Chemical exchange saturation transfer imaging in hepatic encephalopathy.

Hepatic encephalopathy (HE) is a common complication in liver cirrhosis and associated with an invasion of ammonia into the brain through the blood-brain barrier. Resulting higher ammonia concentrations in the brain are suggested to lead to a dose-dependent gradual increase of HE severity and an associated impairment of brain function. Amide proton transfer-weighted (APTw) chemical exchange saturation transfer (CEST) imaging has been found to be sensitive to ammonia concentration. The aim of this work was to study APTw CEST imaging in patients with HE and to investigate the relationship between disease severity, critical flicker frequency (CFF), psychometric test scores, blood ammonia, and APTw signals in different brain regions. Whole-brain APTw CEST images were acquired in 34 participants (14 controls, 20 patients (10 minimal HE, 10 manifest HE)) on a 3 T clinical MRI system accompanied by T1 mapping and structural images. T1 normalized magnetization transfer ratio asymmetry analysis was performed around 3 ppm after B0 and B1 correction to create APTw images. All APTw images were spatially normalized into a cohort space to allow direct comparison. APTw images in 6 brain regions (cerebellum, occipital cortex, putamen, thalamus, caudate, white matter) were tested for group differences as well as the link to CFF, psychometric test scores, and blood ammonia. A decrease in APTw intensities was found in the cerebellum and the occipital cortex of manifest HE patients. In addition, APTw intensities in the cerebellum correlated positively with several psychometric scores, such as the fine motor performance scores MLS1 for hand steadiness / tremor (r = 0.466; p = .044) and WRT2 for motor reaction time (r = 0.523; p = .022). Moreover, a negative correlation between APTw intensities and blood ammonia was found for the cerebellum (r = -0.615; p = .007) and the occipital cortex (r = -0.478; p = .045). An increase of APTw intensities was observed in the putamen of patients with minimal HE and correlated negatively with the CFF (r = -0.423; p = .013). Our findings demonstrate that HE is associated with regional differential alterations in APTw signals. These variations are most likely a consequence of hyperammonemia or hepatocerebral degeneration processes, and develop in parallel with disease severity.

Introduction

Liver cirrhosis is known to manifest in systemic effects. In particular, the most common neurological manifestation of liver cirrhosis is hepatic encephalopathy (HE), which comprises a variety of symptoms. Patients suffering from HE exhibit alterations in cognitive and motor function as well as behavioral changes. The clinical symptoms of HE vary with disease severity, starting with mild attentional deficits and disorientation, cognitive deterioration, and disturbed motor control. They may develop into somnolence and stupor and in the most severe case hepatic coma (Butterworth, 2000; Felipo, 2013; Ferenci et al., 2002; Prakash and Mullen, 2010).

The pathophysiology of HE is not finally understood, but is assumed to be multifactorial (Cichoz-Lach and Michalak, 2013; Felipo, 2013; Häussinger and Sies, 2013). A key element in the pathophysiology is the invasion of ammonia into the brain through the blood-brain barrier. Previous studies using 13NH3-PET imaging have revealed an increased uptake of circulating ammonia into the brain of cirrhosis patients (Keiding et al., 2006; Lockwood et al., 1991).

In addition to increased oxidative stress caused by ammonia accumulation (Norenberg et al., 2005), an excess of glutamine is produced in the astrocytes as a result of ammonia detoxification by glutamine synthetase. The increased glutamine concentration in the astrocytes triggers cell swelling via osmotic imbalance, finally leading to alteration of brain water homeostasis and emergence of a low-grade edema (Detry et al., 2006; Häussinger and Schliess, 2008). The emergence of a low-grade edema was further investigated by different studies using advanced MR imaging: Quantitative T1 mapping and semi-solid magnetization transfer (MTC) imaging studies attributed increased T1 values (Shah et al., 2003) and alterations in the calculated MTC effect (Miese et al., 2006) to increased water content in HE patients. Additionally, a quantitative water mapping approach has demonstrated a small increase of water content in white matter areas (Shah et al., 2008), while these findings remained absent in another quantitative water mapping study including patients with low-grade HE (Oeltzschner et al., 2016). In general, T1-weighted (Butterworth et al., 1995; Klos et al., 2006; Pujol et al., 1993; Rovira et al., 2008) or quantitative T1 (Shah et al., 2003) imaging may be altered by increased water content. However, T1 changes especially within the basal ganglia are more likely to be mediated by manganese deposition, which is a common neurotoxin in HE (Rose et al., 1999).

All MR visible effects described above are based on the notion of ammonia accumulation in the patients' brains. Therefore, it is of paramount interest to measure ammonia in a most direct way, without the use of radiation, and with increased resolution compared to 13NH3-PET imaging.

Chemical exchange saturation transfer (CEST) represents a suitable tool for the assessment of changes of in vivo ammonia levels. CEST provides an advanced MRI contrast depending on diluted labile protons, which are usually undetectable by conventional MRI. The CEST contrast is based on the mitigation of the bulk water signal due to magnetization transfer between the bulk water and frequency-selectively saturated labile protons (Wolff and Balaban, 1989). Amide proton transfer-weighted (APTw) imaging, which is based on magnetization transfer from exchangeable amide protons of mobile tissue proteins and peptides (Jones et al., 2012), is an emerging field in CEST imaging. It allows several applications, such as amide proton quantification, detection of pH changes in the amide proton environment (Mori et al., 1998; van Zijl et al., 2003; Zhou et al., 2003a), and measurement of global in vitro protein folding (Goerke et al., 2017, Goerke et al., 2015) with possible applications in neurodegenerative diseases. Additionally, CEST depends on non-exchange-related factors, such as direct water saturation, MTC, and water longitudinal relaxation time (Zu, 2018). Thus, these factors have to be included in the interpretation of possible HE related changes in APTw imaging.

In our recent in vitro study, we were able to link increasing ammonia concentrations to a decreasing APTw CEST signal through protein denaturation (Zöllner et al., 2018). We could also apply this finding to example cases of HE patients. The present study aimed to systematically investigate the sensitivity of APTw imaging to HE-related brain ammonia levels, and to gauge its potential to monitor disease severity and progression via the HE-related signal alteration in APTw imaging. To this end, APTw imaging was performed in a cohort of clinically well-characterized HE patients in different grades of severity, as well as healthy age-matched controls. To investigate the relationship between disease severity and APTw imaging, the critical flicker frequency (CFF), psychometric testing scores, and blood ammonia levels were assessed. In addition, T1 maps were acquired to control for T1 effects on the measured MTRasym values by T1 normalization in the region-of-interest analysis.

Material and methods

Participants and HE grading

The study was performed in accordance with the Declaration of Helsinki in its current version (“), and was approved by the local institutional review board (study number 5179R). All participants gave written informed consent prior to the examination. 15 controls and 20 patients with clinically confirmed HE, graded as minimal HE (mHE) and manifest HE (HE), were examined. Exclusion criteria for both patients and controls included neurological or psychiatric diseases excluding the diagnosis of HE for the patient group, severe intestinal diseases, the use of any medication acting on the central nervous system and diagnosed peripheral/retinal neuropathy. If alcohol abuse was part of the medical history, the patient had to remain abstinent for ≥4 weeks prior to examination. The grading was performed in line with the West-Haven criteria (Ferenci et al., 2002; Kircheis et al., 2002). In addition, neuropsychological testing, critical flicker frequency (CFF) assessment with portable CFF goggles (NEVOlab, Maierhöfen, Germany), and blood sample tests were carried out. In the CFF test, the tested individual has to indicate with a button press when the impression of a virtual light source in 12 m distance changes from a fused light source to a flickering one. After a short training session, the CFF is assessed eight times, and the mean and standard deviation are calculated for further analysis. The CFF was included as it has been shown to be a reliable clinical parameter for HE monitoring, and accounts for the continuous nature of symptom severity (Kircheis et al., 2014, Kircheis et al., 2002). One control had to be excluded due to a CFF value <39 Hz, which is the cut-off frequency for mHE. The remaining study population is summarized in Table 1.

Table 1

Remaining study cohort.

	N	Sex [M/F]	Age [years]Mean ± SD	CFF [Hz]Mean ± SD	Etiology of liver cirrhosis
Controls	14	8/6	60.1 ± 5.3	44.2 ± 3.4	–
mHE	10	7/3	55.1 ± 10.4	41.0 ± 3.2⁎	5 ALC, 1 HCV, 1 NASH, 1 OS, 2 U
HE	10	6/4	59.5 ± 6.7	36.4 ± 2.3⁎⁎	5 ALC, 2 HCV, 2 NASH, 1 OS

Significantly different from controls (p < .05) with non-parametric Wilcoxon rank sum test.

Significantly different from controls (p < .01) and mHE (p < .01) with non-parametric Wilcoxon rank sum test (ALC alcoholic, HCV hepatitis C virus, NASH non-alcoholic steatohepatitis, OS overlap syndrome, U unknown).

HE severity was determined by an experienced clinician and included a clinical assessment regarding the mental state and consciousness of the patients and neuropsychometric testing. Computer-based neuropsychometric tests from the Vienna Test System (Dr. Schuhfried GmbH, Mödling, Austria) consisted of five test batteries and reported 22 age-validated scores (percentile rank values in comparison to an age-matched control cohort) reflecting motor and cognitive performance. Better performance was graded with higher scores. The parameter values were considered as abnormal in case of a percentile rank lower than 15.9 compared to the control cohort implemented in the test system. Patients without clinical symptoms of manifest HE, but with >1 abnormal psychometric test result were classified as mHE (Kircheis et al., 2002). Selected scores were incorporated for further analysis, including cognitive (COG1: time to reject a geometric shape not matching control shapes; COG2: time to confirm a geometric shape matching a control shape), fine motor performance (line following test: LVT1: time per item; LVT2: overall score), motor/precision/speed (MLS1: hand steadiness/tremor; MLS2: arm/hand precision; MLS3: arm/hand speed; MLS4: finger tapping speed), and reaction performance (WRT1: reaction time; WRT2: motor reaction time) scores.

MR measurements

All MR investigations were performed on a clinical whole-body 3 T MRI (Siemens MAGNETOM Trio A TIM System, Siemens Healthcare AG, Erlangen, Germany) using a 12-channel head coil for receive, and the internal body coil for transmit.

Structural MRI

Following a scout scan, a high-resolution 3D anatomical transversal T1-weighted magnetization prepared gradient echo (MP-RAGE) scan (TR/TE = 1950/4.6 ms; isotropic resolution of 1 mm; 176 slices) was performed aligned to the AC-PC line.

CEST MRI

The CEST images were acquired using a prototype transversal 3-dimensional gradient echo sequence (TR/TE = 1390/3.06 ms; flip angle = 10°; FoV = (230 × 230) mm2; matrix size 192 × 192; 24 slices; slice thickness 5 mm; gap 1 mm) to obtain full brain coverage. 22 equidistant frequency offsets were sampled between −5 and 5 ppm using a pulse train of 5 Gaussian shaped pulses (pulse duration 100 ms; inter pulse delay 100 ms, B1 = 1.5 μT) for saturation.

The unsaturated S0 image was acquired by turning off the saturation pulse. Additionally, data for B0 inhomogeneity correction with a water saturation reference map (WASSR) (Kim et al., 2009) were recorded.

GRE MRI

Two 3-dimensional gradient echo scans with different flip angles (FA = 4°, 15) were acquired with the same spatial dimensions as the CEST images. Relative B1 maps were created by heavily smoothing the low flip angle (FA = 4°) volume (Sabati and Maudsley, 2013). In addition, T1 maps were calculated using the volumes (FA = 4°, 15°) (Sabati and Maudsley, 2013) leading to a measurement time of 25 min in total.

Data processing

Motion correction and brain masking

Motion correction was performed using the co-registration method of the MATLAB (MathWorks Inc., Natrick, MA, USA) toolbox SPM12 (Friston, 2007). The 3D volumes of the CEST datasets of each saturation frequency were co-registered to the 3D volume of the 3.5 ppm saturation frequency, as described by Zhang et al. (Zhang et al., 2016). Subsequently, all WASSR datasets were co-registered to the S0 volume. Tissue compounds and cerebrospinal fluid (CSF) were segmented with the SPM newSegment function, and a brain mask was created, which included all pixels with probabilities >0.8 for grey matter (GM), white matter (WM), and CSF.

MTRasym evaluation

Z-spectra and magnetization transfer ratio asymmetry (MTRasym) images were calculated for all pixels within the brain mask using an in-house written MATLAB script. All images were smoothed using a nonlocal means (NLM) filter to improve the signal-to-noise ratio, and to avoid blurring of the images (Yuan et al., 2016). The used filter is a MATLAB reimplementation (https://de.mathworks.com/matlabcentral/fileexchange/52018-simple-non-local-means-nlm-filter) of the NLM filter described by Manjón et al. (Manjón et al., 2008). The filtered pixel value is calculated by the weighted average of all pixels in the image. The weights originate from the similarity between the user-defined neighborhood of the filtered pixel and are defined by an exponential decay control parameter h and the Gaussian weighted Euclidian distance of the pixels in the neighborhood. The parameters within the MATLAB toolbox were chosen as follows: search window t = 3, similarity window f = 2, standard deviation of the Gaussian low-pass h1 = 0.01, decay control parameter h2 = 18 = 1.18σ with σ being estimated from the noise standard deviation of the background of all images. The factor 1.18 is suggested in the original publication to be optimal for proton-density weighted images. More details are described in Manjón et al. (Manjón et al., 2008).

The Piecewise Cubic Hermite Interpolating Polynomial (pchip) algorithm was chosen in the interp1 MATLAB to interpolate the spectra in steps of 0.05 ppm. Asymmetry analysis (Zhou and van Zijl, 2006) was calculated as follows:

With labelled proton scan Zlab, a reference scan at the opposite frequency Zref and the unsaturated image S0.

MTRasym curves were calculated over a frequency range between 0 and 5 ppm within the z-spectra. Subsequently, the APT-weighted MTRasym maps were calculated by averaging over a frequency range from 3 to 4 ppm and it will be referred to as APTw imaging. B1-one-point-contrast correction is performed as described in detail by Windschuh et al. (Windschuh et al., 2015). Additionally, T1 normalization of the MTRasym was performed in the region-of-interest (ROI) analysis to correct for a T1 relaxation effects (Zhou et al., 2018). It will be referred to as APTwT1 MTRasym.

Normalization

Data normalization was performed with the open-source software package Advanced Normalization Tools (ANTs) (Avants et al., 2009). The package includes algorithms for bias correction, template construction, and image normalization. A template from the current study population was created with the ANTSNormTemp.sh script using the structural images (Avants et al., 2011). The script uses an iterative approach including two steps: At first, all individual structural images are spatially transformed onto one reference image. Initially, an average of all structural images was used as reference and 15 iterative steps of affine transformations were performed to align the volumes. In the second step, the inverse transformation matrix is applied to the reference image to update its shape. These two steps are applied iteratively, starting with the updated reference image of the previous iteration. Subsequently, 15 diffeomorphism transformations were performed iteratively with a Greedy Syn algorithm to create the template. The maximum iterations (parameter -m) within the registrations were as followed: Iterations 1 to 5 with -m 30 × 90 × 40, iterations 5 to 10 with -m 60 × 180 × 80 × 45 and iteration 10 to 15 with -m 120 × 360 × 160 × 90. The step size of the image registrations in every sub iteration (−m) decreased with each level. Afterwards, all individual scans were normalized to the template using the diffeomorphism transformation, and the transformation matrix was applied to the CEST and T1 maps. Thus, all individual CEST and T1 maps could be assessed with an atlas-based statistical analysis.

Statistical evaluations

Atlas-based analysis

The APTw images and T1 maps were analyzed by an atlas-based approach, using the Neuromorphometrics atlas integrated in SPM12 (Friston, 2007). The atlas was co-registered to the cohorts template obtained from ANTs. A set of 6 regions including both hemispheres at once was included in the analysis. The regions are summarized in Table 2 and the selection was based on their suggested involvement in HE pathophysiology as referenced in previous studies (Cauli et al., 2009; Kril et al., 1997; Oeltzschner et al., 2016; Rovira et al., 2008; Shah et al., 2003, Shah et al., 2008). Group level differences (control, mHE, HE) in APTwT1 MTRasym were investigated with a pairwise non-parametric Wilcoxon rank sum test. Group differences were considered significant for p < .05.

Table 2

Atlas based statistics of APTwT1 MTRasym in HE (minimal HE (mHE), manifest HE (HE)).

	T1 normalized MTRasym (%)Mean (± SD)			Correlation with
	Control	Patients		CFF	Blood ammonia	GGT
		mHE	HE	rp	rp	rp
Cerebellum	0.94(0.48)	0.92(1.16)	0.61#(0.75)	0.0430.807	−0.6150.007	−0.2510.300
Occipital cortex	0.82(0.60)	0.88 (0.50)	0.39*(0.50)	0.1020.564	−0.4780.045	−0.2670.269
Putamen	0.61(0.11)	0.76*(0.09)	0.81(0.24)	−0.4230.013	−0.2850.252	−0.3930.096
Thalamus	0.43(0.08)	0.45(0.08)	0.48(0.18)	−0.3230.062	−0.3650.137	−0.4540.049
Caudate	0.35(0.12)	0.37(0.13)	0.42(0.16)	−0.2000.257	−0.2210.379	−0.0080.975
White matter	0.49(0.08)	0.51(0.12)	0.57(0.14)	−0.2310.190	−0.4090.137	−0.0480.847

Asterisks indicate significant differences from controls (* = p < .05). Pound signs indicate significant differences between patient groups (# = p < .05). Bold numbers indicate significant correlations over all participants. Reduced APTwT1 MTRasym and its correlation with blood ammonia in cerebellar and occipital regions indicate either increased ammonia accumulation or hepatocerebral degeneration. Increased putaminal APTwT1 MTRasym are related to other contrast mechanisms, such as strong alterations in metabolite concentrations.

Relationships between CFF and APTwT1 MTRasym, blood ammonia and APTwT1 MTRasym, and between psychometric scores and APTwT1 MTRasym were compared with a bivariate two-sided Pearson correlation test. Correlations were considered significant in case p < .05. No multiple comparison correction was employed in the analysis.

All statistical analyses were performed using IBM SPSS Statistics for Windows, Version 24.0 (IBM Corp., Armonk, NY, USA).

Results

All 34 MRI datasets (14 controls, 10 mHE, 10 HE) remained in the final analysis. The manifest HE group included 9 HE I patients and one patient graded as HE II. One HE I patient did not undergo psychometric testing, and two patients did not undergo blood testing (1 mHE, 1 HE).

APTw atlases

Mean group APTw atlases are depicted in Fig. 1. The atlases indicate decreasing APTw MTRasym in the cerebellum and the occipital region of the manifest HE patients. In addition, slight alterations in deep grey matter regions are occurring, as APTw MTRasym increases within the Putamen. An absence of changes is visible for caudate, thalamus, and the selected white matter region.

Fig. 1

Mean group atlases of APTw MTRasym in healthy controls and HE patients. The rows depict three representative slices including the six selected regions of interests (ROI) (Red: Cerebellum; Yellow: Occipital cortex; Cyan: Putamen; Magenta: Thalamus; Green: Caudate; Blue: White matter). The template of the current study population is depicted in the first column. ROI are illustrated in the second column. The last three columns display the mean atlases of the APTw imaging of each group (control, minimal HE (mHE), manifest HE (HE)). The signal reduction could either be linked to increased ammonia accumulation or hepatocerebral degeneration.

Supplementary Fig. 1 represents the mean APTw atlases, as well as the corresponding slices of mean T1 atlases. To avoid any contamination of APTw effects by water longitudinal relaxation, a T1 normalization was performed in the ROI based analysis. Group results and correlations with critical flicker frequency (CFF), blood ammonia, and gamma-glutamyltransferase (GGT) concentration are summarized in Table 2. The ROI based analysis is summarized in the following sections. The figures are focused on the ROI with significant differences in the measured APTwT1 MTRasym.

Cerebellum

Mean cerebellar APTwT1 MTRasym values were reduced in HE patients compared to mHE patients (p < .05; Fig. 2a) indicating higher levels of ammonia or hepatocerebral degeneration in the patients with manifest HE compared to minimal HE. No differences were found between controls compared to both patient groups. A negative correlation of cerebellar APTw imaging blood ammonia levels (r = −0.615; p = .013; Fig. 2b) and a positive correlation with the psychometric MLS1 (hand steadiness / tremor) score (r = 0.466; p = .044; Fig. 2c) were found. Furthermore, a positive correlation of the psychometric WRT2 score and mean APTwT1 MTRasym could be observed (r = 0.523; p = .022; Fig. 2d).

Fig. 2

APTw imaging of the cerebellum in healthy controls and HE patients (minimal HE (mHE), manifest HE (HE)). (a) Group boxplots including mean APTwT1 MTRasym values. Asterisks indicate significant differences between groups (p < .05). Correlations between blood ammonia levels (b), psychometric MLS1 score (hand steadiness/tremor) (c), psychometric WRT2 score (motor reaction time) (d), and mean APTwT1 MTRasym. The data suggests a strong cerebellar involvement in HE explaining poor motor performance of HE patients, which could possibly be linked to ammonia accumulation or hepatocerebral degeneration.

Occipital cortex

A decrease in mean APTwT1 MTRasym was observed between controls and HE patients (p < .05), whereas no significant differences were detected between controls and mHE patients and between both patient groups (Fig. 3a). A negative correlation of occipital APTw imaging and blood ammonia levels (r = −0.476; p = .045; Fig. 3b) and a positive correlation with the psychometric MLS2 (arm/hand precision) score (r = 0.544; p = .016; Fig. 3c) were present. A positive correlation of occipital APTw imaging and psychometric MLS2 (arm/hand precision) score (r = 0.544; p = .016; Fig. 3c) and a negative correlation with blood ammonia levels (r = −0.476; p = .045; Fig. 3b) were present. Additionally, a positive correlation between the psychometric LVT1 score and the mean APTwT1 MTRasym was found (Fig. 3d).

Fig. 3

APTw imaging of the occipital cortex in healthy controls and HE patients (minimal HE (mHE), manifest HE (HE)). (a) Group boxplots including mean APTwT1 MTRasym values. Asterisks indicate significant differences between groups (p < .05). Correlations between blood ammonia levels (b), psychometric MLS2 score (arm/hand precision) (c), psychometric LVT1 score (time per item) (d), and mean APTwT1 MTRasym values. This might explain alterations in visual perception of HE patients with increasing diseases severity due to ammonia accumulation.

Putamen

Mean putaminal APTwT1 MTRasym were increased in mHE patients compared to controls (p < .05), whereas no differences were found between controls and HE patients and between the two patient groups (Fig. 4a). Additionally, the mean values in the HE patient group showed a greater variability. A negative correlation of CFF and APTwT1 MTRasym was observed (Fig. 4b). In contrast to the cerebellar and occipital region, the increased putaminal APTwT1 MTRasym suggests the domination of other contrast mechanism, such as increased glutamine concentrations, in this region. Alterations through increased water or the accumulation of manganese were ruled out by T1 normalization.

Fig. 4

APTw imaging of the putamen in healthy controls and HE patients (minimal HE (mHE), manifest HE (HE)). (a) Group boxplots including mean APTwT1 MTRasym. Asterisks indicate significant differences between groups (p < .05). (b) Correlation between critical flicker frequency (CFF) and mean APTwT1 MTRasym values. The data suggests the domination of other contrast mechanisms, such as metabolite concertation changes.

Psychometric testing

Results of the correlation analysis of the psychometric tests are summarized in Table 3. The motor score MLS1 and the motor reaction time score WRT2 correlated positively with the APTwT1 intensities in the cerebellum. This indicates a connection between motor deficits of HE patients and reduced APTwT1 MTRasym in the cerebellum. The line following scores LVT1 and LVT2, the motor scores MLS2 and MLS3, as well as the motor reaction time WRT2 correlated positively with the observed APTwT1 intensities in the occipital cortex.

Table 3

Correlations of psychometric test scores and mean APTwT1 MTRasym (COG1: time to reject a geometric shape not matching control shapes; COG2: time to confirm a geometric shape matching a control shape; line following: LVT1: time per item; LVT2: overall score; MLS1: hand steadiness/tremor; MLS2: arm/hand precision; MLS3: arm/hand speed; MLS4: finger tapping speed; WRT1: reaction time; WRT2: motor reaction time). Bold numbers indicate significant correlations.

	COG1	COG2	LVT1	LVT2	MLS1	MLS2	MLS3	MLS4	WRT1	WRT2
	rp	rp	rp	rp	rp	rp	rp	rp	rp	rp
Cerebellum	0.164	0.162	0.299	0.277	0.466	0.443	0.357	0.156	0.282	0.523
	0.503	0.508	0.213	0.250	0.044	0.058	0.133	0.525	0.243	0.022
Occipital cortex	0.378	0.400	0.547	0.469	0.399	0.544	0.526	−0.092	0.356	0.531
	0.110	0.089	0.015	0.043	0.091	0.016	0.021	0.708	0.019	0.019
Putamen	−0.093	−0.105	0.318	0.391	0.314	−0.034	−0.013	0.258	−0.139	0.285
	0.704	0.668	0.184	0.098	0.191	0.890	0.958	0.286	0.569	0.237
Thalamus	−0.070	−0.102	0.124	0.185	0.181	0.272	0.181	0.038	−0.056	0.188
	0.776	0.678	0.614	0.449	0.458	0.260	0.458	0.876	0.821	0.441
Caudate	−0.335	−0.381	−0.150	−0.146	0.151	−0.174	−0.301	0.384	−0.373	0.005
	0.161	0.108	0.539	0.552	0.538	0.476	0.210	0.104	0.116	0.985
White matter	−0.140	−0.170	0.038	0.114	0.236	0.356	0.108	−0.021	0.002	0.099
	0.567	0.488	0.876	0.643	0.330	0.135	0.659	0.930	0.994	0.688

Discussion

In the present in vivo CEST study, we applied chemical exchange saturation transfer imaging to investigate the link between hepatic encephalopathy, blood ammonia levels, and APTw measures which reflect cerebral ammonia levels. Our results reveal reduced APTwT1 MTRasym within the cerebellum and occipital regions correlating both with blood ammonia and psychometric scores. Additionally, APTw intensities are increased within the putamen of mHE patients. Furthermore, putaminal APTwT1 MTRasym correlates negatively with CFF.

CEST and clinical parameters in HE

Psychometric testing, CFF, and blood ammonia are common clinical parameters to assess the severity of HE, especially in cases of minimal HE, when no clinical overt signs of HE are apparent (Kircheis et al., 2014). Decreased fine motor abilities in HE patients are reflected in decreased MLS1 score, which correlate with APTwT1 MTRasym in the cerebellum in the present work. Increased ammonia accumulation or hepatocerebral degeneration, reflected in decreasing APTwT1 MTRasym, might be associated with these findings. This notion tallies previous findings demonstrating that cerebellar damage is linked to sensorimotor deficits (Miall et al., 2007) and slower upper limb movements (Manto et al., 2012). In minimal HE, motor impairments manifest in decreasing finger movement frequency and increasing movement amplitudes even prior to alterations in psychometric test scores (Butz et al., 2010). Therefore, APTwT1 MTRasym could shed light on cerebellar alterations leading to impaired motor performance in HE. An additional hint to this interpretation is the correlation with the motor reaction time (WRT2) score.

The contrast mechanisms behind APTw imaging are assumed to be multifactorial and are a matter of debate in the literature today. Possible confounders are changes in pH (Zhou et al., 2003b), water content and water T1 (Lee et al., 2017), and concentration changes in metabolites and proteins (Zaiss et al., 2015), as well as altered protein conformation within intracellular protein compounds (Goerke et al., 2015), to mention a few. These factors are discussed further in 4.3. Based on our previous study and the supplementary data (Supplementary Fig. 2) we assume changes in the protein structure to be the main contributor to the observed contrast change in our study. Whether these changes are mediated through direct conformational changes in the protein structure or indirect mechanisms of ammonia, however, cannot be disentangled with the present findings. Thus, hepatocerebral degeneration (Victor et al., 1965; Yalçın et al., 2016) might contribute to the signal reduction in APTw images in more severe HE affecting the cerebellum specifically (Butterworth, 2007). Cerebellar degeneration in HE is characterized by loss of Purkinje cells, and alcoholic abuse is associated with a greater degree of severity of loss of these cells, and a higher HE prevalence (Kril et al., 1997). In light of the etiology of our patient cohort (50% alcoholic liver cirrhosis, 50% non-alcoholic etiology), APTwT1 MTRasym may be altered by an alcohol-induced high degree of cell loss in the cerebellum. Moreover, the appearance of Alzheimer type II astrocytes is likely in those patients (Kril et al., 1997). These cells undergo morphological changes and thus, might alter the APTwT1 MTRasym by changing the number of exchangeable amide groups within the measurement volume. Further in vitro CEST studies including different cell types with HE-related morphological changes might unravel their specific contribution to our findings at hand.

Previous studies reported an involvement of the visual cortex in HE. This includes worse performance in CFF tests (Kircheis et al., 2014), reduced visual GABA/Cr levels (Oeltzschner et al., 2015), abnormal neuronal activity in resting-state fMRI of the visual cortex (Chen et al., 2012), slowed frequency of of alpha and gamma band oscillations recorded with MEG (Baumgarten et al., 2018; Kahlbrock et al., 2012), and alterations in visual evoked potentials (Zeneroli et al., 1984). This involvement is reflected in significant reduction of APTw signals within the occipital cortex and its correlation with the psychometric scores. Increased ammonia accumulation, reflected by decreasing APTwT1 MTRasym, could lead to an impairment of visual perception through neurotransmitter imbalance by ammonia detoxification. A link to CFF was only found within the putamen. This could be due to the fact of relatively large dispersion of the measured APTw MTRasym signals. The strong correlation of blood ammonia levels with APTw signals in cerebellum and occipital cortex substantiates our findings as markers for HE severity.

Further liver disease related markers from the blood test, such as gamma-glutamyltransferase (GGT) concentrations displayed significant correlation with APTwT1 signals in the thalamus. Future CEST studies including larger number of patients and especially patients at higher diseases stage (HE II) may help better clarifying links between clinical parameters and APTw signals in different brain regions.

Ammonia in HE

The multifactorial nature of the pathophysiology of HE includes several alterations in brain metabolism, which interact with the APTw imaging contrast. A key role in the pathophysiology is attributed to the invasion of ammonia in the patients' brains (Cichoz-Lach and Michalak, 2013; Häussinger and Sies, 2013). At physiological pH levels, ammonia remains invisible to the classical CEST approach, due to its high exchange rate. Nevertheless, our recent study could demonstrate a link between ammonia concentration and APTw MTRasym protein signal (Zöllner et al., 2018).

In vitro, the contrast mechanisms were mainly driven by an induced protein denaturation through ammonia, while contrast mechanisms remained unclear in the in vivo cases (Zöllner et al., 2018). Our data describes APTw signal reductions in the cerebellum and the occipital cortex of HE patients, including correlations with blood ammonia levels and psychometric test scores. A possible interpretation is to link this reduction to an increased ammonia load in these regions, as we already assumed in our in vitro experiments.

Our findings are in line with an earlier 13NH3-PET study (Keiding et al., 2006) depicting a correlation between blood ammonia and the metabolic flux of ammonia. In this work, the metabolic flux was defined as the product of the netto metabolic clearance in blood from intracellular metabolites and arterial ammonia concentration, which was deduced from radioactively marked ammonia. Intracellular glutamine is linked to ammonia removal through glutamine synthase within the brain. As APTw signal predominantly reflects intracellular protein compounds we assume the contrast changes in our study to be mediated through ammonia within the astrocytes, which is underlined by comparable correlations in cerebellum and cortex (Keiding et al., 2006). Therefore, we speculate APTw signal to reflect the metabolic flux of ammonia. At present, a more elaborate differentiation of ammonia kinetic remains inaccessible with APTw imaging as it is measured without contrast agents and reflects an averaged metabolism during the whole acquisition. In the present study, we found an increased APTw signal in the putamen and a negative correlation with the CFF, which is in contrast to the findings reported by Keiding et al. for the basal ganglia (Keiding et al., 2006). This result indicates the potentially larger contribution of another contrast mechanism in this region. Such changes might be driven by increased manganese levels within the basal ganglia (Felipo, 2013; Häussinger and Sies, 2013; Prakash and Mullen, 2010; Rose et al., 1999; Rovira et al., 2008).

Earlier PET studies (Keiding et al., 2006; Lockwood et al., 1991) also reported increased cerebral metabolic rates and radioactivity of ammonia in the thalamus, respectively. Additionally, thalamic volume changes were found via voxel-based morphometry studies in HE (Lu et al., 2018). In the present study, thalamic APTw signals only correlated with GGT concentrations. Firstly, increased glutamine concentrations could interfere with the effect of ammonia, which is complementary to our additional phantom experiments (Supplementary Fig. 2). Secondly, the changes in thalamic volume could alter the APTw signal due to changes in the protein concentration. Thirdly, the changes in APTw signal could be dominated by hepatocerebral degeneration, which possibly differs between various regions. Yet, these explanations need to remain speculative.

Ammonia detoxification by glutamine synthase triggers the accumulation of glutamine within astrocytes (Felipo, 2013; Häussinger and Sies, 2013; Prakash and Mullen, 2010). Several studies reported disturbance in glutamate/glutamine balance, such as increasing glutamine concentrations after chronic liver failure in rat models (Swain et al., 1992), and elevated glutamine concentrations in basal ganglia (Miese et al., 2006) and cortical brain regions of HE patients (Häussinger et al., 1994; Kreis et al., 1991; Oeltzschner et al., 2015). Glutamine and glutamate are both CEST-sensitive, and resonate between 3 and 4 ppm (Schmidt et al., 2016).

As a result, signal contributions of both metabolites might interact with the APTw mechanism used in this study. In this case, the increasing APTw signal and the negative correlation with the CFF in the putamen could be interpreted as increased glutamate + glutamine (Glx) ratios. In additional phantom experiments (Supplementary Fig. 2), we found that ammonia dominates the contrast mechanism with its presence, by adding in vivo realistic glutamate, glutamine (control and HE concentrations), and ammonia concentrations in one solution.

Besides, a 3 T system does not have a high sensitivity to depict changes in either glutamate or glutamine (Lee et al., 2016), but showed sensitivity to ammonia load in protein based solutions (Zöllner et al., 2018). Hence, the underlying mechanisms in the putamen remain unclear, but are likely to be linked to HE disease severity.

CEST-sensitive confounders

Starting from APTw imaging, we adapted our saturation parameters in two ways (Zöllner et al., 2018). In in vitro experiments, the saturation parameters were selected to maximize the contrast gained from the effect of ammonia on the protein signal. Then, these saturation parameters were used to create full brain coverage within the in vivo measurements. In vitro, we were able to link the alterations in APTw signals directly to ammonia, but in vivo, ammonia could only be one of a few contributors to the alterations. Another physiological change described for HE is the osmotic imbalance, which is triggered through the ammonia depletion. Finally, alterations in brain water homeostasis and the emergence of a low-grade edema are reported (Häussinger and Schliess, 2008). Regarding the MR visibility of these changes, one study reported water content changes of about 2% in several brain regions including the putamen (Shah et al., 2008), whereas an absence of MR visible water content changes was reported in another study in HE patients in less severe stages (Oeltzschner et al., 2016).

Several APTw CEST studies at 4.7 T and 7 T emphasize that water content plays a minor role in the contrast formation of APT only (Khlebnikov et al., 2016; Lee et al., 2017). However, as the APTw values were normalized by the T1 relaxation time, any possible T1 effect can be neglected.

Quantitative T1 (Shah et al., 2003) and T1-weighted (Butterworth et al., 1995; Klos et al., 2006; Pujol et al., 1993; Rovira et al., 2008) changes are known to be a key finding in MR imaging of hepatic encephalopathy. To rule out the contamination of our MTRasym we employed a T1 normalization, which resolves in pure APTw contrast mechanisms. In addition, a recent study claimed that MTRasym of amide protons at the saturation parameters (B1, tsat) used in the present study is roughly insensitive to water longitudinal relaxation time (Zu, 2018).

Limitations

One limitation of the present study is the number of included patients. As our data suggests that the APTw signals are strongly altered in manifest HE, the inclusion of more patients in the higher HE grade could have strengthened our findings. However, the inclusion of patients in higher disease stages is hard to achieve, as patient compliance is needed both to perform psychometric testing and to achieve sufficient quality of MR data without a substantial amount of movement artifacts.

Another limitation is the number of analyzed ROIs, which included prior knowledge and assumptions within the analysis and led to the fact that some effects might be missed in other brain regions. Thus, all interpretations remain descriptive in the first place and final conclusions about affected brain regions and APTw imaging as marker for HE severity have to be confirmed in future larger studies. In addition, the inclusion of larger cohorts would allow the implementation of non-parametric voxel based analysis (Holmes et al., 1996). This technique excludes user-biased ROI analysis and could give further insight in the disease progression and spatial distribution of HE-related changes. Nevertheless, the present study indicates that APTw imaging is a possible marker for HE, as correlations with HE-related clinical markers were evident.

The sensitivity of the CEST technique is another limitation and a possible explanation for the absence of thalamic alterations in the present study. As discussed above, several confounders to the APTw signal remain elusive, which could lead to increasing type II error rates. In future studies, a combination of MRS with the CEST technique could shed light on the underlying contrast mechanisms. By using the voxel-based acquisition technique EXPRESS (Walker-Samuel et al., 2012) in combination with J-difference-edited MEGA-PRESS spectroscopy (Mescher et al., 1998) or a novel accelerated spectral editing sequence allowing the measurement of multiple compounds at the same time in the same brain region (Saleh et al., 2016), the interplay between neurotransmitter metabolism (GABA, glutamate, glutamine), osmolytes (myo-inositol), oxidative stress markers (glutathione), and changes in the protein signals could be investigated. These studies may focus on alterations in cerebellum and thalamus, as the interplay of metabolism and HE is yet unclear in these regions. Additional acquisition of water references would further allow drawing conclusions about water content.

To reach sufficient acquisition times and full brain coverage, relatively short saturation times, small numbers of saturation frequencies, and a non-steady state acquisition were chosen in the present study. Faster imaging sequences would allow improving those parameters, which could even include steady-state CEST measurements within the acquisition. As a result, further correction of the CEST signal could be implemented, such as AREX (Zaiss et al., 2015) or EMR (Heo et al., 2016). More advanced quantification approaches like Lorentzian fitting (Zaiss et al., 2011) could also improve data quality and strengthen the findings. Additionally, a Lorentzian fitting model could be able to distinguish between APT, Nuclear Overhauser effect (NOE), and MTC, if a suitable model is chosen. Apart from the fact that ammonia only affects the APT exchange in in vitro experiments (Zöllner et al., 2018), some in vivo studies reported MTC to be altered in HE patients (Miese et al., 2006). Moreover, morphological changes might affect the NOE, thus Lorentzian fitting could give further insight to the underlying contrast mechanisms in HE. It was not implemented in the current study, as it requires a large number of saturation frequencies leading to long total acquisition times. The implemented MTRasym analysis combined with the full brain coverage already permits the distinction between controls and patients, based on APTw signal alterations in several brain regions.

Conclusions

Hepatic encephalopathy is associated with a region-specific decrease of APTw signals, in particularly in the cerebellum and the occipital cortex. These signal changes are linked to increased blood ammonia concentrations, and clinical scores of cognitive and motor function. These variations are most likely a consequence of hyperammonemia or hepatocerebral degeneration processes and develop in parallel with disease severity. Therefore, APTw CEST imaging could be a possible tool to investigate HE and advance the understanding of region-specific alterations in HE and its clinical equivalents. By including additional methods to quantify metabolite levels and water content, the interplay between metabolism and protein signal alterations in HE could be assessed in more detail in future studies.

Conflict of interest

D.H. belongs to a group of patent holders for the bedside measurement device determining the critical flicker frequency.