Integration of constraint-based modelling with faecal metabolomics reveals large deleterious effects ofFusobacteriaspecies on community butyrate production

Abstract

SummaryIntegrating constraint-based community modelling with population statistics, we introduce new theoretical concepts for interrogating the metabolic functions of the microbiome, applying them to a public metagenomic dataset consisting of 365 colorectal cancer cases (CRC) and 251 healthy controls. We found that 1) glutarate production capability was significantly enriched in CRC microbiomes and mechanistically linked to lysine fermentation in Fusobacteria species, 2) acetate and butyrate production potentials were lowered in CRC, 3) Fusobacteria presence had large negative ecological effects on community butyrate production in CRC and healthy controls. Validating the model predictions against faecal metabolomics, our in silico frameworks correctly predicted in vivo species metabolite correlations with high accuracy. In conclusion, highlighting the value of combining statistical association studies with in silico modelling, this study delivers insights on the metabolic role of Fusobacteria in the gut, while providing a proof of concept for the validity of constraint-based community modelling.

Causal inference on microbiome-metabolome relations viain silico in vivoassociation pattern analyses

“…The easiest explanation for this pattern is therefore to interpret the secretion potential as a metric of net consumption. In an earlier work [25], the validity of this interpretation was demonstrated for glutarate.…”

“…COBRA microbial community modelling allows calculating the direct contribution of a species to the metabolic net production profile of a community, thereby quantifying the direct and indirect metabolic causal effects [17]. Additionally, it allows for quantifying the ecological effects as well, although no inference on the nature (i.e., causal vs. confounding) of the ecological effects can be made from community modelling alone [25]. In essence, all confounding and causal pathways, which lead to correlation among species abundances, impact the output of COBRA microbial community models, explicitly including ecological causality and the two types of ecological confounding (Fig 1).…”

“…However, one must be careful, as a species can have a positive direct contribution to the net secretion potentials, while displaying large negative ecological effects. An example of this was given in Hertel et al [25], where Fusobacterium sp. were demonstrated to contribute small amounts of butyrate to faecal butyrate pools, while having large deleterious effects on community butyrate production.…”

“…containing communities resulted in an overall negative impact of Fusobacterium sp. on community butyrate production [25].…”

“…It is worth noting that in silico in vivo pattern analyses, strictly spoken, allows only for an inference on whether the microbial community as a whole is causally related to a metabolite. As COBRA microbial community modelling, at least in the herein applied form, cannot differentiate between ecological causation and ecological confounding [25], causal inference on the species level, strictly spoken, is not possible. One also may argue that causal inference on single species is not sensible due to systems nature of microbial communities, making the concepts of causal inference on the species level more difficult to apply.…”