pH-sensitive MRI demarcates graded tissue acidification during acute stroke — pH specificity enhancement with magnetization transfer and relaxation-normalized amide proton transfer (APT) MRI

Abstract

pH-sensitive amide proton transfer (APT) MRI provides a surrogate metabolic biomarker that complements the widely-used perfusion and diffusion imaging. However, the endogenous APT MRI is often calculated using the asymmetry analysis (MTR_asym), which is susceptible to an inhomogeneous shift due to concomitant semisolid magnetization transfer (MT) and nuclear overhauser (NOE) effects. Although the intact brain tissue has little pH variation, white and gray matter appears distinct in the MTR_asym image. Herein we showed that the heterogeneous MTR_asym shift not related to pH highly correlates with MT ratio (MTR) and longitudinal relaxation rate (R_1w), which can be reasonably corrected using the multiple regression analysis. Because there are relatively small MT and R_1w changes during acute stroke, we postulate that magnetization transfer and relaxation-normalized APT (MRAPT) analysis increases MRI specificity to acidosis over the routine MTR_asym image, hence facilitates ischemic lesion segmentation. We found significant differences in perfusion, pH and diffusion lesion volumes (P<0.001, ANOVA). Furthermore, MRAPT MRI depicted graded ischemic acidosis, with the most severe acidosis in the diffusion lesion (−1.05±0.29%/s), moderate acidification within the pH/diffusion mismatch (i.e., metabolic penumbra, −0.67±0.27%/s) and little pH change in the perfusion/pH mismatch (i.e., benign oligemia, −0.04±0.14%/s), providing refined stratification of ischemic tissue injury.

Graphical Abstract

1. Introduction

Amide proton transfer (APT) MRI is promising in capturing tissue pH changes in disorders such as acute ischemia and tumor by depicting the chemical exchange saturation transfer (CEST) effect between the endogenous protein/peptide amide protons and bulk water (McVicar et al., 2014; Sheth et al., 2012; Sun and Sorensen, 2008; Sun et al., 2012; Ward et al., 2000; Zhou et al., 2003). As tissue acidosis has been postulated as a surrogate metabolic biomarker, pH imaging augments conventional MR spectroscopy (MRS)-based techniques by refining tissue classification (Chang et al., 1990; Moon and Richards, 1973; Ojugo et al., 1999). Indeed, it has been shown that pH MRI detects metabolic penumbra, complementing the commonly used perfusion and diffusion MRI for mapping metabolic disruption following ischemia (Harston et al., 2015; Sun et al., 2011b; Sun et al., 2007c).

APT MRI is often measured using the magnetization transfer (MT) asymmetry (MTR_asym) to compensate for RF spillover effect. In addition to pH, MTR_asym however, is susceptible to semisolid magnetization transfer, nuclear overhauser effect (NOE) and relaxation (Heo et al., 2016; Jokivarsi et al., 2007; Sun et al., 2005; Woessner et al., 2005). Particularly, because MT and NOE contributions are asymmetric, MTR_asym image is of limited pH specificity (Desmond and Stanisz, 2012; Heo et al., 2016; Sun et al., 2007b; Zhou et al., 2004; Zong et al., 2014). Although normal brain white (WM) and gray matter (GM) have similar intracellular pH (Back et al., 2000), they appear drastically different in the pH-weighted MTR_asym image. Consequently, it has been very challenging to resolve graded metabolic disruption within the heterogeneous perfusion/diffusion lesion mismatch (Harston et al., 2015; Sun et al., 2007c). Jin et al. estimated that the mobile amide proton concentration is about 10–20% higher in brain GM than WM (Jin et al., 2013). Concurrently, brain GM longitudinal relaxation time is slightly longer than that of WM (de Graaf et al., 2006). Hence, we postulated that concomitant saturation transfer effects (i.e., MT and NOE) and bulk water relaxation variation across the brain may contribute substantially to the pH-independent heterogeneity in the routine MTR_asym image, correction of which could improve the pH specificity of APT MRI. Because the relaxation and MTR changes during acute stroke are relatively small, we proposed magnetization transfer and relaxation-normalized APT (MRAPT) analysis for acute stroke imaging, in complementary to routine stroke MRI for improved stratification of graded metabolic injury.

2. Theory

The in vivo MTR_asym can be generally described by

$$\mathrm{M}\mathrm{T}{\mathrm{R}}_{\text{asym}}=\mathrm{A}\mathrm{P}\mathrm{T}\mathrm{R}+\mathrm{M}\mathrm{T}\mathrm{R}{\text{'}}_{\text{asym}}$$ (1)

which includes pH-sensitive APT effect (APTR) and an intrinsic MTR asymmetry shift (MTR’_asym) not related to pH. The APT effect (APTR) can be described by an empirical solution (Sun and Sorensen, 2008)

$$\mathrm{A}\mathrm{P}\mathrm{T}\mathrm{R}=\frac{{\mathrm{f}}_{\text{amide}}*{\mathrm{k}}_{\text{amide}}(\mathrm{p}\mathrm{H})}{{\mathrm{R}}_{1\mathrm{w}}+{\mathrm{f}}_{\text{amide}}*{\mathrm{k}}_{\text{amide}}(\mathrm{p}\mathrm{H})}*\mathrm{\alpha }*(1-\mathrm{\sigma })$$ (2)

where α is the amide proton labeling coefficient, σ is the bulk water spillover factor, f_amide and k_amide are labile amide proton concentration and pH-dependent exchange rate, respectively, and R_1w is the bulk water longitudinal relaxation rate. For typical endogenous amide protons, because of the relatively slow amide proton exchange rate and its dilute concentration, we have f_amide · k_amide << R_1w (Sun et al., 2007b; Zhou et al., 2003). Moreover, we have α≈1 and σ≈0 under RF irradiation level of 0.75 µT for amide proton exchange at 4.7 Tesla (Sun et al., 2012), and Eq. 1 can be simplified as

$$\mathrm{M}\mathrm{T}{\mathrm{R}}_{\text{asym}}\approx \frac{{\mathrm{f}}_{\text{amide}}*{\mathrm{k}}_{\text{amide}}(\mathrm{p}\mathrm{H})}{{\mathrm{R}}_{1\mathrm{w}}}+\mathrm{M}\mathrm{T}\mathrm{R}{\text{'}}_{\text{asym}}$$ (3)

Hence, MRAPTR can be shown to be (Wu et al., 2012)

$$\mathrm{M}\mathrm{R}\mathrm{A}\mathrm{P}\mathrm{T}\mathrm{R}={\mathrm{R}}_{1\mathrm{w}}*\mathrm{M}\mathrm{T}{\mathrm{R}}_{\text{asym}}(\mathrm{p}\mathrm{H})\approx {\mathrm{f}}_{\text{amide}}*{\mathrm{k}}_{\text{amide}}(\mathrm{p}\mathrm{H})+{\mathrm{R}}_{1\mathrm{w}}*\mathrm{M}\mathrm{T}\mathrm{R}{\text{'}}_{\text{asym}}\approx \mathrm{f}({\mathrm{R}}_{1\mathrm{w},2\mathrm{w}},\mathrm{M}\mathrm{T}\mathrm{R})$$ (4)

As there is no easy means to quickly measure MTR'_asym, we used linear regression analysis of measurable terms such as R_1w,2w and MTR to estimate MRAPTR, and denoted it as f(R_1w,2w, MTR). Note that MRAPTR depends on MTR_asym and is pH sensitive. Because normal brain has uniform pH distribution, MRAPTR can be estimated from its linear regression relationship with MTR and R_1w voxel by voxel from the intact brain tissue, which accounts for the baseline MRAPTR heterogeneity and uniform brain pH. During acute stroke, the difference between the experimentally measured pH sensitive R_1w*MTR_asym map and MRAPTR prediction, based on R_1w and MTR that show little change immediately following ischemia, should be dominated by ischemic acidosis-induced signal change. We have

$$\mathrm{\Delta }\mathrm{M}\mathrm{R}\mathrm{A}\mathrm{P}\mathrm{T}\mathrm{R}={\mathrm{R}}_{1\mathrm{w}}·\mathrm{M}\mathrm{T}{\mathrm{R}}_{\text{asym}}-\mathrm{f}({\mathrm{R}}_{1\mathrm{w},2\mathrm{w}},\mathrm{M}\mathrm{T}\mathrm{R})\propto \mathrm{\Delta }({\mathrm{f}}_{\text{amide}}·{\mathrm{k}}_{\text{amide}}(\mathrm{p}\mathrm{H}))$$ (5)

We derived the change of MRAPTR with the relationship of ΔMRAPTR ∝ f_amide · k_amide (ΔpH), given that there is little change in amide proton concentration immediately following ischemia.

3. METHODS

3.1. Animal Stroke Model

The animal experiments have been approved by the Institutional Animal Care and Use Committee. Adult male Wistar rats (Charles River Laboratory, Wilmington, MA) were anesthetized with 1.5–2.0% isoflurane/air mixture throughout the study. Heart rate and oxygen content of blood (SpO₂) were monitored online (Nonin Pulse Oximeter 8600, Plymouth, MN), and body temperature was maintained by a circulating warm water jacket. Ten normal (n=10) and twenty acute stroke rats (n=20) following the standard intraluminal middle cerebral artery occlusion (MCAO) surgery were imaged. Briefly, after exposure of the right carotid bifurcation and suturing of the common carotid and distal external carotid arteries, a silicone-coated 4-0 nylon suture (Doccol Corp., Sharon, MA) was advanced to block the middle cerebral artery. The filament was secured and the incision was closed with sutures.

3.2. Simulation

We simulated APT effect in the presence of semi-solid MT using 3-pool Bloch-McConnell equations in Matlab (Woessner et al., 2005). To simulate how bulk water relaxation affects APTR and T_1w-normalzied APTR effect, we simulated typical range of T_1w from 1.25 to 2 s, and T_2w from 50 to 100 ms. Parameters we used include: exchange rates k_amide=30 and 10 s⁻¹ for normal and ischemic pH, respectively, and k_MT=20 s⁻¹; concentrations f_amide=1/1000, f_MT=1/40; chemical shifts δ_amide=3.5 ppm, δ_MT=0 ppm with relaxation constants T_1amide,MT=1 s, T_2amide=20 ms and T_2MT=200 µs (Sun et al., 2007b; Zhou et al., 2003). Note that we chose a relatively long T_2MT similar to that used by Scheidegger et al., which allows the use of a simplified Lorentzian line shape to approximate the semi-solid MT contribution (Scheidegger et al., 2014). Because there is little overlap between NOE and APT effects, it is not necessary to include NOE in the simulation for the demonstration of relaxation-normalized APT MRI measurement (Jin et al., 2013; Zhang et al., 2016).

3.3. MRI

MRI scans were performed on a 4.7 Tesla small-bore scanner (Bruker Biospec, Erlangen Germany). Multi-slice MRI (5 slices, slice thickness/gap=1.8/0.2 mm, field of view=20×20 mm², image matrix=48×48) was acquired with echo-planar imaging (EPI). Our study chose a dual RF coil setup, including a 70 mm volume transmitter coil and an actively-decoupled 20 mm surface receiver coil, to simultaneously achieve homogeneous B₁ field and increased sensitivity in signal detection. In addition, we applied high order Fastmap shimming and reset the bulk water frequency using the point resolved spectroscopy (PRESS) protocol without water suppression. Diffusion MRI was obtained using single-shot isotropic diffusion-weighted MRI with two b-values of 250 and 1,000 s/mm² (repetition time (TR)/echo time (TE) = 3250/54 ms, 16 averages, scan time=2 min) (Mori and van Zijl, 1995). For pH-weighted APT MRI, we used the recovery time of 5,000 ms, primary RF saturation duration of 4,500 ms, and secondary RF saturation duration of 500 ms for an RF irradiation amplitude of 0.75 µT applied at ±3.5ppm (Sun et al., 2011a). The unsaturated control scan was signal-averaged 8 times, while the saturated images were averaged 32 times (scan time=4 min). T₁-weighted images were acquired using inversion recovery EPI, with seven inversion delays ranging from 250 ms to 3,000 ms (TR/TE = 6,500/15 ms, 4 averages, scan time=3 min); T₂-weigthed SE images were obtained with two TE of 30 and 100 ms (TR = 3,250 ms, 16 averages; scan time=2 min) (Cheung et al., 2012). In addition, amplitude modulated arterial spin labeling (ASL) perfusion MRI was acquired in acute stroke rats (TR/TE = 6,500/15 ms, time of saturation = 3,250 ms and 32 averages; scan time=7 min). We used B₁=4.7 µT, labeling distance of 15 mm, modulation frequency of 250 Hz (Alsop and Detre, 1998; Utting et al., 2005).

3.4. Data Analysis

Parametric T_1w map was obtained with mono-exponential fitting of the signal intensity as a function of the inversion time (I(i) = I₀⌊1−(1−η)e^−TI_i/T_1w⌋), where η is the inversion efficiency and TI_i is the ith inversion time. T_2w and apparent diffusion coefficient (ADC) maps were calculated as ${\mathrm{T}}_{2\mathrm{w}}=\frac{\mathrm{\Delta }\mathrm{T}\mathrm{E}}{\text{ln}(\mathrm{I}(\mathrm{T}{\mathrm{E}}_1)/\mathrm{I}(\mathrm{T}{\mathrm{E}}_2))}$ and $\mathrm{A}\mathrm{D}\mathrm{C}=\frac{\text{ln}(\mathrm{I}({\mathrm{b}}_1)/\mathrm{I}({\mathrm{b}}_2))}{\mathrm{\Delta }\mathrm{b}}$, where TE_1,2 and b_1,2 are two TEs and diffusion b values, respectively, with ΔTE and Δb being their differences. In addition, cerebral blood flow (CBF) was derived as $\mathrm{C}\mathrm{B}\mathrm{F}=\frac{\mathrm{\lambda }({\mathrm{I}}_{\text{ref}}-{\mathrm{I}}_{\text{tag}})}{2\mathrm{\alpha }·{\mathrm{I}}_{\text{ref}}}·\frac{{\mathrm{e}}^{\mathrm{w}/{\mathrm{T}}_{1\mathrm{a}}}}{{\mathrm{T}}_{1\mathrm{w}}}$, where I_tag is the label image, I_ref is the reference image, λ is the brain-blood partition coefficient for water, α is the degree of inversion with transient time correction, w is the post-labeling delay, and T_1a is the arterial blood longitudinal relaxation time. In addition, MTR was calculated as $\mathrm{M}\mathrm{T}\mathrm{R}(\pm 3.5\mathrm{p}\mathrm{p}\mathrm{m})=1-\frac{\mathrm{I}(\pm 3.5\mathrm{p}\mathrm{p}\mathrm{m})}{{\mathrm{I}}_0}$, where I(±3.5 ppm) are the label and reference images with RF irradiation applied at ±3.5 ppm, respectively, and I₀ is the control image without RF irradiation. The mean MTR (MMTR) was calculated as the average of MTRs at ±3.5 ppm, and pH-weighted MTR_asym was calculated as $\mathrm{M}\mathrm{T}{\mathrm{R}}_{\text{asym}}=\frac{\mathrm{I}(-3.5\mathrm{p}\mathrm{p}\mathrm{m})-\mathrm{I}(+3.5\mathrm{p}\mathrm{p}\mathrm{m})}{{\mathrm{I}}_0}$. In addition, MRAPTR was calculated based on the multiple regression analysis between R_1w*MTR_asym, and R_1w and MMTR (i.e. MRAPTR=C₀+C₁₁*R_1w+C₁₂*MMTR+C₂* R_1w*MMTR, where Cs are regression coefficients). Note that we used MMTR instead of an individual MTR to harness the image sensitivity. In addition, ΔMRAPTR was calculated as the difference between experimentally measured R_1w*MTR_asym and MRAPTR estimated from R_1w and MMTR maps, using coefficients determined from the intact brain tissue (i.e. ΔMRAPTR= R_1w*MTR_asym − MRAPTR, Eq. 5). In addition, ischemic lesions in rCBF, ΔMRAPTR and ADC maps were automatically defined using a K-means clustering-based algorithm which categorized all the pixels within the brain into ischemic and normal clusters (Cheung et al., 2011).

4. Results

Fig. 1a simulates the experimental factor (i.e., α*(1-σ), B₁=0.75 µT at 4.7 Tesla, δ_amide=3.5 ppm) from the empirical solution (Eq. 2), as functions of T_1w and T_2w, for amide exchange rates of 30 and 10 s⁻¹, typical exchange rates under normal and acidic pH, respectively (Zhou et al., 2003). The experimental factor shows little change with relaxation and exchange rate, hence, can be treated as a constant that is near unity across the brain. Although APTR increases with exchange rate and hence pH, both APTR (Fig. 1b) and APTR contrast (i.e., ΔAPTR=APTR(k_amide=30 s⁻¹)−APTR(k_amide=10 s⁻¹)) between two amide exchange rates (Fig. 1c) show sizable T_1w dependence. In comparison, T_1w-normalized APTR contrast (i.e., R_1w* ΔAPTR) shows reduced T_1w dependency, with the coefficient of variation (COV) as a function of T_1w decreased from 9.3% (Fig. 1c) to 5.5% (Fig. 1d).

Fig. 1.

Simulation of T_1w-normalized APT MRI for two representative exchange rates of 30 and 10 s⁻¹ at 4.7 Tesla (B₁=0.75 µT). a) APT MRI experimental factor as a function of T_1w and T_2w. b) pH-sensitive APT ratio (APTR) as a function of T_1w and T_2w. c) APTR pH contrast (ΔAPTR) as a function of T_1w and T_2w. d) T_1wnormalized ΔAPTR as a function of T_1w and T_2w, showing reduced sensitivity to relaxation rates.

Fig. 2 shows multi-parametric images from a representative normal rat brain. T_1w map (Fig. 2a) appears hypointense in striatum and corpus callosum when compared to cortex. In addition, ventricle shows the highest intensity in T_1w due to increased water content. In comparison, T_2w map (Fig. 2b) appears relatively homogeneous except the hyperintensity in the piriform cortex and ventricle. MTR images at the reference frequency (MTR(−) at −3.5 ppm) and amide proton labile frequency (MTR(+) at +3.5 ppm) are shown in Figs. 2c and 2d, respectively. Corpus callosum and striatum show hyperintense in MTR maps, indicating an increased contribution of semisolid macromolecules and myelin MT. Moreover, MTR(−) shows higher intensity than MTR(+), mainly due to NOE contribution around the reference offset of −3.5 ppm and to a lesser degree, asymmetric MT effect (Heo et al., 2016). Figs. 2e and 2f show the mean MTR ([MTR(+)+MTR(−)]/2) and MTR_asym ([MTR(−)−MTR(+)]) maps, respectively. Note that striatum and corpus callosum appear hyperintensive in MMTR images while they display hypointensity in MTR_asym map. It is important to point out that there is little pH difference between normal brain WM and GM, and therefore the heterogeneity in the routine MTR_asym image is pH-independent, correction of which should improve specificity of APT MRI to pH (Back et al., 2000; Jin et al., 2013).

Fig. 2.

Multi-parametric MRI images from a representative normal rat. a) Longitudinal relaxation time T_1w map. b) Transverse relaxation time T_2w map. c) MTR(−) image (i.e., MTR at −3.5 ppm). d) MTR(+) image (i.e., MTR at +3.5 ppm). e) MMTR image (i.e., mean of MTR(±) images). f) pH-weighted MTR_asym image.

Fig. 3 evaluates pixel-wise correlation between MTR_asym and relaxation as well as MTR. There was significant correlation between R_1w-scaled MTR_asym and R_1w (Fig. 3a), R_2w (Fig. 3b) and MMTR (Fig. 3c). Specifically, we found R_1w*MTR_asym= 6.1% −13.0%*R_1w (R²=0.79, P<0.01), R_1w*MTR_asym=5.3% − 0.4%*R_2w (R²=0.42, P<0.01) and R_1w*MTR_asym= 9.6% −41.6%*MMTR (R²=0.75, P<0.01), with the linear regression fitting shown in dashed lines. With the group analysis of all normal animals, R² was found to be 0.76 ± 0.04, 0.39 ± 0.08 and 0.68 ± 0.10 between R_1w*MTR_asym and R_1w, R_2w and MMTR, respectively. We further tested whether multiple regression can enhance the prediction of MTR_asym (STATA, StataCorp LP, College Station, TX). We found that R_2w is not a significant predicator, and we had R_1w*MTR_asym=−13.2% +25.6%*R_1w +55.7% *MMTR −115.9%*MMTR*R_1w (R²=0.88, P<0.001). When all normal animals were analyzed, R² was 0.83±0.05, showing that the majority of the intrinsic heterogeneity in the routine MTR_asym map is due to longitudinal relaxation and magnetization transfer/NOE contributions. We further tested whether the proposed MRAPT MRI reduces the intrinsic pH-independent heterogeneity. R_1w-scaled pH-weighted MTR_asym map (Fig. 3e) shows noticeable contrast between brain WM and GM. Fig. 3f shows the predicted MTR_asym map based on multiple regression from MT and T_1w images. Fig. 3g shows ΔMRAPTR image, which appears relatively homogeneous across the brain. Indeed, the contrast-to-noise ratio between the striatum and cortex decreased from 2.8±0.8 to 0.3±0.4 (Figs. 3e vs. 3g, P<0.01, ANOVA), which confirms the effectiveness of MRAPT analysis in the removal of intrinsic heterogeneity not related to pH in the intact brain.

Fig. 3.

Pixel-wise regression analysis between R_1w-scaled MTR_asym with multiple MRI parameters from a representative normal rat. a) R_1w*MTR_asym as a function of R_1w. b) R_1w*MTR_asym as a function of R_2w. c) R_1w*MTR_asym as a function of MMTR. d) R_1w*MTR_asym as a function of R_1w and MMTR. e) R_1w*MTR_asym image. f) R_1w and MMTR-based multiple regression prediction of R_1w*MTR_asym image. g) ΔMRAPTR image, the difference between the experimentally measured R_1w*MTR_asym image and the multiple regression prediction.

We tested the MRAPT analysis during acute stroke. Parametric T_1w (Fig. 4a) and MMTR (Fig. 4b) maps show relatively little change immediately following ischemia. Both MTR_asym and relaxation-normalized MTR_asym maps (Figs. 4c and 4d) show hypointensity in the ipsilateral ischemic brain, indicating acidosis. However, Figs. 4c and 4d display noticeable brain WM/GM contrast over the contralateral intact brain tissue, suggesting poor pH specificity of the routine MTR_asym image. The MRAPTR from the contralateral normal brain can be well described with multiple regression analysis of R_1w and MMTR (Fig. 4e). The MRAPT analysis shows substantially improved tissue homogeneity in the contralateral normal brain, which enhances the specificity of ΔMRAPTR to pH (Fig. 4f). Indeed, acidic lesion can be reliably segmented using k-means clustering (Fig. 4f) from ΔMRAPTR image while the tissue classification in a routine MTR_asym map is poor (data not shown). Interestingly, the ΔMRAPTR image clearly demonstrated graded tissue acidification; the striatum was demonstrated with the most severe acidification which is concurrent with diffusion lesion, while the cortical region showed a moderate pH reduction. In comparison, CBF map (Fig. 4g) shows reduced blood flow in the ipsilateral ischemic area while diffusion MRI (Fig. 4h) depicts infarction predominantly in striatum. Fig. 4i overlays three lesion areas: diffusion lesion (black), pH/diffusion lesion mismatch (green) and perfusion/pH lesion mismatch (red).

Fig. 4.

MRAPT MRI analysis from a representative acute stroke rat. a) Parametric T_1w map. b) MMTR image c) pH-weighted MTR_asym image. d) pH-weighted R_1w*MTR_asym image. e) Multiple regression analysis of R_1w*MTR_asym as a function of R_1w and MMTR. f) pH-sensitive ΔMRAPTR image. g) CBF map. h) ADC map. i) Color-coded perfusion (red), pH (green) and diffusion (black) lesions.

Fig. 5 shows multi-slice perfusion, pH and diffusion lesions mismatch (Fig. 5a), and the corresponding 3D volume rendering overlaid on a structural scan from a representative acute stroke rat (Fig. 5b). Note that for the 3D view, we made the lesion colors moderately transparent so we can see through different layers of lesions, resulting in slightly fainted apparent color. We found significant differences in the volume between perfusion (323±41 mm³), pH (229±34 mm³) and diffusion lesions (132±52 mm³) (P<0.001, repeated measures One-way ANOVA, Bonferroni post-hoc test). The conventional perfusion/diffusion mismatch paradigm (Fig. 6a) demarcates the ischemic tissue into the most severely injured diffusion lesion (Area I) and the viable perfusion/diffusion lesion mismatch (Area II). We found that the diffusion lesion has slight yet significant rCBF decrease from that of perfusion/diffusion mismatch (Paired-t test, Table 1). Notably, the diffusion lesion shows significantly worsened acidosis when compared to the perfusion/diffusion mismatch. Our work shows that the conventional perfusion/diffusion lesion mismatch is pH heterogeneous, which can be further segmented into hypoperfused acidic tissue (i.e. pH/diffusion mismatch, Area IIa) and hypoperfused tissue with little acidification (i.e. perfusion/pH mismatch, Area IIb), as illustrated in Fig. 6b. We compared rCBF, ADC and ΔMRAPTR in the diffusion lesion, pH/diffusion and perfusion/pH mismatch (Table 2a). One-way ANOVA shows significant rCBF difference only between the diffusion lesion and perfusion/pH mismatch (Areas I vs. IIb, Table 2b). Although perfusion/pH and pH/ADC mismatches show significantly higher ADC value than the diffusion lesion (Areas I vs. IIa, and I vs. IIb), the ADC value for the perfusion/pH and pH/ADC mismatches are not significantly different (Areas IIa vs. IIb). Importantly, pH MRI captures graded acidification within the ischemic lesion, progressively worsened from the perfusion/pH mismatch (i.e., benign oligemia, −0.04±0.14%/s), pH/diffusion mismatch (i.e., metabolic penumbra, −0.67±0.27%/s) to diffusion lesion (−1.05±0.29%/s, P<0.0001).

Fig. 5.

Multi-parametric MRI segmentation of ischemic tissue injury. a) Multi-slice perfusion, pH and diffusion lesion segmentation. b) 3D volume rendering of ischemic lesions overlaid on a structural scan.

Fig. 6.

pH-aided ischemic lesion classification. a) Routine perfusion/diffusion MRI-based tissue stratification. b) pH MRI refines perfusion/diffusion lesion mismatch into two areas: 1) hypoperfused and acidic penumbra (Area IIa) and 2) benign oligemia area with little acidosis (Area IIb).

Table 1.

The rCBF, pH and ADC show significant difference between ADC lesion and PWI/ADC mismatch.

	DWI lesion (Area I)	PWI/DWI mismatch (Area II)	PWI/DWI mismatch vs. DWI lesion Mean Difference (P value)
rCBF	0.51 ± 0.11	0.58 ± 0.12	0.07 (0.001*)
ADC(µm2/ms)	0.64 ± 0.04	0.85 ± 0.04	0.21 (<0.001*)
ΔMRAPTR (%/s)	−1.05 ± 0.29	−0.32 ± 0.18	0.72 (<0.001*)

Table 2.

Comparison of multiple indices in refined lesion segmentation. a) One-way ANOVA shows significant change in rCBF, pH and ADC among ADC lesion, pH/ADC and PWI/pH mismatch. b) One-way ANOVA shows pH resolves all three regions: ADC lesion, pH/ADC mismatch and PWI/pH mismatch while rCBF and ADC have limited ability to refine the lesion zones.

a
	DWI lesion (Area I)	pH/DWI mismatch (Area IIa)	PWI/pH mismatch (Area IIb)	One-way ANOVA F (P value)
rCBF	0.51 ± 0.11	0.56 ± 0.14	0.62 ± 0.11	4.24 (0.019*)
ADC(µm2/ms)	0.64 ± 0.04	0.83 ± 0.05	0.84 ± 0.05	128.07 (<0.001*)
ΔMRAPTR (%/s)	−1.05 ± 0.29	−0.67 ± 0.27	−0.04 ± 0.14	89.17 (<0.001*)

b
	DWI lesion vs. pH/DWI mismatch (Areas I vs. IIa) Mean Difference (P value)	DWI lesion vs. PWI/pH mismatch (Areas I vs. IIb) Mean Difference (P value)	PWI/pH vs. pH/DWI mismatch (Areas IIa vs. IIb) Mean Difference (P value)
rCBF	0.05 (0.370)	0.11 (0.01*)	0.06 (0.274)
ADC(µm2/ms)	0.18 (<0.001*)	0.20 (<0.001*)	0.02 (0.490)
ΔMRAPTR (%/s)	0.37 (<0.001*)	1.01(<0.001*)	0.63 (<0.001*)

5. Discussion

Our study demonstrated that the MRAPT analysis substantially enhances imaging specificity to ischemic acidification during acute stroke, permitting refined tissue classification. We found that diffusion lesion suffers more aggravated acidification than the perfusion/diffusion lesion mismatch. Importantly, the development of MRAPT analysis allows semi-automatic lesion segmentation, which refines the heterogeneous perfusion/diffusion mismatch into hypoperfused acidic lesion (Area IIa) and hypoperfused tissue with little acidification (Area IIb) to capture graded ischemic tissue metabolic disruption. The finding of heterogeneous tissue metabolic injury is consistent with the hypothesis that the perfusion/diffusion mismatch includes not only ischemic penumbra but also benign oligemia that is not at risk to infarction (Kidwell et al., 2004; Sun et al., 2007c). It is helpful to briefly discuss and compare perfusion and pH measurements during acute stroke. Because CBF is under the influences of multiple factors including anesthesia and blood pressure regulation, we calculated the rCBF to reasonably account for such variabilities. As rCBF directly maps the severity of hypoperfusion, it is very sensitive to ischemia. Yet owing to the heterogeneous collateral flow and tissue susceptibility to ischemia, the rCBF measurement is associated with but not very specific to tissue metabolic disruption. In comparison, despite the fluctuation of cerebral perfusion and oxygen saturation, tissue pH is well regulated under the normal physiological conditions. Tissue pH shift is associated with glucose/oxygen delivery and consumption imbalance during the acute stroke (An et al., 2015; Shen et al., 2016; Shu et al., 2016; Zhu et al., 2013). Because tissue acidification is exacerbated by the reduced buffering capacity of bicarbonate and hypoperfusion, often leading to cell death and tissue damage (Siesjo, 1992), pH imaging can serve as an important surrogate metabolic imaging biomarker during acute ischemic stroke. Notably, Back et al. measured tissue pH with umbelliferone fluorescence and found that penumbral pH changes reflect the local disturbance of pH regulation and, possibly, the differential fate of penumbral tissue (Back et al., 2000). Specifically, they found that pH in the peri-infarct rim was 6.53±0.24, about 50% less when compared to the pH change in the infarct core, which decreased to 6.03±0.36. Because the amide proton exchange rate is dominantly base-catalyzed, the initial pH change will yield slightly larger APT MRI contrast change, suggesting that our results are in good agreement with previous findings (Back et al., 2000).

We would like to briefly discuss the confounding NOE contribution to pH-sensitive APT MRI in acute stroke. Jones et al. showed that in concentrated bovine serum albumin solution, both NOE and APT effects show pH dependence (Jones et al., 2013). On the other hand, Jin et al. (Jin et al., 2013) and Zhang et al. (Zhang et al., 2016) found very small NOE effect change following acute stroke. We would like to point out that Jones et al. (Jones et al., 2013) used pulsed RF saturation scheme in volunteers and tumor patients, while Jin et al. (Jin et al., 2013), Zhang et al. (Zhang et al., 2016) and our study here used continuous wave saturation in acute ischemic animals. The differences in saturation schemes, diseases of interest, and the magnitude of pH change may partially explain the different conclusions of NOE contribution, which should be explored in the future. Although the routine MTR_asym image includes contributions from both ±3.5 ppm, it is most likely that the APT contrast change during acute ischemia is dominated by base-catalyzed amide proton exchange at 3.5 ppm.

We used a relatively weak RF irradiation power, which balances APT and concomitant RF saturation effects at 4.7 Tesla yet not overly strong to saturate amine proton exchange that is two orders of magnitude faster than that of amide protons (Cai et al., 2012; Sun et al., 2007b). Under such conditions, simulation with and without MT pool shows that MT effect dilutes the CEST effect by about 20%. This explains why adding MT into the multiple regression (T_1w and MMTR) significantly improves the correlation from that using T_1w alone. Because the MRAPT analysis requires just an additional T_1w map in addition to the raw CEST data, the proposed analysis can be quickly tested for different experimental settings. Briefly, it is feasible to simultaneously adjust multiple variables, such as B₁, saturation time and repetition time to optimize the APT MRI sensitivity (Sun et al., 2013). For example, Zhao et al. demonstrated that under a short saturation duration of 500 ms, an RF power of 2 µT provides higher ΔMTR_asym(3.5ppm) than 1 µT (Zhao et al., 2011). It helps to keep in mind that when relatively large RF power levels are used, there may be overlapping CEST effects from multiple exchangeable groups, which requires Z spectrum acquisition and line fitting to resolve their independent contribution to enhance pH MRI. This, however, extends the scan and processing time, not desirable in the acute stroke setting.

Our study found relatively small MMTR and T_1w changes of under 1% and 7%, respectively, during acute ischemia, consistent with those in the previous studies (Jokivarsi et al., 2010; Makela et al., 2002). As such, MRAPT approach performs well for acute stroke imaging. Notably, the experimentally measured T_1w is complex, susceptible to the underlying CEST, NOE and MT effects as well as intracellular viscosity, fraction of bound and free water molecule. For dilute amide protons and under the condition of weak RF saturation, the observed T_1w provides a reasonable estimation of the effective relaxation rate under which the dilute CEST effect relaxes. This is confirmed by the finding that the multiple regression analysis based on the experimentally measured T_1w highly correlates with the intrinsic APTR inhomogeneity. In addition, there may be means to shorten the scan time. Our study used an inversion recovery sequence and measured quantitative T₁ map, which may be replaced by the look-locker MRI approach (Freeman et al., 1998). An alternative means is to test whether a T_1w-weighted image can provide satisfactory correction to further reduce the total scan time. However, depending on the stroke onset time, there may be presence of edema, which may cause water content change and heterogeneous amide proton distribution in the brain. In addition, APT MRI is susceptible to MT and NOE effects, which confound its quantification. Under such conditions, the emerging field of quantitative CEST (qCEST) MRI may eventually allow us to measure tissue acidosis independent of water content and relaxation changes. However, absolute pH imaging has to fully account for all these confounding factors within a minimal scan time, which could be challenging. The proposed MRAPT analysis is expeditious to acquire, simple to implement, yet provides refined lesion stratification. Future studies will investigate whether ΔMRAPTR can map the amount of pH change (ΔpH) from the baseline.

There are a few limitations of our current study. Our study here chose a relatively reproducible filament stroke model to refine pH imaging and aid ischemic tissue classification. Because the commonly used staining techniques (e.g. H&E and TTC) are not suitable for characterizing tissue acidification and injury in the first few hours, no pH histology was obtained in our current study. Nevertheless, our findings of severity of pH drop from penumbral tissue and ischemic core are in good agreement with that of Back et al. (Back et al., 2000). Our study also showed ischemic penumbra and benign oligemia have significant pH difference, not cerebral blood flow, consistent with the clinical observation of Harston et al. (Harston et al., 2015) It is worthwhile to mention that our animal system has excellent field homogeneity. Indeed, after relaxation and MT corrections, MRAPT images appear homogeneous across the intact brain tissue, and we did not perform additional field inhomogeneity correction. Field correction may be necessary when translating pH MRI to study large animal models and stroke patients (Kim et al., 2009; Scheidegger et al., 2011; Sun et al., 2007a; Zhou et al., 2008). In addition, our study chose a filament stroke model, which induces severe hypoperfusion with poor collateral circulation, causing large infarction days later. Nevertheless, our acute stroke imaging results are consistent with those reported by Harston et al., who showed that when confined to the grey matter perfusion deficit, intracellular pH, but not cerebral blood flow, differs between tissue that infarcted and tissue that survived (Harston et al., 2015). Although embolic stroke model may better mimic pathophysiological events in acute stroke patients, it is more variable than filament stroke model. Because our study aims to first develop the MRAPT MRI technique to enhance the pH specificity of APT MRI for refined lesion delineation, we chose a relatively reproducible stroke model and imaged at a single time point. Built on our results that demonstrate the superiority of the proposed MRAPT MRI over the routine MTR_asym image in the stratification of heterogeneous metabolic disruption in the ischemic tissue, our future studies will investigate how the enhanced pH MRI helps to elucidate regional ischemic tissue response to recanalization and neuroprotective treatments and ultimately, translate it to the acute stroke setting (Leslie-Mazwi et al., 2016; Li et al., 2010).

6. Conclusion

Our study here shows that the proposed MRAPT MRI effectively minimized the non-acidosis related heterogeneity commonly observed in the routine pH-weighted MTR_asym map, resulting in substantially enhanced pH specificity. We found that diffusion lesion suffered more aggravated acidosis than the perfusion/diffusion lesion mismatch. Importantly, the MRAPT analysis allows semi-automatic lesion segmentation, which refines the conventional perfusion/diffusion lesion mismatch into hypoperfused and metabolically challenged penumbra (i.e., pH/diffusion mismatch) and benign oligemia with little acidification.

• ▪Magnetization transfer and relaxation-normalized APT (MRAPT) MRI is pH specific.
• ▪pH MRI shows graded metabolic disruption in acute ischemic tissue.
• ▪pH MRI delineates metabolic penumbra from benign oligemia.
• ▪pH MRI augments perfusion and diffusion imaging for refined tissue classification.