Compositional alterations of gut microbiota in nonalcoholic fatty liver disease patients: a systematic review and Meta-analysis

This systematic review and meta-analysis is reported following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) Statement [11] and the PRISMA 2009 checklist, which has been with a registration code in PROSPERO registry (registration number is: CRD42020220632, https://www.crd.york.ac.uk/PROSPERO↗). A complete and computerized search (last update: December 30, 2020) was performed without restriction based on region, language or publication type in the following electronic databases: PubMed, EMBASE and Cochrane Library. The Medical Subject Heading (MeSH) database was used to acquire MeSH terms and their relevant entry terms based on the following formatted terms: (NAFLD OR NASH OR NAFL OR “nonalcoholic fatty liver disease” OR “nonalcoholic steatohepatitis” OR “nonalcoholic fatty liver disease” OR “nonalcoholic steatohepatitis” OR “fatty liver” OR “steatohepatitis” OR “steatosis”) AND (microbiota OR microbes OR microbiome OR microflora OR flora OR bacteria). The query was adapted to different databases with minimal differences as required. The literature search was also supplemented with a manual search using the Related Articles Function and the reference lists of all retrieved articles.

Any study that provides necessary data on detecting gut microbiota for NAFLD patients and comparable controls was eligible for this systematic review and meta-analysis. The inclusion criteria for this study were as follows: 1) original full-text publications, 2) NAFLD diagnosed with sonographic or histologic evidence, 3) healthy controls with comparable age and sex proportions as the case group, 4) studies comparing the gut microbiota between NAFLD patients and healthy controls, and 5) available and sufficient data (sample size, mean and standard deviation or any data that can be converted to) to calculate the standardized mean difference (SMD) with 95% confidence interval (CI) in these two groups.

The exclusion criteria were as follows: 1) editorials, letters to the editor, reviews, case reports, hypotheses, book chapters, conference abstracts or studies on animals or cell lines; 2) studies that included NAFLD patients with other liver diseases, such as alcoholic fatty liver disease, autoimmune liver disease and viral hepatitis; and 3) studies that performed any intervention on participants at baseline.

Two investigators (FX L and JZ Y) were responsible for selecting studies and extracting data from the included studies independently according to predesigned inclusion and exclusion criteria. Any disagreement was discussed and finally resolved by the adjudicating senior author (BH Z). If additional data or data transformations were required for analysis, the corresponding or first authors were contacted by email for assistance. When the authors did not reply, standard statistical formulas were applied for data transformation [12, 13]. If important additional data were not available from the corresponding or first author, the study was excluded.

A full-text review of the included studies was performed independently by the two aforementioned investigators, and the following variables were collected: 1) first author; 2) year of publication; 3) location; 4) study design; 5) number, age and sex of NAFLD (or NAFL or NASH) patients and healthy controls; 6) methods for diagnosing NAFLD; 7) methods for detecting and analyzing gut microbiota; and 8) bacterial counts and units for expressing the values.

To evaluate the risk of bias of individual studies, the methodological quality of the included studies was assessed using the Newcastle-Ottawa Scale (NOS) by two investigators (FX L and CX S). A senior author (BH Z) who was not involved in the initial assessment was responsible for adjudication on any disagreement between the two investigators. The NOS is a widely and frequently used tool for assessing the quality of nonrandomized studies included in systematic reviews and/or meta-analyses. Using this tool, study quality is assessed based on eight items that are categorized into three groups: study group selection, group comparability (a maximum of two stars can be given for comparability), and exposure or outcome of interest determination for case-controlled or cohort studies. A star rating system was used to rate the quality of the included studies based on a scale ranging from 0 (low quality) to 9 (high quality).

In this study, statistical analyses were performed using R language Version 3.4.3 (The package of meta and metafor) and STATA/SE 12.0 (StataCorp, Texas, USA) with a Meta module for Windows. After extraction and transformation, all bacterial count data are presented as the mean (x¯), standard deviation (SD) and total sample size (n). The standardized mean difference (SMD) with 95% confidence interval (CI) was adopted to evaluate the effect size when a bacterial genus was detected by different techniques in the included studies [14]. The α criterion was set at 0.05. Accordingly, when the total SMD of a bacterial genus was more than 0 (favors NAFLD) with a 95% CI not including the value 0, we assumed that this bacterial genus exhibited overgrowth in NAFLD patients. Otherwise, the bacterial genus was assumed to be deficient in NAFLD patients. Additionally, subgroup analysis was performed based on the areas where these studies were conducted (Western and Eastern). Meta regression analysis was also performed to evaluate the effect of alanine aminotransferase (ALT) and body mass index (BMI) levels on mean different abundance of bacterial genus for NAFLD versus control patients. Those univariate significant variances would be entered in the multivariate models.

Heterogeneity across the included studies was evaluated using χ² and I² statistics. Higher χ² and I² statistics indicated greater heterogeneity, and an I² value greater than 50% with a P-value less than 0.1 represented substantial heterogeneity in this analysis. A random-effects model was reported when substantial heterogeneity existed. The 95% prediction intervals which reflected the true effects of future similar studies with 95% certainties were calculated based on the formulas reported by Joanna IntHout et al. [15].

To assess the risk of publication bias, funnel plot analyses, Egger’s test and Begg’s test were conducted for genus-specific gut microbiota. Publication bias can cause asymmetry in funnel plots, which can be quantified with Egger’s and Begg’s tests. When P-values derived from Egger’s or Begg’s regressions were greater than 0.05, we assumed that no publication bias was present. Moreover, a sensitivity analysis was performed to evaluate the impact of each study on the overall results. In addition, considering the observational design limitations of each included study despite of its large sample size and quality, there would be the maximum certainty of an effect of case-control comparison, which was defined as credibility ceiling. Credibility ceilings tests setting different degrees of credibility to further adjust the included study in a meta-analysis would favor lessening the reporting of exaggerated associations with statistic significant results, therefore decrease the unmeasured bias. A sensitivity analysis based on ceiling value(s) of 0, 20 and 40% were conducted for ccredibility ceilings tests as the literature reported [16].

Using the predefined search strategy, 2352 citations were identified, as shown in Fig. 1. No additional records were identified through other sources. After removing duplicates, 1890 citations were screened by reviewing titles and abstracts. Of the remaining citations, 1862 articles were excluded because of unsuitable article type or studies on animals or cell lines or irrelevance. Then, the full texts of the remaining 28 articles were retrieved and assessed for eligibility. Of these full-text studies, 2 were excluded for using culture-dependent methods for detecting gut microbiota, 6 were excluded for not including a control group, and 5 were excluded because the authors did not provide necessary data. Finally, 15 studies published from 2012 to 2020 were included in the meta-analysis [7, 17–30]. Additional examination of the references listed in these studies did not provide any additional eligible studies. Agreement between the two investigators on study selection was 93.3%.

The main characteristics of the 15 included studies are listed in Table 1. Among the 15 studies consisting of 1265 individuals (577 NAFLD patients and 688 controls), six studies were performed in China, including one in Taiwan, one in Hong Kong, two in the USA, two in Canada, one in Thailand, one in Turkey, one in Spain, one in Italy and one in Korea. Nine of the 15 included studies adopted liver biopsy to diagnose NAFLD. All of the studies collected stools of the patients and healthy controls as samples for gut microbiota detection and utilized 16S rRNA sequencing.

As shown in Table 2, quality assessment of the included studies was carefully performed using the NOS, and the general quality was moderate (six studies scored 8 points, seven scored 7 points, and two scored 6 points, mean ± SD: 7.27 ± 0.70). Agreement between the two investigators for quality assessment was 93.3%. No study was excluded due to poor NOS score.

The included studies analyzed different taxa of gut microbiota in NAFLD patients at different levels. Based on available data, this meta-analysis primarily focused on alterations at the genus level, including Escherichia, Prevotella, Streptococcus, Coprococcus, Faecalibacterium, Ruminococcus, Bacteroides, Bifidobacterium, Blautia, Clostridium, Dorea, Lactobacillus, Parabacteroides and Roseburia.

As noted in Fig. 2, the total SMDs of the genera Escherichia, Prevotella and Streptococcus were 1.55 [95% CI: 0.57, 2.54], 1.89 [95% CI: 0.02, 3.76] and 1.33 [95% CI: 0.62, 2.05], respectively, revealing that all these genera exhibited increased abundance in NAFLD. Significant heterogeneities were noted in the included studies of the three genera with I² values greater than 50% and p values less than 0.1. In the subgroup analysis, the results of the two subgroups of the genus Escherichia were consistent with the overall results, while in the analysis of the genera Prevotella and Streptococcus, the results of the Eastern subgroups were not significantly different.

In Fig. 3, the total SMDs of the genera Coprococcus, Faecalibacterium and Ruminococcus were − 1.75 [95% CI: − 3.13, − 0.37], − 13.23 [95% CI: − 17.59, − 8.87] and − 1.84 [95% CI: − 2.41, − 1.27], respectively, revealing that these genera were deficient in NAFLD patients. Significant heterogeneities were also noted. In the subgroup analysis, neither subgroup of the genus Coprococcus showed significant differences, the Western subgroup of the genus Faecalibacterium did not reach a significant difference either, and two subgroups of the genus Ruminococcus were consistent with the overall results.

Regarding the genera Bacteroides, Bifidobacterium, Blautia, Clostridium, Dorea, Lactobacillus, Parabacteroides and Roseburia (Fig. 4), no significant differences were observed between the NAFLD patients and healthy controls in the overall analysis. Interestingly, in the subgroup analysis, the abundance of the genus Bacteroides in the NAFLD patients was decreased in the Western population but increased in the Eastern population. Moreover, the genera Parabacteroides and Roseburia in the NAFLD patients presented decreased abundance in the Western and Eastern subgroups, respectively.

Supplemental Figureand Supplemental Tableprovide an analysis of the funnel plots and Egger’s and Begg’s tests for publication bias. According to Egger’s and Begg’s tests, none of the analyses presented publication bias. 1 1

The sensitivity analysis (Supplemental Figure 2) showed that omitting each single study did not significantly change the results in the analysis for the genera Escherichia, Streptococcus, Faecalibacterium and Ruminococcus. However, after removing some of the included studies, the results of the genera Prevotella and Coprococcus did not reach significant differences. The results of the credibility ceiling test for ceilings of 0–40% are shown in Supplemental Table 2. All 3 meta-analyses that retained statistical significance with c = 20% or 40% for abundance changes of genera [Coprococcus, Faecalibacterium, Ruminococcus] in NAFLD. The meta-analysis on Escherichia, Prevotella and NAFLD none survived a 20% ceiling, whereas the Coprococcus remained statistically significant even with 20% ceiling but not 40% ceiling.

On univariate meta-regression model (Supplemental Figure 3), when plotting log odds ratio of mean different abundance of bacterial genus for NAFLD versus control patients (y-axis) against alanine aminotransferase (ALT) or body mass index (BMI) levels age (x-axis), significant association was found between BMI and Faecalibacterium (log regression coefficient β = 2.38, P = 0.00001), and Prevotella (regression coefficient β = 1.0554, P = 0.0005). While ALT levels was associated with different abundance of Faecalibacterium (log regression coefficient β = 1.11, P = 0.001), Prevotella (log regression coefficient β = 0.2018, P = 0.031), Streptococcus (log regression coefficient β = 0.09, P = 0.041). After multivariate adjustments, BMI remained significant for Prevotella (log regression coefficient β = 1.20, P = 0.019), while both BMI and ALT were significant for Faecalibacterium (log regression coefficient β = − 6.2, P = 0.00001 and β = 2.51, respectively, both P < 0.00001).

This study is the first to identify microbial signatures in NAFLD patients via a meta-analysis. A subgroup analysis was also conducted by classifying the research locations to validate the generalities of the results in the overall generation. Certain limitations were observed in this meta-analysis. First, the included studies only analyzed gut microbiomes recovered in the stool; however, several studies have indicated that mucosa-associated bacteria might greatly differ from those recovered in stool and could play a more important role in the pathogenesis of associated diseases [31, 32]. In addition, most of the analyses presented significant heterogeneities, which can only be minimized but not eliminated using a random-effects model. Third, several eligible studies [37–41] were excluded because of lack of necessary data. Fourth, the sensitivity of Egger’s and Begg’s test is low in the context of small sample size (most of which does not exceed 100) of the included studies. Last but not least, meta-regression was performed for BMI and ALT values based on aggregate data; therefore, the risk of ecollogic fallacy exists.

All data generated or analysed during the current study are included in this published article and its supplementary information files.