A challenge in understanding the gut microbiome’s role in health and neurological disease is linking changes in the microbiome itself and its derived metabolome to changes in host metabolism and cognitive symptoms. Here, we present a personalised metabolic modelling approach for the interrogation of the host-microbiome co-metabolism in AD. Through this framework, we used whole-body models of human metabolism to generate insights about the consequences of altered formate microbial secretion and its association with urine output through host-microbiome co-metabolism. These in silico investigations were motivated by findings of alterations in formate urine levels in AD from an independent cohort. Taken together, our analyses point towards altered gut microbiome secretion capacities and host-microbiome interactions on formate production, results that could help understand the crosstalk between host and microbiome metabolisms in AD.
The use of urine metabolomics, as a non-invasively collectable biofluid, in the context of AD and the identification of early markers of AD could be crucial to developing successful therapies 34,35. Notably, urinary formate has been already suggested as a new potential biomarker for Alzheimer’s disease by an independent study 36. As a final breakdown product of human and microbial metabolism, formate is typically found in human urine 37. Levels are affected by both environmental and dietary exposures, and our results suggest that the urinary formate secretion could be a direct effect of altered host and microbial formate co-metabolism 25. The interrogation of microbiome-personalised sex-specific WBM models, through computation of maximum formate urine secretions (Fig. 4), highlighted that microbial formate secretion capacity was lower in AD microbiomes. Our results suggested the involvement of host-microbiota co-metabolism in the overall formate urine secretion catabolising microbially derived metabolites, such as glucose, L-serine, L-glycine, L-tryptophan, L-cysteine, L-tyrosine, and ornithine. The tyrosine pathway has been repeatedly implicated in AD 38,39 and tryptophan metabolites have been shown to regulate the cerebral activity of neprilysin, a metalloproteinase that controls the degradation and clearance of Aβ peptides in the brain 40.
Our results further underline altered metabolism as a hallmark of AD 41,42. In particular, the WBMs revealed the role of amino acid degradation in formate production, highlighting microbial-derived tryptophan degradation as one of the primary microbial sources of formate. Tryptophan is not only an important precursor of neurotransmitters and neuroactive metabolites, such as serotonin and kynurenine 43, but it also plays a role in immunoregulation 4. Moreover, tryptophan depletion increases cognitive deficits among people with AD 44 and the bioavailability of metabolites in the serotonin and kynurenine pathways are altered in both the urine and serum of AD patients 45. The microbiome modelling implicated that microbial tryptophan production may also be reduced in AD, concurring with earlier work indicating that the microbiome contributes to human tryptophan pools 21. Thus, in conjunction with the WBMs, our finding of decreased formate among individuals with AD and MCI suggests alterations in tryptophan degradation in AD. These results are also in line with a recently formulated hypothesis of AD being a tryptophan metabolism-correlated disease 46.
Formate is also involved in different pathways and is a precursor of purine synthesis 47. Our study also highlighted the association of formate with folate metabolism, a pathway that has been found associated with AD and DNA methylation 48. The importance of host-microbiome formate co-metabolism is further highlighted by our examination of genes associated with AD, where we found that most of the reactions involved in formate metabolism (Fig. 4) belong to genes, whose expression was altered in AD participants compared to healthy controls. Five reactions associated with the folate metabolism (MTHFDm, MTHFDm2, MTHFCm, MTHFD, MTHFD2) were associated with genes differently expressed in AD, three reactions (VMH IDs: PGCD, PSERT, and PSP_L), which are involved in the catabolism of 3-phosphoglycerate through an alternative pathway from glycolysis with the production of L-serine, a possible precursor of formate, were associated with genes overexpressed in AD; this result could corroborate the reported reduction in glycolysis intermediate concentrations in AD participants 49. Overall, these results suggest that formate metabolism is altered in individuals with AD also due to genetic variations, which could lead to changes in formate urine secretion in addition to microbial formate production. Notably, this inference would have not been possible without the WBM modelling, which clarified the role of the host formate metabolism. Without this additional in silico analysis, one could have falsely concluded that changes in microbial formate production, due to differences in microbiome composition, would be responsible for the reduced urine formate secretion in AD patients as measured in the metabolomic data. Since our WBM models were not further personalised using an individual’s genomic, metabolomic, or transcriptomic data, the aforementioned host genetic factors should be considered in future in silico studies, potentially increasing the validity of the in silico results regarding host urinary formate secretion.
While the COBRA modelling approach is a very valuable approach for investigating host-microbiome co-metabolism involvement in AD, certain limitations should be noted. For instance, we used a relatively small cohort of subjects and controls. Hence, our results need to be validated in larger independent cohorts. It has to be noted that discarded microbes not accounted for by the AGORA2 resource could lead to loss of metabolic capacities in the correspondent microbiome models. Additionally, the models were built on microbial relative abundance data generated from the reads count data, and results were subject to the intrinsic compositional structure of the models. Genome-scale metabolic reconstructions are also continuously updated as new experimental data and biochemical knowledge become available 50–52. The incompleteness of the metabolic reconstructions is particularly true for gut microbes, for which only limited data are available. Computational reconstruction tools, such as DEMETER 53, which has been used for the construction of the AGORA2 microbial reconstructions, permit the inclusion of experimental data, e.g., from BacDive 54, during the reconstruction process. Similarly, refined genome annotations that correct missing and mis-annotations should be performed to minimise the errors in the reconstruction, and thus increase the fidelity of the predictions. Such reannotation has been done for most of the microbes in AGORA2. Moreover, an inherent limitation of the COBRA approach is that it assumes the biological system to be in a steady-state condition, thereby ignoring the dynamic nature of microbial communities. Notably, the predicted secretion capacities obtained through microbiome modelling are not confounded by different factors, such as age and sex, improving the identification of secretion-microbial correlations, since they are derived from deterministic modelling rather than inferred from statistical dependence patterns of observational data. Additionally, while this study highlighted the role of gut microbes and host metabolism and genetics, differences between lifestyle factors (e.g., diet and exercise) and medications are also likely to contribute to changes in host-microbiome co-metabolism, urine metabolome, and AD pathology.
In conclusion, in this study, we combined omics data with COBRA modelling on the level of the microbiome and the whole human supra-organism and highlighted the role of microbiome-host interplay on formate-producing pathways. In particular, the microbiome’s role in linking aminoacidic and glucose metabolism with formate, a possible early marker for AD, could be of clinical importance, potentially contributing to the AD phenotype. The underlying mechanism suggested by our model, that both gut microbes and host genetics contribute to an altered formate metabolism in AD, needs to be assessed with more targeted validation studies. Our study delivers proof of the concept of personalised whole-body modelling in the context of a complex human disease. As such, the paradigm has demonstrated promise in uncovering host-microbiome co-metabolism involving biomarkers found in metabolomic studies validating or suggesting pathology hypotheses.