Nonalcoholic Fatty Liver Disease: Correlation of the Liver Parenchyma Fatty Acid with Intravoxel Incoherent Motion MR Imaging–An Experimental Study in a Rat Model

Purpose To prospectively evaluate the changes in fatty acid concentration after administrating a 60% high-fat diet to a non-alcoholic fatty liver disease rat model and to perform a correlation analysis between fatty acid with molecular diffusion (Dtrue), perfusion-related diffusion (Dfast), and perfusion fraction (Pfraction). Material and Methods This prospective study was approved by the appropriate ethics committee. Ten male Sprague-Dawley rats were fed a 60% high-fat diet until the study was finished. Point-resolved spectroscopy sequence 1H-MRS with TR = 1,500 msec, TE = 35 msec, NEX = 64, and 8×8×8 mm3 voxel was used to acquire magnetic resonance spectroscopy (MRS) data. Diffusion-weighted imaging was performed on a two-dimensional multi-b value spin echo planar image with the following parameters: repetition time msec/echo time msec, 4500 /63; field of view, 120×120 msec2; matrix, 128×128; section thickness, 3 mm; number of repetition, 8; and multiple b value, 0, 25, 50, 75, 100, 200, 500, 1000 sec/mm2. Baseline magnetic resonance imaging and magnetic resonance spectroscopy data (control) were acquired. 1H proton MRS and diffusion-weighted imaging were obtained every 2 weeks for 8 weeks. The individual contributions of the true molecular diffusion and the incoherent motions of water molecules in the capillary network to the apparent diffusion changes were estimated using a least-square nonlinear fitting in MatLab. A Wilcoxon signed-rank test with the Kruskal-Wallis test was used to compare each week’s fatty acid mean quantification. Spearman’s correlation coefficient was used to evaluate the correlation between each fatty acid (e.g., total lipid (TL), total saturated fatty acid (TSFA), total unsaturated fatty acid (TUSFA), total unsaturated bond (TUSB), and polyunsaturated bond (PUSB)) and intravoxel incoherent motion (IVIM) mapping images (e.g., Dtrue, Dfast, and Pfraction). Results The highest mean TL value was at week 8 (0.278 ± 0.10) after the administration of the 60% high-fat diet, followed by weeks 6, 4, 2, and 0. The concentration level (16.99±2.29) of TSFA at week 4 was the highest. No significant differences in the concentrations of TUSFA, TUSB, or PUSB were observed in different weeks. Conclusion After the administration of the 60% high-fat diet in nonalcoholic fatty liver disease model, TL and TSFA depositions had significant changes. The mean concentrations of TUSFA, TUSB, PUSB did not significantly change. Total unsaturated fatty acid and polyunsaturated bond showed positive correlations with Dtrue and Pfraction.


Material and Methods
This prospective study was approved by the appropriate ethics committee. Ten male Sprague-Dawley rats were fed a 60% high-fat diet until the study was finished. Point-resolved spectroscopy sequence 1 H-MRS with TR = 1,500 msec, TE = 35 msec, NEX = 64, and 8×8×8 mm 3 voxel was used to acquire magnetic resonance spectroscopy (MRS) data. Diffusion-weighted imaging was performed on a two-dimensional multi-b value spin echo planar image with the following parameters: repetition time msec/echo time msec, 4500 /63; field of view, 120×120 msec 2 ; matrix, 128×128; section thickness, 3 mm; number of repetition, 8; and multiple b value, 0, 25, 50, 75, 100, 200, 500, 1000 sec/mm 2 . Baseline magnetic resonance imaging and magnetic resonance spectroscopy data (control) were acquired. 1 H proton MRS and diffusion-weighted imaging were obtained every 2 weeks for 8 weeks. The individual contributions of the true molecular diffusion and the incoherent motions of water molecules in the capillary network to the apparent diffusion changes were estimated using a least-square nonlinear fitting in MatLab. A Wilcoxon signed-rank test with the Kruskal-Wallis test was used to compare each week's fatty acid mean quantification. Spearman's correlation coefficient was used to evaluate the correlation between each fatty acid (e.g., total lipid (TL), total saturated fatty acid (TSFA), total unsaturated fatty acid (TUSFA), total Introduction Non-alcoholic fatty liver is a common disease that is currently experienced bu 10-24% of the world's population [1]. It can progress into chronic liver diseases such as liver fibrosis and cirrhosis [2][3][4]. In the USA, 32-37% of nonalcoholic fatty liver disease (NAFLD) patients progress to liver fibrosis in 3-6 years, with 12% experiencing liver cirrhosis in 8-10 years [5][6][7][8]. A simple scoring system for predicting cirrhosis in nonalcoholic fatty liver disease has been used that measured platelet, albumin, and AST/ALT ratios in blood samples [9]. NAFLD is usually associated with metabolic syndromes (MetS), such as type 2 diabetes, insulin resistance, hypertension, and dyslipidemia, that are accompanied by abdominal obesity [10][11]. Accordingly, clinical approaches to investigating the relationship between NAFLD and the prevalence of atherosclerosis and cardiovascular disease have been conducted [12]. NAFLD is not a simple risk factor of type 2 diabetes, cardiovascular disease, or liver cirrhosis.
Liver biopsy has been accepted as the gold standard for diagnosing NAFLD. However, liver biopsy is not only invasive but also clinician-dependent, which often causes sampling errors [13]. Moreover, it is impossible to use this method over a long period to monitor the progress of NAFLD. In a 2014 study, Davide et al. examined whether extracellular vesicles are increased in the liver and blood during experimental NAFLD during searchers for bio-markers [14]. Magnetic resonance spectroscopy (MRS) has been widely used to diagnose NAFLD [15][16][17][18]. MRS is a noninvasive and in vivo method with high sensitivity and specificity. It is frequently used to diagnose various diseases such as NAFLD, liver fibrosis, and cirrhosis [19][20][21][22]. Total lipid (TL), total saturated fatty acid (TSFA), total unsaturated fatty acid (TUSFA), total unsaturated bond (TUSB), and polyunsaturated bond (PUSB) can be calculated with liver fatty acid analysis using 1 H-MRS through the signal integration of lipid methyl protons (-CH 3, 0.9 ppm), methylene protons ((-CH 2 -)n, 1.3 ppm), allylic protons (-CH 2 -C = C-CH 2 -, 2.0 ppm), diallylic protons (= C-CH 2 -C =, 2.8 ppm), and methene protons (-CH = CH-, 5.3 ppm) [19]. Many studies have been conducted on fatty acid deposition in liver parenchyma and fatty acid changes in liver fibrosis using animal models [19,23].
Diffusion-weighted imaging (DWI) techniques have been recently attempted for diagnosing fatty liver, liver fibrosis, and cirrhosis [24][25][26]. In particular, the intravoxel incoherent motion (IVIM) is a method of obtaining multi b values to encompass both low-b-value and high-bvalue diffusion-weighted images to reflect the random microscopic motion that occurs in voxels on MRI of either intracellular or extracellular water molecules and the micro-circulation of blood through non-linear bi-exponential graph fitting [27]. Previous studies have shown that pure molecular diffusion and perfusion fraction values in the liver parenchyma of liver fibrosis or cirrhosis were lower than those in a control group [24][25][26][27][28]. In fatty liver patients, fat deposition in the liver parenchyma was similar to that in liver fibrosis patients because of distortion and sinus compression of the microcirculatory anatomy and reduced pure molecular diffusion [27][28]. However, the correlation of changes in fatty acid in the liver parenchyma with fat deposition and true diffusion with blood microcirculation has not yet been studied.
Therefore, the objectives of this study were to examine the changes in fatty acid concentration after high-fat diet administration and to determine factors that affect NAFLD through correlation analysis between fatty acid and pure molecular diffusion (D true ), perfusion-related diffusion (D fast ), or perfusion fraction (P fraction ).

Materials and Methods
This study was carried out in strict accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health. The protocol was approved by the Committee on the Ethics of Animal Experiments of the Korea Basic Science Institute (KBSI-AEC 1305). All surgery was performed under sodium pentobarbital anesthesia, and all efforts were made to minimize suffering.

Animal Model
Ten male 8-week-old Sprague-Dawley rats that weighed 100-150 g were housed with ad libitum access to water. The animal care facility was controlled for humidity and temperature on a 12 h light-dark schedule. All Sprague-Dawley rats were fed a 60% high-fat diet that contained 60% fat, 20% protein, and 20% carbohydrate (D12492, Research Diets, New Brunswick, NJ) until the experiment was complete [29]. Baseline MRI and MRS data were acquired before the rats were fed the high-fat diet. MRI and MRS data were also acquired every 2 weeks for 8 weeks according to the schedule of the 1 H-MRS experiments. The experimental rats were anesthetized for all surgical and imaging procedures by general inhalation anesthesia (isofluorane 1.5 to 2.5% vol., plus O 2 ). After completion of the imaging study, mice were sacrificed under deep anesthesia and the livers were excised and processed for further histological analysis.

In vivo Liver 1 H MRS
All MRI and 1 H-MRS experiments were performed on a 3.0Tesla MRI scanner (Achiva Tx 3.0 T; Philips Medical Systems, Netherlands) with a maximum gradient of 200 mT/m using a 4-channel animal coil (CG-MUC18-H300-AP, Shanghai Chenguang Medical Technologies Co., Ltd., China). During liver MRS and imaging, all NAFLD model rats were anesthetized with isoflurane/air at 1.0 to 1.5% via a nose cone with respiratory monitoring [30]. Using T 2weighted fast spin echo, whole liver parenchyma images were acquired in three transverse axial (FOV 60 mm×60 mm, slice thickness = 1.5 mm), coronal (FOV 6 cm×6 cm, slice thickness = 1.5 mm), and sagittal (FOV 6 cm×6 cm, slice thickness = 1.5 mm) planes to localize voxels or volume of interest for MRS.
We used a point-resolved spectroscopy (PRESS) sequence for localized 1 H-MRS with TR = 1,500 msec, TE = 35 msec, NEX = 64, and total scan time = 10 minute. An 8×8×8 mm 3 voxel was placed within a homogeneous liver parenchyma to avoid large blood vessels, as shown in Fig 1. We applied iterative VOI shim [31]. The water signal of each VOI was suppressed by variable pulse power and optimized relaxation delays before the scan. The signal was shimmed to a line width of lipid (4 to 6 Hz) over VOI (30).

IVIM MR imaging
The multi b value DWI was conducted immediately after the 1 H-MRS data were obtained by using the same MR system. Whole liver parenchyma transverse axial images were obtained. DWI was performed on a two-dimensional multi-b value spin echo planar imaging with the following parameters: prepetition time /echo time, 4500 /63 msec; field of view, 120×120 mm 2 ; matrix, 128×128; section thickness, 3 mm; number of repetition, 8; and multiple b value, 0, 25, 50, 75, 100, 200, 500, and 1000 sec/mm 2 . Individual contributions of true molecular diffusion and incoherent motions of water molecules in the capillary network to the apparent diffusion changes, D true , D fast , and P fraction were estimated as shown in Fig 2 using a least-square nonlinear fitting in MatLab (Mathworks, Natick, MA, USA) by fitting, pixel by pixel, the DWI signal decay in the region of interest (ROI) to the IVIM by-compartmental model as follows: SI/SI 0 = (1-P fraction )×exp (-bD true ) + P fraction ×exp (-bD fast ), where SI was the signal intensity and P fraction was the perfusion fraction linked to blood volume.

Histology
After 8 weeks of MRI and 1 H-MRS study, all rats were sacrificed for histological evaluation. One additional normal animal was sacrificed as a control. All liver specimens were fixed in formalin, embedded in paraffin, sectioned, and examined under light microscopy after standard hematoxylin-eosin staining and Prussian blue staining to confirm fat and iron deposition in the liver parenchyma.

MRS Analysis
Raw 1 H-MRS data were analyzed using LCModel software (version 6.3-1H, Stephen W. Provencher). Spectrum type with lipid-11 was selected for quantifying all lipids, with water data as references. All lipid peaks were automatically calculated in LCModel. The LC model parameters were set according to the LCModel & LCMgui user's manual (http://s-provencher.com/ pages/lcm-manual.shtml). Less than 10% standard deviation (%SD) of metabolite quantification data was allowed. The %SD called the Cramér-Rao lower bound of useful reliability indicators was used for error estimates. The integrating areas under peaks were measured as follows: signal integrals in lipid methyl protons at 0.90 ppm, methylene proton at 1.30 ppm, allylic protons at 2.02 ppm, diallylic protons at 2.77 ppm, and methane protons at 5.30 ppm. For relative quantification, total lipid ((-CH 2 -)n / noise), total saturated fatty acid, total unsaturated fatty acid, total unsaturated bond, and polyunsaturated bond were quantified by separating each peak area of (-CH 2 -)n, -CH 2 -C = C-CH 2 -, = C-CH 2 -C =, and -CH = CH-by -CH 3 as shown in Table 1.

IVIM MR Analysis
Extracted mapping imaging diffusion coefficient, pseudo diffusion coefficient, and perfusion fraction were analyzed using standard software on the workstation. All ROIs were manually positioned by one author (Yu). The region of interest of each image was a set circle in the right lobe that was just about the same as the VOI region of MRS. Every ROI size was manipulated differently during measurement to avoid IVIM mapping imaging artifacts. The Wilcoxon signed-rank test with the Kruskal-Wallis test was used to compare each week's fatty acid mean quantifications with significance level p<0.05. Results are expressed as mean ± standard deviation. Spearman's correlation coefficient was used to evaluate the correlations between each fatty acid (e.g., TL, TSFA, TUSFA, TUSB, and PUSB) and IVIM mapping images (e.g., D true , D fast , and P fraction ) with a significance level of p<0.05. All statistical analyses were performed using SPSS version 20.0 (SPSS Incorporated, Chicago, IL, USA).

Comparison of the mean of hepatic fatty acids
The differences in the N number shown under the X axis (Fig 3) were observed because 10% or lower % SD values were used.
The highest mean TL value was measured at week 8 (0.278±0.10) after the administration of the 60% high-fat diet, followed by weeks 6, 4, 2, and 0. The TL value at week 2 was 0.201±0.07, which was insignificantly (p>0.05) lower than that of week 8. The TL value increase at week 2 from baseline week 0 was the most significant (p<0.01) among all of the increases. The mean increases in the TL value at weeks 2 and 4 were also significant (p<0.05). The TSFA concentration level (16.99±2.29) at week 4 was the highest. A statistically significant mean difference in the levels was observed only between the baseline and week 2 (p<0.05). No significant differences in the concentration levels of TUSFA, TUSB, or polyunsaturated bond PUSB were observed among different weeks.

Correlation between hepatic fatty acid and IVIM mapping images
The results of the non-parametric analysis of the correlations between hepatic fatty acid and IVIM mapping images are summarized in Table 2.
A correlation between total lipid and D fast was observed (r = 0.333, p<0.05). P fraction had no correlation with TL. TUSFA showed a positive correlation with D true. . PUSB was significantly correlated with D fast (r = 0.413, p<0.01) and P fraction (r = 0.346, p<0.05) in the liver parenchyma, confirming that D fast and P fraction are closely related to liver parenchymal microcirculation. PUSB also correlated with D true (r = 0.413, p<0.05, Fig 4).

Discussion
In this study, changes in fatty acid with fat deposition were observed in liver parenchyma. Correlations among fatty acid, intracellular or extracellular water molecules, and blood microcirculation were investigated. As the NAFLD progresses into chronic liver diseases such as liver fibrosis and cirrhosis, the fat deposition in the liver parenchyma is known to have a negative effect on micro-circulation. In this study, there was great significance in revealing the specific fatty acid through the correlate fatty acid and representing the micro-circulation factor (D true , D fast , and P fraction ) in NAFLD.  Table 2. Spearman correlation analysis between hepatic fatty acid and IVIM data mapping images.  After the administration of a 60% high-fat diet, the TL level consistently increased, with the peak at week 8. However, the mean difference between week 6 and week 8 was statistically insignificant. TL concentration was saturated at week 6. The NAFLD modeling through the high-fat diet was completed at week 6 and afterwards.
The TSFA concentration calculated through the 1.3 ppm and 0.9 ppm lipid proton ratio increased consistently from week 0 to week 4 and decreased thereafter. The mean difference in TSFA concentrations between week 2 and afterwards was statistically insignificant. The mean levels of TUSFA and PUSB increased from week 0 to week 6 and decreased thereafter. However, the differences were statistically insignificant. The mean differences in TUSB concentrations were also statistically insignificant. Changes in TL and TSFA deposition were the most significant after the administration of the 60% high-fat diet, whereas the changes in the other unsaturated acids and bonds were insignificant. Excellent liver fat deposition was confirmed after hematoxylin-eosin staining for histological evaluation. According to studies by Pacifico et al. [32] and Scorletti et al. [33], the administration of n-3 PUFA positively affected Mets and NAFLD [32][33]. In this study, the long-term administration of the 60% high-fat diet did not affect the changes in PUFA, TUSFA, or TUSB. Therefore, a high-fat diet might be an indirect factor that could adversely affect the liver.
Studies on liver diseases using DWI have been conducted previously [25][26][27][28]34]. Particularly, D true , D fast , and P fraction data can be analyzed by applying various b-values on the same slice using signal reduction in the low and high b-values through bi-exponential data fitting. In previous studies, a slightly negative correlation between the increase in fat fraction value and the D true value was confirmed with statistical significance [29]. In another study, D true and D fast values were low in the hepatic steatosis group without hepatic steatosis [28]. Based on the results of previous studies, the TL and TSFA were expected to show negative correlations with D true and D fast because the liver fat content calculation from the equation of Guiu et al.
(100 Á CH 2 )/(CH 2 + water) was similar to the TL in this study, which was calculated using the 1.3 ppm proton lipid metabolite and water ratios. In this study, no fatty acid showed a correlation with D true . Rather, a low correlation was observed between total lipid and D fast (r = 0.333, p<0.05). The etiology in the previous studies that targeted human bodies differed from that in other studies because of fat deposition caused by type b and c type hepatitis as well as by the fatty liver that was induced by the administration of the high-fat diet. Accordingly, studies on microcirculation in the liver parenchyma according to the type of fatty acid caused by etiology are required. In a 2014 study by Joo et al. that used a NAFLD rabbit model, the D true value in the severe NAFLD group decreased slightly without statistical significance. However, D fast decreased significantly [35]. Because the NAFLD model P fraction reflects the capillary perfusion contribution in ADC, the D true reduction and the D fast increase might have been influenced by the fatting equation that meant a P fraction decrease. The positive correlation of D true with D fast in this study could have been for the same reason. The high D true meant an increase in the pure water diffusivity and in the capillary perfusion [28,35]. In this study, the fatty acid that showed a positive correlation with D true and P fraction was TUSFA. The correlation coefficient with D true was 0.535 (p<0.01), a significantly high correlation. A positive correlation with P fraction was also confirmed with statistical significance. The increase in the pure water diffusivity positively affected capillary perfusion. In particular, PUSB was confirmed to have positive correlations with D fast , D true , and P fraction with statistical significance. The positive correlation between D fast and P fraction was attributed to the fact that D fast represented the perfusion-related incoherent micro-circulation whereas P fraction represented the blood volume. In a study by Monteiro J et al. [36], n-6 and n-3 polyunsaturated fatty acids had beneficial effects on cardiovascular diseases and non-alcoholic fatty liver diseases. In this study, positive effects on D fast , D true , and P fraction in USFA and PUSB were confirmed. However, the mean concentrations of USFA and PUSB did not significantly increase after the administration of the 60% high-fat diet. Additional studies on the positive effects of increased TUSFA and PUSB concentrations on blood perfusion in liver parenchyma are merited.
The limitations of this study are as follows: For accurate metabolite quantifications, spin-lattice (T 1 ) and spin-spin (T 2 ) relaxation must be corrected. T 2 Ã correction and accurate fat fraction calculation using chemical shift imaging techniques have been attempted not only in MRS but also in multi-echo Dixon [37][38][39][40][41]. In a study by Andrew et al., new methods of in vivo T 2 time measurement through echo-time averaging were suggested [42]. Iron deposition in the liver parenchyma was confirmed to be one of the major factors that triggered T 2 Ã that was closely correlated with the development of liver fibrosis and cirrhosis [43][44][45]. In the fatty liver model in this study, iron deposition through Prussian blue staining did not occur at week 8 or afterwards. The absence of T 2 Ã correction for accurate metabolite quantification was the most significant limitation of this study.

Conclusions
In conclusion, the administration of the 60% high-fat diet to rats gradually increased the TL and TSFA levels in the liver parenchyma, after which the liver parenchyma became saturated.
In the meantime, the levels of TUSFA, TUSFB, and PUSB did not show statistically significant changes. The fatty acids that showed positive correlations with D true and P fraction in the liver parenchyma were TUSFA and PUSB.
Supporting Information S1 Dataset. The raw data of lipid proton concentration MRS and IVIM parameters. The lipid signals were analyzed using the LCModel algorithm, and a fitting error (the standard deviation) was estimated using the Cramer-Rao lower bounds (CRLB) of the signal amplitude (e.g., at 0.9, 1.3, 2.02, 2.77, and 5.30 ppm). In this work, a CRLB threshold of 10% was used for quality control. (XLSX)

Author Contributions
Conceived and designed the experiments: SMY SHK HMB. Performed the experiments: SMY HMB. Analyzed the data: SMY SHK HMB. Contributed reagents/materials/analysis tools: SMY HMB. Wrote the paper: SMY SHK HMB.