Correlations of amide proton transfer-weighted MRI of cerebral infarction with clinico-radiological findings

Objective To clarify the relationship between amide proton transfer-weighted (APTW) signal, which reflects intracellular pH, and clinico-radiological findings in patients with hyperacute to subacute cerebral infarction. Materials and methods Twenty-nine patients (median age, 70 years [IQR, 54 to 74]; 15 men) were retrospectively examined. The 10th, 25th, 50th, 75th, and 90th percentiles of APTW signal (APT10, APT25, APT50, APT75 and APT90, respectively) were measured within the infarction region-of-interest (ROI), and compared between poor prognosis and good prognosis groups (modified Rankin Scale [mRS] score ≥2 and mRS score <2, respectively). Correlations between APTW signal and time after onset, lesion size, National Institutes of Health Stroke Scale (NIHSS) score, mRS score, and mean apparent diffusion coefficient (ADC) were evaluated. Results The poor prognosis group had lower APT50, APT75, and APT90 than the good prognosis group (–0.66 [–1.19 to –0.27] vs. –0.09 [–0.62 to –0.21]; –0.27 [–0.63 to –0.01] vs. 0.31 [–0.15 to 1.06]; 0.06 [–0.21 to 0.34] vs. 0.93 [0.36 to 1.50] %; p <0.05, respectively). APT50 was positively correlated with time after onset (r = 0.37, p = 0.0471) and negatively with lesion size (r = –0.39, p = 0.0388). APT75 and APT90 were negatively correlated with NIHSS (r = –0.41 and –0.43; p <0.05, respectively). APT50, APT75 and APT90 were negatively correlated with mRS (r = –0.37, –0.52 and –0.57; p <0.05, respectively). APT10 and APT25 were positively correlated with mean ADC (r = 0.37 and 0.38; p <0.05, respectively). Conclusion We demonstrated correlations between APTW signals of infarctions and clinico-radiological findings in patients with hyperacute to subacute infarctions. The poor prognosis group had a lower APTW signal than the good prognosis group. APTW signal was reduced in large infarctions, infarctions with low ADC, and in patients with high NIHSS and mRS scores.

Introduction Ischemic stroke is a leading cause of various neurological sequelae or death throughout the world [1]. A critical reduction of cerebral blood flow leads to a failure in normal cerebral oxygen and glucose metabolism, which causes a shift to anaerobic glycolysis and results in the accumulation of lactic acid and a concomitant decrease in intracellular pH [2]. The intracellular pH of brain tissue varies according to metabolic energy status after stroke [3]. Thus, tissue pH may serve as a potential surrogate biomarker to reflect the metabolic state and tissue damage of ischemic brain tissue [4]. Previous phosphorus-31 MR resonance spectroscopy studies have revealed that early acidosis and subacute alkalosis occurs during ischemic stroke [5,6]. However, the clinical application of phosphorus-31 MR spectroscopy to brain ischemia is limited by its low spatial and temporal resolution, and the requirement for special coils which are not available in most hospitals [7]. Thus, novel imaging methods to evaluate the changes of tissue pH after ischemic stroke in the clinical setting are required.
Amide proton transfer-weighted (APTW) MRI is a chemical exchange saturation transfer (CEST) imaging method [8]. In the past two decades, APTW MRI has been employed to evaluate the changes of mobile abnormal proteins and peptides in brain tumors [9,10]. The CEST effect due to amide protons, referred to as the amide proton transfer ratio (APTR), highly depends on tissue pH as shown in the following formula: where κ is proton exchange rate, which is the pH-dependent term as shown in Eq (2); T 1 is the T1 relaxation time of water; and t sat is the duration of saturation [11]. Zhou et al. were the first to observe the pH-sensitive amide proton transfer effects in an animal model of brain ischemia in 2003 [12]. Since then, there have been increasing number of animal studies in which pHsensitive APTW imaging has shown promise in detecting ischemic tissue acidosis following impaired aerobic metabolism [13,14]. Despite these promising results, its translation to clinical use is still limited, and there have been only a few clinical studies using APTW imaging for ischemic stroke [15][16][17].
Clarifying the relationship between APTW signal changes and clinical information should enhance our understanding of the pathophysiology of ischemic stroke, and may benefit its future treatment strategies. Therefore, the purpose of this study was to clarify the relationship between APTW imaging characteristics and the clinico-radiological findings in patients with hyperacute to subacute cerebral infarction.

Participants
The institutional clinical research ethics committee of Kyushu University approved the study protocol (IRB Clinical Research number 2019-233). The requirement for informed consent was waived due to the retrospective nature of the study. The inclusion criteria were as follows: (a) patients who underwent MRI at our institution with clinical suspicions for acute or subacute infarctions (<3 weeks) from August 2014 to September 2019. The exclusion criteria were as follows: (a) patients who could not undergo APTW imaging because of clinical demand and availability of MRI unit, (b) lesions less than 10 mm in diameter on the transverse diffusion-weighted image (DWI), (c) insufficient image quality, (d) patient's age <18 years, (e) patients who received endovascular therapy for brain infarction before MRI, (f) hemorrhagic infarction, and (g) other concurrent brain disorders.
Twenty-nine patients (median age, 70 years; interquartile range [IQR], 54 to 74 years) including 15 males (median age, 71 years; IQR, 59 to 75) and 14 females (median age, 63.5 years; IQR, 42.8 to 73.3) met the inclusion and exclusion criteria (Fig 1). In all patients, the clinical subtypes of infarctions were classified according to the TOAST stroke subtype classification system [18]. The time after onset was defined as the duration of symptom onset to MRI. The patients' pre-stroke functional ability was assessed using the pre-stroke modified Rankin Scale (mRS) score [19]. Clinical severity was assessed with the National Institute of Health Stroke Scale (NIHSS) score when patients were admitted to the hospital [20]. The patients' functional outcome was assessed using the mRS score [21] at the end of the follow-up period, at least four weeks after symptom onset. Pre-stroke mRS scores, NIHSS scores on admission, and mRS scores at the chronic phase were determined according to the medical records by a board-certified stroke physician (Y.W., 23 years of experience in stroke medicine).

MRI data acquisition
MRI was conducted on a 3.0-T MRI system (Achieva TX, Philips Medical Systems, Best, the Netherlands) with an 8-channel head coil.
To correct the B0 inhomogeneity of the APTW MRI, a B0 map was acquired separately using the same imaging geometry and spatial resolution as the APTW MRI via a 2D gradientecho sequence with the following parameters: repetition time = 15 ms, echo time = 8.1 ms, flip angle = 30˚, dual echo (ΔTE = 1 ms), number of acquisitions = 16, scan time = 33 sec.

Data processing
All image processing was executed using Image J (version 1.48, National Institutes of Health, Bethesda, MD). To assess the APTR, we obtained the magnetization transfer ratio asymmetry (MTR asym ) on a pixel-by-pixel basis from a Z-spectrum plot of the water signal attenuation versus offset saturation frequency (offset = 0 for water resonance). Because the saturation frequency of amide protons is primarily around +3.5 ppm [12], the APTR was assessed by calculating the MTR asym at ±3.5 ppm according to the following formula [12]: where S sat and S 0 were the signal intensities obtained with and without selective radiofrequency saturation pulse irradiation, respectively. MTR' asym (3.5 ppm) represents the upfield nuclear Overhauser enhancement (NOE) effect of various non-exchangeable protons [22]. Eq (3) shows that MTR asym (3.5 ppm) contains not only the APTR, but also contributions from the NOE effect. Thus, we refer to MTR asym (3.5 ppm) as the APTW signal (%), and MTR asym (3.5 ppm) maps as APTW images.

Qualitative analysis of APTW images
As a qualitative analysis, the visibility of infarctions on APTW images was independently evaluated by two board-certified neuroradiologists (observer 1, D.M., 5 years of experience; observer 2, O.T., 19 years of experience) using a three-level visual grading scale as follows: clear, the infarction can be clearly identified as a hypointense lesion on an APTW image without reference to DWI; moderate, an infarction can be recognized as a hypointense lesion on APTW images, but it is difficult to identify the lesion without reference to DWI; unclear, a lesion cannot be recognized on an APTW image even with reference to DWI.
The inter-rater agreement for the visibility between the two observers was evaluated using Cohen's kappa coefficient [23].

Quantitative analysis of APTW images
The quantitative analysis was performed by a board-certified neuroradiologist (D.M., 5 years of experience). A region-of-interest (ROI) was manually drawn to include an entire hyperintense lesion on DWI; this ROI was then copied and pasted onto the APTW image. Another ROI was placed on the contralateral normal-appearing white matter (CNAWM).
We obtained the 10th, 25th, 50th, 75th, and 90th percentiles of the histogram for the APTW signal (%) within the ROIs (APT 10 , APT 25 , APT 50 , APT 75 , and APT 90 , respectively); the lowest 10% of the signal within an ROI was found below APT 10 , and the highest 10% was found above APT 90 . Histogram analysis is a quantitative technique has been used in previous CEST studies, e.g. Liu et al. applied a histogram distribution-based analysis for CEST signals of a mouse model of multiple sclerosis [24].
The mean apparent diffusion coefficient (ADC) (mm 2 /sec) within the infarction ROI was also obtained. Lesion size was measured as the maximum diameter of the infarction on the transverse DWI.
For the qualitative analyses, the APTW signal percentiles within the infarction ROI were compared to those within the CNAWM ROI using the Wilcoxon signed-rank test. Demographic and baseline clinical characteristics and the APTW signal percentiles within the infarction ROI were compared between cardioembolic infarctions and non-cardioembolic infarctions, and between the poor prognosis group (mRS score �2) and the good prognosis group (mRS score <2), using the Chi-squared and Mann-Whitney U tests. The correlations between the APTW signal percentiles within the infarction ROI and the time after onset, lesion size, NIHSS score on admission, mRS score at the chronic phase, and mean ADC within the infarction ROI were evaluated using Spearman's rank correlation coefficient.
The statistical analyses were performed using JMP Pro 14.2.0 (SAS Institute, Cary, NC) under supervision of a statistician (J.K.). P-values <0.05 was considered statistically significant.

Quantitative analysis of APTW images
The APTW signal percentiles within the infarction and CNAWM ROIs are presented in Table 1. The APT 10 , APT 25 , and APT 50 of the infarctions were significantly lower than those of the CNAWM (median APT 10  The baseline clinical characteristics and the percentiles in the cardioembolic and other subtypes of cerebral infarction are presented in Table 2. There were no significant differences in the clinical characteristics and APTW signals within the infarction ROI between the two groups (p >0.05). Table 3 shows the baseline clinical characteristics and percentiles of the two groups divided according to their mRS scores at the chronic period. The baseline clinical characteristics were not significantly different between the poor and good prognosis groups. The APT 50 , APT 75 , and APT 90 within the infarction ROI were significantly lower in the poor prognosis group than those in the good prognosis group (median APT 50 , -0.66 [IQR, -1. The correlations between APTW signal within the infarction ROI and time after onset, lesion size, NIHSS score, mRS score, and mean ADC are summarized in Table 4. The APT 50 was positively correlated with time after onset (r = 0.37, p = 0.0471), and inversely correlated with lesion

Sensitivity analysis of study participants without pre-stroke dependency
The sensitivity analysis of a subpopulation of patients without pre-stroke dependency (n = 18, pre-stroke mRS≦1) showed the same trend for an association between the APTW signal within the infarction ROI and the stroke outcome (see details in S1 Appendix): the APT 50 , APT 75 , and APT 90 were significantly lower in the poor prognosis group than those in the good prognosis group (median APT  Table 3. Demographic and baseline clinical characteristics and percentiles of APTW signal of the poor prognosis group with mRS score �2 (n = 21) and good prognosis group with mRS score <2 (n = 8).    S sat (ppm) and S 0 , the signal intensities obtained with and without selective radiofrequency saturation pulse irradiation, respectively; APT 10 , APT 25 , APT 50 , APT 75 , and APT 90 , correspond to the 10th, 25th, 50th, 75th, and 90th percentiles of APTW signal value, respectively.

Discussion
In this study, we observed a significantly reduced APTW signal in the infarct area compared to that in the CNAWM, in agreement with previous studies [12,13,17]. As shown in Eqs (1) and (2), the APTR depends on not only tissue pH, but also T1 relaxation time, water content, and amide proton content. However, it has been previously reported that APTR changes are primarily determined by pH changes [12]. In addition, a previous study using phosphorus-31 MR spectroscopy has demonstrated that pH abnormalities of the infarct area remain, even in the subacute phase [6]. T1 relaxation time and water content of the infarct area have been shown to both increase over time during the acute to subacute phase due to cytotoxic edema [25]; as the T1 relaxation time has a strong positive correlation with the water content in the infarct area [26], the two changes may cancel each other out to some extent. There has been no previous study focusing on dynamic changes of the amide proton content within the infarction. However, based on a study investigating amide proton content in peritumoral brain edema [27], we speculate that amide proton content might decline in infarctions with cytotoxic edema, which would affect the APTR.
As shown in Eq (3), the APTW signal includes not only the APTR but also MTR' asym (3.5 ppm), which is attributed to the upfield NOE effect [22]. However, a recent quantitative CEST study using animal models confirmed that the APTR accounts for a greater part of the MTR asym (3.5 ppm) changes than the NOE effect [28]. Collectively, this suggests that the APTW signal would primarily depend on tissue pH of the infarction, even in the subacute phase. Therefore, the reduction in APTW signal in the infarction observed in the present study likely reflects the reduced CEST effect of amide protons due to pH reduction.
The APTW signal within the infarction ROI was significantly lower in the poor prognosis group than in the good prognosis group, which was similar to the findings of Lin et al. [29]. Although infarct volume could be a prognostic factor for patients with cerebral infarction [30], there were no significant differences in the lesion size between the two groups. Thus, the APTW signal of the infarct area might be a new prognostic factor, independent from a lesion size.
Our analyses showed that the APTW signal within the infarction was positively correlated with the time after onset. This might reflect the alleviation of tissue acidification that occurs over time after the onset, as previously demonstrated in both animal [13] and clinical [17] studies. This result is consistent with previous studies using phosphorus-31 MR spectroscopy [5,6]. The increase in pH after stroke could be explained by active compensatory mechanisms within the ischemic tissue, such as microglial invasion and alternate buffering mechanisms [6].
The APTW signal within the infarction was inversely correlated with the lesion size. We assume that larger infarctions could have a smaller surface area to volume ratio, which would reduce gas-diffusion from surrounding area and result in severe metabolic dysfunction.
The APTW signal within the infarction was significantly inversely correlated with both the NIHSS score on admission and the mRS score at the chronic phase. Lin et al. [29] have previously shown that the difference of the APTW signal between the acute ischemic region and the contralateral side is linearly correlated with both the NHISS score and mRS score. Thus, these results indicate that the APTW signal may be a useful tool to assess stroke severity and predict outcome.
A previous study has demonstrated that a high pre-stroke mRS score is significantly associated with poor functional outcomes at the chronic phase [19]. Thus, we were concerned that the pre-stroke mRS score could be a confounding factor when evaluating the relationship between the APTW signal and mRS score at the chronic phase. Therefore, we conducted an additional sensitivity analysis of a patient subpopulation comprising only those with low prestroke mRS scores (pre-stroke mRS �1), to exclude the influence of a high pre-stroke mRS score. This sensitivity analysis showed that APTW signal within the infarction in the poor prognosis group was significantly lower than that in the good prognosis group, and was inversely correlated with the mRS score at the chronic phase, which agreed with the main results. These findings suggest that a low APTW signal within the infarction could predict poor outcome, regardless of pre-stroke mRS score.
We found that the APTW signal was positively correlated with ADC in infarctions, indicating that the pH reduction may correlate with the severity of cytotoxic edema. One explanation for this is that acid-sensing ion channels, which are hydrogen ion-gated cation channels that are permeable to Na + , are activated as pH falls but are inactive at physiological pH [31]. Thus, pH reduction would promote intracellular Na + accumulation via the activation of those channels, worsening the cytotoxic edema.
The two observers judged 11 and 12 out of the 29 infarctions as unclear on the visual grading scale. This may reflect an insufficient sensitivity to pH changes in the protocol used in this study. Concomitantly, undetected lesions might have less tissue acidosis compared to the detected lesions. Optimization of the acquisition parameters, application of a fitting method [32], correction with T1-relaxation time [33,34] and reduction of motion effect [35] would be helpful to improve the sensitivity of APTW images to tissue pH.
There were no significant differences in the APTW signals within the infarctions between cardioembolic infarctions and the other subtypes. The pH reduction might be less influenced by clinical subtypes compared to other factors such as lesion size or time after onset. Since the number of patients in each group was small, the effect of clinical subtype should be evaluated in a larger patient population.
Our study had several limitations. First, only a limited number of the records of patients with suspected acute infarctions included APTW MRI. When patients are treated for infarctions, the opportunity to perform new treatment and assessment protocols is limited. Therefore, the sample size was small (n = 29) in this study period. Second, although we excluded cases with intense participant motion, it remains possible that participant motion negatively affected the image quality. Third, we determined pre-stroke mRS scores, NIHSS scores on admission, and mRS scores at the chronic phase retrospectively with a stroke physician supervising to avoid inaccurate scoring. Fourth, to date, the optimal saturation pulse strength for APTW imaging has not been established, and saturation pulse strengths of 0.5-3 μT have been used in previous studies [14-16, 36, 37]. We selected a saturation power of 1.5 μT based on a previous animal study that reported the relationship between APTW signal and tissue lactic acidosis [14]. However, a human study has demonstrated that a saturation power of either 2 or 3 μT can result in a better APTW image contrast between the infarction and the CNAWM than that of 1 μT [15]. Considering these facts, we feel that the saturation pulse of 1.5 μT used in the present study was acceptable; however, slightly stronger saturation powers might have improved the APTW image contrast.
In conclusion, we demonstrated correlations between the APTW signal of infarctions and the clinico-radiological findings in patients with hyperacute to subacute infarctions. The poor prognosis group had a lower APTW signal than good prognosis group. APTW signal was reduced in large infarctions, infarctions with low ADC, and in patients with high NIHSS and mRS scores.