BACKGROUND

Fatigue is a prevalent and debilitating symptom of non-communicable diseases (NCDs); however, its biological basis are not well-defined. This exploratory study aimed to identify key biological drivers of fatigue by integrating metabolomic, microbiome, and genetic data from blood and saliva samples using a multi-omics approach.

METHODS

Metabolomic, microbiome, and single nucleotide polymorphisim analyses were conducted on saliva and blood samples from 52 patients with NCDs. Fatigue dimensions were assessed using the Multidimensional Fatigue Inventory and correlated with biological markers. LightGBM, a gradient boosting algorithm, was used for fatigue prediction, and model performance was evaluated using the F1-score, accuracy, and receiver operating characteristic area under the curve using leave-one-out cross-validation. Statistical analyses included correlation tests and multiple comparison adjustments (p < 0.05; false discovery rate <0.05). This study was approved by the Yokohama City University Hospital Ethics Committee (F230100022).

RESULTS

Plasmalogen synthesis was significantly associated with physical fatigue in both blood and saliva samples. Additionally, homocysteine degradation and catecholamine biosynthesis in the blood were significantly associated with mental fatigue (Holm p < 0.05). Microbial imbalances, including reduced levels of Firmicutes negativicutes and Patescibacteria saccharimonadia, correlated with general and physical fatigue (r = - 0.379, p = 0.006). Genetic variants in genes, such as GPR180, NOTCH3, SVIL, HSD17B11, and PLXNA1, were linked to various fatigue dimensions (r range: -0.539-0.517, p < 0.05). Machine learning models based on blood and salivary biomarkers achieved an F1-score of approximately 0.7 in predicting fatigue dimensions.

CONCLUSION

This study provides preliminary insights into the potential involvement of alterations in lipid metabolism, catecholamine biosynthesis disruptions, microbial imbalances, and specific genetic variants in fatigue in patients with NCDs. These findings lay the groundwork for personalized interventions, although further validation and model refinement across diverse populations are needed to enhance the prediction performance and clinical applicability.