The Effects of Foods on Blood Lipids in Non-alcoholic Fatty Liver Disease (NAFLD)—A Systematic Review and Meta-Analysis

Background: Non-alcoholic fatty liver disease (NAFLD) is associated with dyslipidemia and increased cardiovascular disease risk. Dietary choices may produce profound effects on blood lipids. Thus, the purpose of this study was to investigate which foods modify blood lipids in NAFLD. Methods: Systematic review of published systematic reviews and randomized controlled trials (RCTs). Searches were performed in PubMed, Cochrane Database of Systematic Reviews, and Cochrane Central Register of Controlled Trials, from inception through March 2020. Studies in populations with NAFLD, which provided data on foods or dietary patterns and blood lipids were included, but not weight loss diets, supplements, nor individual nutrients. The strength of evidence was evaluated using The Grading of Recommendations Assessment, Development, and Evaluation (GRADE). Results: No relevant systematic reviews were identified. Eleven RCTs were included in the qualitative synthesis. Two RCTs were included in meta-analyses, regarding the comparison between Mediterranean and Low-fat diets, in which there were no clear effects on either high-density lipoprotein cholesterol or triglycerides, with Low evidence. From single RCTs, there was Moderate evidence for reduced triglycerides by a healthy dietary pattern, compared with usual care; and for reduced total cholesterol by a probiotic yogurt, enriched with Lactobacillus acidophilus La5 and Bifidobacterium lactis Bb12, compared with conventional yogurt. For all other comparisons, the evidence was considered as Low or Very low. Conclusion: Few studies were identified which reported effects of foods on blood lipids in subjects with NAFLD. The possible beneficial effect of probiotics warrants further study. PROSPERO identifier: CRD42020178927.


INTRODUCTION
Non-alcoholic fatty liver disease (NAFLD) is a prevalent and emerging public health concern, closely associated with obesity, metabolic syndrome, insulin resistance, and type 2 diabetes (1). Individuals with NAFLD frequently present with dyslipidemia, typically elevated fasting plasma triglycerides and/or low highdensity lipoprotein (HDL) cholesterol, but not necessarily elevated low-density lipoprotein (LDL) cholesterol (2); and may have high, or even very high, cardiovascular disease (CVD) risk (3). Dietary choices may produce profound and rapid effects on plasma lipoproteins, even under isocaloric conditions, and is thus possible to determine with high evidence in well-conducted short-term randomized controlled trials (RCTs). When several foods with cholesterol-lowering effects are combined (e.g., in the Portfolio diet) in strictly controlled settings, effects approaching 30% may be achieved on the group level (4). However, the effects of dietary fat quality (e.g., saturated and polyunsaturated fat) on LDL cholesterol and levels of apolipoprotein B are less pronounced in individuals with obesity than without (5). Considering the strong association between NAFLD and obesity, it can thus be speculated that the effect of dietary fat type may be less effective in individuals with compared to without NAFLD (due to concomitant overweight), however this requires further investigation. Factors proposed to blunt the effectiveness of dietary fat modification on LDL cholesterol in obesity (and thus speculatively in NAFLD) are increased inflammation, insulin resistance and endogenous cholesterol Population: Adults with NAFLD, not taking lipid-lowering medication Intervention: Intake of foods and dietary patterns, but not weight loss diets, supplements, nor individual nutrients.
Control: Background diet or comparison foods, but not lower dose of the same food.
Searches in PubMed, Cochrane database of systematic reviews and cochrane central register of controlled trials, through March 2020 Filter: RCTs, systematic reviews and meta-analyses. Modified  OR "Plant Oils" OR "Fatty Acids" OR "Fatty Acids, Unsaturated" OR "Fatty Acids, Monounsaturated" OR "Fatty Acids, Omega-3" OR "Fish Products" OR "Fish Oils" OR Avena OR Hordeum OR "Edible Grain" OR Psyllium OR Flax OR "Soy Foods" OR "Prunus dulcis" OR Nuts OR Phytosterols OR "Lycopersicon esculentum" OR Sugars OR Carbohydrates OR Alcohols AND Title/Abstract: NAFLD OR "Fatty liver" OR "Liver steatosis" OR MeSH terms: "Non-alcoholic Fatty Liver Disease" apo, apolipoprotein; HDL, high-density lipoprotein; LDL, low-density lipoprotein; NAFLD, non-alcoholic fatty liver disease; MUFA, monounsaturated fatty acids; RCT, randomized controlled trial. *Wildcard caracter. synthesis (6). Triglycerides may also be causally associated with CVD risk and can be clearly affected by diet and/or weight loss, although by distinctly different food choices than regarding hypercholesterolemia (7). The mechanisms through which foods influence plasma lipoproteins are partly unknown and may also be independent of the overall dietary fatty acid composition (8). The main routes involve intestinal absorption and excretion of lipids (including bile acids), or hepatic synthesis or clearance from the circulation. However, it is unknown whether hepatic steatosis modifies these effects. For instance, foods affecting gut microbiota (e.g., probiotics and prebiotics) may be of special importance. Also, foods affecting intestinal cholesterol uptake (e.g., foods enriched with phytosterols and/or stanols) could theoretically have less importance in NAFLD compared with foods believed to mainly exert direct effects on the liver, e.g., foods high in polyunsaturated fatty acids (PUFA). Thus, we aimed to systematically evaluate the literature for the effects of foods on blood lipids in persons with NAFLD.

METHODS
A systematic search for systematic reviews and RCTs was performed, in PubMed, Cochrane Database of Systematic Reviews, and Cochrane Central Register of Controlled Trials, from inception through March 2020. The PICO and full search strategy are provided in Table 1. Included search terms were selected based on dyslipidemia guidelines and studies in other populations. The target population was adults with NAFLD, not taking lipid-lowering medication. In order to be included, studies must include information on NAFLD diagnosis or provide other data of liver fat content of participants, and a clear majority of participants must conform with inclusion criteria. Interventions of interest were foods and dietary patterns, but not single nutrients (not clearly associated with specific foods), supplements, nor weight loss diets. As comparison, other foods or background diets were acceptable, but not different doses of the same food. The outcomes of interest were LDL cholesterol, total cholesterol, HDL cholesterol, apolipoprotein B, apolipoprotein AI, triglycerides, non-HDL cholesterol, oxidized LDL cholesterol, and other lipoproteins and subfractions. If data was not reported, authors were not contacted for additional information, however, for included studies, if data was ambiguous, authors were contacted for clarification. Bibliographies of included RCTs were screened for additional studies. Language was restricted to English. The study protocol is available at www.crd.york.ac.uk/PROSPERO, identifier CRD42018089661. No ethics approval was required.

STUDY SELECTION, DATA EXTRACTION, RISK OF BIAS, AND EVIDENCE ASSESSMENTS
Two reviewers screened abstracts and in cases of disagreement, the paper was included for full-text review. From RCTs that fulfilled inclusion criteria, data was extracted by at least two researchers using standardized forms including: author, year, study type (parallel/crossover), population (NAFLD/other), n analyzed overall and per group, % male, age in years (mean ± SD and/or range), intervention/daily dose (mean ± SD and/or range), control/daily dose (mean ± SD and/or range), effects on lipids (mean difference, 95%CI and/or end mean for intervention and control groups), duration, country, and funding. Risk of bias in RCTs was assessed by two authors using the SBU Risk of Bias tool for intervention studies (version 6 May 2020). The quality of evidence was evaluated using The Grading of Recommendations Assessment, Development, and Evaluation (GRADE), by predetermined criteria for risk of bias, inconsistency, indirectness, imprecision, publication bias, large effects, dose response relationships, and opposing bias. For instance, evidence was downgraded (-1) for inconsistency if results were not replicated, as consistent results of low heterogeneity are impossible to attain from a single RCT. In indeterminate cases, discussions were extended among all authors and overall judgments were employed.

DATA SYNTHESIS AND ANALYSIS
Results from included RCTs were included in meta-analyses, when considered appropriate, using ReviewManager 5.3 software. The Cochrane Handbook (9) was adhered to. In cases of at least moderate (I²>50%) unexplained heterogeneity, random effects models were preferred. If mean changes from baseline and their standard deviations were unavailable and impossible to calculate, end-of-study means and standard deviations were used. Cross-over studies were included together with parallel studies, with modified weights, calculated from available data. In cases where the variances of mean differences were unavailable, a conservative correlation R = 0.5 was imputed and standard deviations adjusted accordingly. Mg/dL was converted to mmol/L by multiplication with 0.02586 for cholesterol and 0.01129 for triglycerides. Data is presented as mean±SD. P-values > 0.05 were considered as non-significant (NS).

RESULTS
Six systematic reviews and 26 RCTs were evaluated in full text. No systematic reviews and 11 RCTs (12 publications) fulfilled inclusion criteria (Figure 1). The characteristics of the 11 included RCTs are presented in Table 2. The excluded papers are presented, with reasons, in Table 3. The overall risk of bias was considered as Low in three RCTs, Some concerns in six RCTs, and High in two RCTs (Table 2 and Figure 2). Four RCTs compared dietary patterns and the remaining seven compared food items. The overall results on blood lipids and the evidence gradings for each available outcome are presented in Table 4. Two studies were considered appropriate to include in metaanalyses, as they both compared a Mediterranean with a Low-fat dietary pattern (Figure 3).

DIETARY PATTERNS Healthy Dietary Pattern vs. Usual Care
In the RCT by Arab et al. (10), there were significant reductions in both total cholesterol (from 5.56 ± 1.6 to 5.01 ± 1.8 mmol/l in intervention vs. from 4.81 ± 0.92 to 4.91 ± 0.89 mmol/l in control group, P = 0.04) and triglycerides (from 2.46 ± 1.9 to 1.86 ± 0.98 vs. from 2.11 ± 1.7 to 2.21 ± 1.4 mmol/L, P = 0.002). There was a tendency toward reduced LDL cholesterol (from 3.11 ± 1.0 to 2.83 ± 1.1 vs. from 2.87 ± 0.84 to 2.96 ± 0.79 mmol/L, P = 0.46), but no clear effect for HDL cholesterol (from 1.20 ± 0.29 to 1.16 ± 0.20 vs. from 1.12 ± 0.26 to 1.07 ± 0.20 mmol/L, P = 0.12). The quality of evidence for the effect on triglycerides was considered as Moderate, upgraded for large effects, whereas all other comparisons were further downgraded to Very low evidence, due to high imprecision.

Mediterranean Diet vs. Low-Fat Diet
From the RCTs by Properzi et al. (19) and Ryan et al. (20), effects on HDL cholesterol and triglycerides were combined in metaanalyses, but there were no clear effects (Figure 3). Effects on total and LDL cholesterol were only reported in the study by Properzi et al. (19), with an indicated slight reduction for total cholesterol (from 4.78 ± 1.3 to 4.53 ± 1.3 vs. from 5.23 ± 0.90 to 5.15 ± 1.1 mmol/L, NS), but no meaningful effect on LDL cholesterol (from 2.83 ± 1.2 to 2.73 ± 1.1 vs. from 3.20 ± 0.78 to 3.17 ± 0.86 mmol/L, NS). The evidence was considered as Low for all outcomes.

High-Monounsaturated Fatty Acid Diet vs. High-Carbohydrate, High-Fiber, Low-Glycemic Index Diet
In the RCT by Bozzetto et al. (12,13), there were tendencies (all NS) toward reductions after the high-monounsaturated fatty acid (MUFA) diet vs. high-carbohydrate (high-CHO) diet, for total (from 4.32 ± 0.65 to 4.29 ± 0.59 vs. from 4.06 ± 0.98 to 4.24 ± 1.2 mmol/L) and LDL cholesterol (from 2.85 ± 0.52 to 2.82 ± 0.54 vs. from 2.53 ± 0.75 to 2.77 ± 1.0 mmol/L), and for triglycerides (from 1.38 ± 0.42 to 1.29 ± 0.37 vs. from 1.24 ± 0.77 to 1.48 ± 1.2 mmol/L). There was no clear effect on HDL cholesterol (from 0.91 ± 0.2 to 0.93 ± 0.1 vs. from 0.96 ± 0.2 to 0.96 ± 0.2 mmol/L). The evidence was considered as Very low. In the report from 2014 (10), results were reported on postprandial lipids combined with 19 (8 + 11 per group) additional subjects that were also prescribed exercise training. Post-prandial (after 2, 4, and 6 h, presented as figure) triglycerides and cholesterol in plasma and chylomicrons/very-low-density lipoproteins were increased after the MUFA diet (P<0.05) and reduced after the high-CHO diet (P < 0.05). We did not perform evidence gradings for these outcomes, as they were not clearly pre-specified.

Olive Oil Enriched With n-3 PUFA vs. Olive Oil
In the RCT by Sofi et al. (15), there was an increase in HDL cholesterol (from 1.15 ± 0.14 to 1.56 ± 0.24 vs. from 1.13 ± 0.12 to 1.07 ± 0.05 mmol/L, P = 0.03) and a reduction in triglycerides (from 1.86 ± 0.97 to 1.50 ± 0.72 vs. from 1.61 ± 0.28 to 1.84 ± 0.34 mmol/L, P = 0.04) in the enriched vs. regular olive oil group. No clear differences were seen for total (from 5.53 ± 0.72 to 5.51 ± 0.60 vs. from 5.11 ± 0.57 to 4.99 ± 0.58 mmol/L, P = 0.9) or LDL cholesterol (from 3.40 ± 0.64 to 3.45 ± 0.84 vs. from 3.24 ± 0.50 to 3.08 ± 0.60 mmol/L, P = 0.2). The evidence was considered as Very low for all outcomes.  The limited number of studies discovered was somewhat surprising, given the prevalence and clinical significance of NAFLD. One plausible reason may be that, in most studies on diet and CVD risk factors, liver fat content is not easily attainable and thus overlooked. Also, many patients with NAFLD go undiagnosed, as it usually presents with little symptoms. The effect of dietary fat type (e.g., saturated/polyunsaturated fat) on plasma lipoproteins is diminished in individuals with obesity compared with normal body weight (5). Considering the strong association between NAFLD and obesity, it can thus be speculated that the effect of dietary fat type may be less effective in individuals with compared to without NAFLD (due to concomitant overweight), however this requires further investigation. Factors proposed to blunt the effectiveness of dietary fat modification on LDL cholesterol in obesity (and thus speculatively in NAFLD) are increased inflammation, insulin resistance, and endogenous cholesterol synthesis (6).
An RCT testing the effects of a synbiotic supplement (fructooligosaccharides plus B. animalis) on liver fat content and liver fibrosis in NAFLD was published during the writing of this manuscript (43). This study showed that this synbiotic combination altered the fecal microbiome but had no effect on liver fat, fibrosis, or plasma lipoproteins compared to prebiotic    20) is a cross-over study and its variance has been adjusted accordingly, in order to increase its weighting.
fructo-oligosaccharides only. This is in contrast to the results reported by Bakhshimoghaddam (11) where a similar synbiotics was administered in the form of yogurt. Whether the mode of administration of synbiotics (supplement vs. yogurt/whole food) modifies the effects on plasma lipoproteins requires further investigation but is not unlikely considering potential matrix effects when part of a whole food. The present systematic review has limitations. Studies on dietary supplements and individual nutrients were not included in our searches, which could have been relevant e.g.. for determining which bacterial strains (in probiotics) may provide beneficial effects on plasma lipoproteins. Also, we did not search gray literature or contact authors for additional information, and only studies published in English were included. Thus, we cannot rule out that additional relevant studies may exist. In addition, the present study could not confidently determine effect sizes or required daily doses for the included foods, because of the scarce data available.

CONCLUSIONS
The results from the included RCTs are mostly in line with current guidelines for the treatment of dyslipidemia or prevention of CVD (7,44) in other populations. However, it is not possible from the present data to determine which foods significantly modify blood lipids in NAFLD. The possible beneficial effect of probiotics enriched with certain bacterial strains on plasma lipoproteins warrants further study.

DATA AVAILABILITY STATEMENT
The original contributions presented in the study are included in the article/supplementary material, further inquiries can be directed to the corresponding author.

AUTHOR CONTRIBUTIONS
The study was conceptualized by DI. AR and FR performed screening. AR, FR, and DI performed data extraction. Risk of bias was assessed by FR and DI. Data analysis and grading of the evidence was performed by DI, with support from AR and FR. The paper was written by AR, FR, and DI. All authors contributed to the article and approved the submitted version.