Nonalcoholic fatty liver disease (NAFLD) is a prevalent chronic liver disease, closely associated with type 2 diabetes (T2D) through insulin resistance. Interventions targeting this shared pathophysiology are crucial. The gut microbiota plays a significant role in metabolic health and is implicated in NAFLD and T2D. Dysbiosis, or imbalance in the gut microbiota, is linked to these conditions, potentially through inflammation, intestinal membrane damage, and bacterial translocation. While currently, increased physical activity and dietary modifications are the only effective therapeutic options for NAFLD management, the mechanisms are not fully elucidated. Studies show a low-carbohydrate diet (LCD) improves fatty liver metabolism and alters gut microbiota, and regular aerobic exercise reduces hepatic fat and increases gut microbial diversity. However, individual responses to these interventions vary, with some subjects showing greater improvement than others. This variability may mask changes in the microbiome using traditional statistical methods. Keystone taxa, which drive microbiome structure and function, and their interaction networks are important for understanding microbial functions and disease progression. This study aimed to analyze gut microbiota composition and metabolic pathways, construct a co-occurrence network at the population level, and develop personalized gut microbial networks to identify microbial signatures and interactions at a personal level, differentiating responders from low/non-responders to interventions.

Existing literature demonstrates a strong link between gut microbiota dysbiosis and NAFLD and T2D. Studies have shown that dietary interventions, particularly low-carbohydrate diets, can positively impact liver fat metabolism and alter the composition of the gut microbiome. Similarly, aerobic exercise has been shown to reduce hepatic fat and improve gut microbial diversity. However, there is a lack of understanding regarding how these interventions interact with the gut microbiome at the individual level, especially in patients with NAFLD and prediabetes. The impact of keystone taxa and their interaction networks within the gut microbiome on the response to interventions remains largely unexplored. Prior research has highlighted the significant inter-individual variability in response to exercise and dietary interventions, suggesting the need for personalized approaches. This study bridges this gap by investigating the effects of exercise and diet on both the population-level and individual-level gut microbiome networks in the context of NAFLD and prediabetes.

This study was an 8.6-month, four-arm, randomized, single-blinded (for researchers) controlled trial. 115 participants with NAFLD and prediabetes were randomized into four groups: aerobic exercise (AEx), fiber-enriched low-carbohydrate diet (Diet), aerobic exercise + diet (AED), and no intervention (NI). 85 participants completed the trial. Fecal samples were collected at baseline and after the intervention for 16S rRNA gene sequencing (n=76) and metagenomics (n=42). Data analysis included assessment of alpha and beta diversity, identification of differentially abundant amplicon sequence variants (ASVs) using LEfSe analysis, correlation analysis between ASVs and clinical biomarkers (HFC, HbA1c, SCFAs), KEGG pathway analysis of metagenomics data, co-occurrence network analysis to identify keystone taxa, and development of personalized gut microbial networks using a Single SparCC network method to predict individual responses to interventions. Responders were defined as those with >5% reduction in hepatic fat content (HFC).

The study found that the combined aerobic exercise and diet intervention (AED) was associated with increased diversification and stabilization of keystone taxa within the gut microbiome. In contrast, exercise and diet interventions alone resulted in increased network connectivity and robustness among gut microbial taxa. The no intervention (NI) group showed a significant decrease in gut microbial alpha diversity (Shannon index). LEfSe analysis revealed significant differences in ASVs between intervention groups and the NI group, with several ASVs from *Bacteroides* and *Ruminococcus* showing enrichment in intervention groups. Correlation analysis showed that certain ASVs were associated with changes in HFC and HbA1c levels. KEGG pathway analysis of metagenomic data indicated significant differences in metabolic pathways between intervention and NI groups, with changes in carbohydrate, energy, lipid, and amino acid metabolism. Co-occurrence network analysis showed increased connectivity and robustness in intervention groups. Importantly, the personalized gut microbial network analysis, using a novel Single SparCC method, demonstrated that baseline network properties (edge number) could predict the individual response to exercise intervention (particularly in AEx and AED groups), while ASV abundance had better predictive power for the Diet group.

This study demonstrates the impact of exercise and diet interventions on the gut microbiome in NAFLD and prediabetes patients at both population and individual levels. The findings highlight the importance of considering both the composition and interaction networks of the gut microbiota when designing personalized interventions. The increased network connectivity and robustness observed in intervention groups suggest a healthier and more stable microbial ecosystem. The ability to predict individual response to intervention using baseline gut microbial network characteristics has significant clinical implications, potentially allowing for tailored treatment strategies to optimize therapeutic outcomes. The discrepancy in predictive power between ASV abundance and network properties suggests that the interaction dynamics within the gut microbial community are critical factors influencing treatment efficacy, highlighting the need to incorporate network-based approaches in future research.

This study demonstrates that exercise and diet interventions differentially impact gut microbiome composition and network structure in NAFLD patients. Combined interventions stabilize keystone taxa, while individual interventions enhance network connectivity. A novel Single SparCC network method enables prediction of individual responses to exercise, highlighting the potential for personalized interventions targeting the gut microbiome. Future research should focus on validating these findings with larger cohorts and investigating the mechanisms underlying these effects.

The relatively small sample size, particularly for the metagenomics data, limits the generalizability of the findings. The retrospective registration of the trial is a limitation. The diverse dietary and exercise intervention methods across different trials pose challenges for cross-study comparisons. The lack of independent validation for the predictive model based on personalized gut microbial networks represents a limitation.