water pool and f_si are the fraction of protons on the metabolites relative to water’s. Similar to the original omega plot equation, Equation 1 indicates a linear relationship between $\mathcal{Y}=\frac{M_z}{M_0-M_z}$ and the inverse of the square of saturation pulse power, $\mathcal{X}=\frac{1}{{\omega }_1^2}$. This means the X-axis intercept of the $\mathcal{Y}$ vs. $\mathcal{X}$ plot (the omega plot) can provide a readout of the convoluted k_ex average of all exchanging components according to the following equation.

$$\frac{-1}{x_{intercept}}=\frac{{\sum }_1^n\frac{k_{s_i}}{f_i}}{{\sum }_1^n\frac{1}{f_i{k}_{s_i}}}=M_1\left(\frac{k_{s_i}}{f_i}\right)\times M^{-1}\left(f_i{k}_{s_i}\right)=p^2\left(f_i{k}_{s_i}\right)=p^2({\overline{k}}_{ex})$$ (2)

where M₁ and M⁻¹ are the arithmetic and the harmonic mean operator, respectively.

Here, the quantified exchange rate value reflects a profile of the average exchange rate of all the contributing components and exchanging metabolites to the Z-spectral signals. Following this procedure for each pixel, the exchanging rate maps of the brains without the DS effect were then constructed.

For comparison, conventional CEST and MT maps were quantified as

$${\text{CEST}}_{asym}=\frac{M_z(-3.5ppm)-M_z(+3.5ppm)}{M_0}\times 100\%$$ (3)

and

$$\text{MT}=(1-\frac{M_z(+6ppm)}{M_0})\times 100\%$$ (4)

Given that stroke lesions were visually differentiable from control tissues in k_ex images, regions of interest (ROIs) of infarcts and the corresponding contralateral healthy tissues were manually drawn in reference to k_ex images by a neuroradiologist (5 years of experience) and double-checked by another senior neuroradiologist (8 years of experience). There were some small hot spots in k_ex images that are potentially small infarct lesions and to be confirmed with thorough investigation. Without further verification, they were not included as lesion ROIs. Values of all pixels within each ROI were averaged for k_ex, CEST and MT contrast maps. DWI appeared to delineate different lesion areas than k_ex images. Therefore, the lesion ROIs were separately manually drawn on the DWI images and then the areas of infarct lesions both in k_ex and DWI were compared. To summarize CEST or MT contrast at different B₁s, the percentage changes of the CEST or MT contrast from ischemic tissues compared to contralateral tissues were also computed as 100% × (Lesion – Control)/Control.

2.4. Statistical analysis

Two-tailed paired Student’s t-tests were used to compare parameters (k_ex, CEST and MT) between ischemic lesions and contralateral normal tissues for the whole patient group. The value of parameters were presented as mean ± standard deviation. P < 0.05 was considered statistically significant. Statistical analysis was performed using Statistical Product and Service Solutions (SPSS, version 19.0, Chicago, IL) software.

3. Results

Fig. 1a demonstrates Z-spectra data from infarct brain tissues at different saturation B₁s. Z-spectrum was fitted for direct saturation (Fig. 1b) and DS-removed residual signal was used to construct omega plots and quantify k_ex for all pixels from lesions or control tissues as demonstrated in Fig. 1c. k_ex map from a representative subject is shown in Fig. 2. Infarct lesion was clearly delineated in the k_ex map with a remarkably elevated signal. Compared to contralateral normal tissues, k_ex from lesions increased significantly (893 ± 52 s⁻¹ vs. 739 ± 34 s⁻¹, P < 0.001, Fig. 2c). As shown in Fig. 3, infarct lesions typically showed larger areas in k_ex maps than they were in DWI images. Quantitatively, the areas of lesions increased with different degrees (from 3.69% to 297%, mean and standard deviation 43.6 ± 67.7%, P < 0.05) in k_ex maps compared to that delineated in DWI images for all patients.

Fig. 1.

Direct saturation removed Omega plots for the quantification of k_ex. a Typical Z-spectra of infarct brain tissues plotted form −6 to 6 ppm for three different saturation powers B₁s (1.5, 2.5, 3.5 μT). b The process of removing water DS by Lorentzian fitting is demonstrated with a Z-spectrum at 1.5 μT. DS was subtracted from the entire Z-spectrum and the residual signals were used for constructing the omega plot. c Typical DS-removed omega plots of infarct (red line) and control brain tissue (blue line). The plots were linear, and the X-intercept of the omega plots provided a direct readout of the exchange rates of infarct or healthy brain tissues.

Fig. 2.

k_ex map from a representative ischemic stroke patient. a DWI exhibited an infarct lesion in the left occipital lobe. b k_ex map also clearly showed the lesion with an elevated signal. c The bar plots showed that k_ex of lesions were significantly increased compared with that of the contralateral tissues for all patients. * represents significant differences between infarct lesions and control brain tissues for k_ex.

Fig. 3.

A representative case showing that the k_ex map defines larger lesion area than DWI. a DWI exhibited a massive infarct in the left frontal lobe and scattered small infarcts in the left occipital lobe. The ROI profiles of infarct lesions in DWI image are delineated with red lines. b k_ex map exhibited increased signal covering the majority of the whole left hemisphere. The ROI profile of infarct lesion in k_ex map is delineated with black line. c Larger lesion areas in k_ex maps (k_ex(area)) were shown from all individual patients compared to lesion areas delineated with DWI images (DWI_(area), normalized to 1).

In Fig. 4, a representative k_ex map was compared with conventional CEST and MT contrast maps. While the k_ex map clearly define lesions from control tissues, the signals of lesions varied greatly with B₁ in CEST and MT maps (Fig. 4d-i). At B₁ = 1.5 μT, compared with contralateral healthy tissues, CEST signals of lesions were reduced, however, they were increased at 2.5 or 3.5 μT. No statistically significant difference was shown between infarcts and contralateral tissues (P > 0.05) for CEST in any B₁ due to large standard deviations (Fig. 5a, b). On the other hand, MT values of lesions were consistently reduced under varying B₁s (Fig. 5c, d).

Fig. 4.

The comparison among k_ex, CEST and MT maps from a representative patient. a-c T2-FLAIR, DWI and k_ex maps clearly showed the location and extent of the infarct. d-i CEST and MT contrasts of lesions were highly variable with B₁.

Fig. 5.

CEST/MT contrasts of infarct lesions varied with saturation amplitude B₁. a ompared to healthy brain tissues, CEST contrast of lesions was reduced at 1.5 μT, but increased at 2.5 or 3.5 μT, confirmed by the percentage changes between lesions and healthy tissues as shown in Figure b. Figure c shows MT of lesions was consistently reduced, however, at decreased degrees with higher B₁ as shown in Figure d. * represents significant differences (P < 0.05) between infarct lesions and control brain tissues for MT.

4. Discussion and Conclusions

In the present study, we have successfully implemented k_ex MRI for clinical patients with ischemic stroke. By comparing to other conventional MRI contrasts, we have investigated k_ex as a novel and unique contrast for the delineation of infarct lesions in human brain.

The k_ex maps were derived from the DS-removed Z-spectral signals using the improved omega plot method according to the previous study⁸. Since omega plots are linear functions, at least two Z-spectra collected with different RF saturation B₁s are needed. Under different B₁s, varied exchanging proton species with CEST, NOE, or MT effects were selected. For example, amide proton transfer (APT) CEST effect is optimally saturated under low B₁ around 1 to 2 μT due to the relatively slow amide proton exchange rate¹⁶. Glutamate CEST effect, on the other hand, is optimal under high saturation powers, for instance B₁ ≥ 3 μT, given the relatively fast amine proton exchange rate in glutamate¹⁷. That is to say, the metric of tissue k_ex as a general physical parameter is the weighted average for all exchangeable proton species in tissues.

Furthermore, the calculation of k_ex from the omega plot is based on the X-intercept of the plot, which depends on the B₁ values of the RF saturation pulse. One can mathematically calculate the influence of B₁ inhomogeneity. A dB₁ inhomogeneity in the pulse strength shifts the omega plot and its X-axis intercept with the amount

$$d\left(\frac{1}{{\omega }_1^2}\right)=-\frac{2d{\omega }_1}{{\omega }_1^3}$$ (5)

where ω₁ = γB₁. As a result, the inaccuracy in the calculated exchange rate equal to

$$d{k}_{ex}=-{\left(\frac{k_{ex}}{2\pi \gamma {\mathrm{B}}_1}\right)}^{3}2\pi \gamma d{B}_1$$ (6)

This means 5% of B₁ variation will lead to ~7% of k_ex deviation when k_ex is 800 s⁻¹, indicating the importance of a homogenous RF field. It is also worth mentioning that the omega plot is not sensitive to tissue relaxation times ( T₁ and T₂) given that those relaxation times are not part of the X-intercept of the omega plot as shown in Equation 1.

Based on our understanding, tissue k_ex could be influenced by pH, temperature, the profile of tissue metabolites containing slow or fast exchangeable protons, or reactive oxygen species (ROS)^8,18. Following ischemic stroke, increased anaerobic metabolism will lead to tissue acidosis and hence pH reduction in ischemic tissues. Given that proton exchange is base catalyzed around physiological pH, k_ex is expected to decrease due to acidosis⁷. In addition, brain temperature is generally well controlled and stable. Particularly, Karaszewski et al. demonstrated that the temperature of ischemic brain tissues changed within 1 °C¹⁹. Besides pH and temperature, the increase of the concentration of small metabolites with relatively fast proton exchange rate may lead to increase in the overall tissue k_ex in infarct lesions. It was reported that some small metabolites, such as excitatory amino acids (glutamate and aspartate) and inhibitory amino acids (GABA and taurine), elevate during the initial stage of stroke (< 24 hours)^20-22. However, the majority of small metabolites with generally higher exchanging protons decreases after 24 hours^23,24. Since all of our measurements in this study were performed after 24 hours, the reduction in small metabolites with a relatively faster proton exchange rate compared to macromolecules will likely reduce the observed tissue k_ex. The fact that we observed elevated k_ex in the infarct lesions eliminates pH, temperature, and the metabolite change as the major factors for the k_ex changes due to ischemic stroke.

Our recent studies have discovered that ROS can uniquely and sensitively enhance tissue k_ex¹⁸. In stroke, cerebral ischemia is accompanied by the imbalance between the production of ROS and the antioxidant capacity, which leads to excessive ROS production directly or indirectly involved in oxidative damage to lipids, proteins, and nucleic acids, etc., resulting in cellular damage and death²⁵. In addition, the complex metabolic changes including anaerobic metabolism, metabolite profile and oxidative stress are mainly induced in the acute ischemic stroke, however, ROS production is further stimulated by the reperfusion after acute stage^26,27. Hence, the significantly elevated k_ex of the infarct regions in the present study may due to the elevated ROS production. The hypothesis needs further investigation with invasive measurements of tissue ROS in ischemic animal models.

It is interesting also to note that the infarct areas in k_ex maps were generally larger than they were in DWI images, indicating that k_ex imaging is unique from diffusion MRI for the definition of infarct lesions. This is likely due to the distinctive contrast mechanism between these two techniques. In stroke clinics, the definition of ischemic penumbra can potentially help with the decision of recanalization therapy. Whether this mismatch may help to define ischemic penumbra remains to be investigated with the aid of longitudinal and histological studies.

We observed that CEST and MT contrast maps for the definition of infarct lesions varied with saturation amplitude B₁. CEST of lesions were less than control brain tissues at 1.5 μT, consistent to previous studies with relatively low B₁s(<1.5 μT)^7,28,29. But, with the saturation powers of 2.5 or 3.5 μT, CEST of lesions were increased to be higher than contralateral tissues. Zhao et al.³⁰ also reported similar variations of the CEST contrast in infarcted tissues due to varied B₁. On the other hand, although MT was significantly lower in lesions compared to contralateral tissues, its percentage changes were different under varying B₁s. k_ex MRI reflects a physical parameter and has been shown here to be independent and differentiable from conventional CEST and MT MRI contrasts.

With recent advances^31,32, magnetic resonance spectroscopy imaging (MRSI) may be used for high-resolution 3-D imaging of brain metabolites with enhanced clinical translatability. Compared to MRSI, molecular CEST MRI is sensitive to a different set of metabolites given to its unique signaling mechanism. For instance, glutamate and glutamine show as one complex peak in MRS while CEST MRI can only detect glutamate in regular MRI settings. The combination of MRS and CEST methods may help to quantify both. Similarly, creatine and phosphorous creatine have different presentations in MRS and CEST MRI. k_ex MRI as a derivative from CEST technique has been found to be sensitive to tissue microenvironment, such as pH and ROS; both are currently not detectable with ‘H-MRS or MRSI.

While the present study has successfully demonstrated the implementation of k_ex MRI for ischemic stroke patients, limitations are listed as follows. First, the scanning time may be a concern. It took close to 10 minutes in this study to acquire the entire three sets of Z-spectra for k_ex mapping of a single slice. However, since omega plots are linear functions, the total scanning time may be greatly reduced by using only two Z-spectra, and further reduced by decreasing the number of saturation offsets in each Z-spectrum⁸. Second, the 1.5 seconds saturation duration that we used in our data collection is not optimal for reaching steady state; however, due to the tissue’s specific absorption rate (SAR) limits in clinical settings, relatively short duration is preferred. According to simulations with Bloch-McConnell equations, 1.5 seconds of saturation can still achieve around 65% of the steady state saturation when B₁ is 1μT and 99% of the steady state value when B₁ is 4μT for the k_ex in the 30 to 3000 s⁻¹ range⁸. Third, it is a limitation of the present study that there is a lack of an automatic and objective algorithm for lesion ROI delineation from DWI and k_ex images. Finally, the sample size used in this preliminary study is relatively small. Larger sample sizes at different stages of stroke would enhance the characterization of stroke progression. Future longitudinal studies will help to delineate temporal evolvement of tissue k_ex post infarction.

In conclusion, we have demonstrated the implementation of k_ex MRI for assessing ischemic stroke patients. The utility of this novel and independent contrast may lead to a new perspective on stroke diagnosis and management.

Highlights.

• Proton exchange rate (k_ex) imaging was proposed based on water direct saturation (DS)-removed omega plots

• k_ex MRI can differentiate infarcts from contralateral normal brain tissues

• k_ex MRI was different from conventional MRI contrasts