MS2 · Six-Tier Evidence Framework

How the Evidence Monitor grades a paper: six tiers from mechanistic in-vitro work (L1a) up to intervention validation in the target host (L4), constrained throughout by two dimensions — strain specificity and target-host validation.

Applies to: the gut microbiome co-metabolism channel (Evidence Monitor) only. The eight bioinformatics channels are graded by AI significance rating (High / Medium / Low) instead, and do not use this hierarchy.

1. Six-Tier Evidence Assessment Framework

TierNameDefinitionQuestion Addressed
L1aMechanistic evidenceIn vitro pure-culture metabolic assays, ex vivo tissue or cell-based experiments, and indirect mechanistic evidence from heterologous species within the same genus. Example: in vitro demonstration that P. succinatutens converts succinate to propionate via the methylmalonyl-CoA pathway.Does this bacterium have the biochemical capacity to perform the claimed function under controlled conditions?
L1bAssociation inferenceGenus-level co-occurrence network analyses, abundance–phenotype statistical associations, metagenomic functional inferences. Example: negative correlation between Phascolarctobacterium relative abundance and Enterobacteriaceae in piglet fecal metagenomes.Is there a statistical signal linking this taxon to the outcome of interest?
L2Moderate inferenceDirect validation of the effector molecule (e.g., propionate) in a porcine model, with mechanistic pathway data, but without strain-specific attribution. Example: porcine dietary fiber intervention → propionate → TLR4/NF-κB suppression → tight junction protein upregulation, without demonstrating that P. succinatutens produced the propionate in situ.Does the effector mechanism operate in the target host?
L3Strong inferenceCausal validation through mono-colonization with this bacterium in murine models using human-derived strains. Example: germ-free mouse mono-colonized with human-derived P. succinatutens, showing causal propionate production, TLR4/NF-κB pathway inhibition, and intestinal barrier maintenance.Can the bacterium, in principle, execute the claimed function in a mammalian gut?
L3.5Transitional validation [NOVEL TIER]Mono-colonization using human-derived strains in a porcine model. Deconvolves host-species transfer from strain-origin transfer. Example: gnotobiotic piglet mono-colonized with the human-derived P. succinatutens strain used in the L3 murine study.Does the mechanism transfer across host species?
L4Established evidenceMono-colonization in weaned piglets using a porcine-derived strain, with independent replication. For metabolically dependent taxa (e.g., P. succinatutens requiring exogenous succinate), requires demonstration of adequate substrate availability; defined co-colonization with a succinate-producing consortium may supplement mono-colonization designs.Does the mechanism operate in the target host under ecologically realistic conditions?
Critical epistemological distinction. Levels 1–3 address mechanistic plausibility under controlled conditions, whereas Level 4 addresses translational efficacy under ecologically realistic conditions. These are qualitatively different scientific questions, not merely points on a continuum of increasing evidential strength. A candidate that achieves L3 in murine models may fail at L4 not because the mechanism is wrong, but because ecological factors in the porcine gut—substrate availability, microbial competition, immunological context—differ fundamentally from the gnotobiotic murine environment.

2. The Level 3.5 Innovation

When a probiotic candidate validated in murine models (L3) fails in the target livestock host, the failure cannot be attributed to either (a) host-species differences between mouse and pig, or (b) strain-origin differences between the human-derived strain used in mice and a hypothetical porcine-derived strain. Both variables changed simultaneously, confounding the interpretation. Level 3.5 deconvolves these two factors:

Deconvolution logic. Hold strain origin constant (human-derived) while varying host species (mouse → pig). A positive L3.5 result demonstrates that the mechanism transfers across host species, isolating strain origin as the remaining barrier to L4. A negative L3.5 result identifies host-species incompatibility as the primary barrier, directing resources away from costly porcine-derived strain isolation toward mechanistic investigation of species-specific factors.

3. Evidence Challenge Principle

The framework includes a provisional mechanism for dynamic evidence updating. Recognizing that evidence tiers are static snapshots at a single point in time, an evidence challenge principle is proposed: any tier assignment may be challenged by new experimental data, and a provisional workflow defines how evidence upgrades and downgrades should be adjudicated. A formal Bayesian belief-updating extension is deferred to future work. This principle ensures the framework remains a living tool rather than a static classification.

4. Scope and Limitations

The framework has been formally applied to one candidate (P. succinatutens) with two illustrative comparators. Broader validation across a larger panel of candidates is needed to establish generalizability. Tier assignments reflect the highest experimental design type available, not study quality within a tier (a murine mono-colonization study with n = 3 receives the same L3 classification as one with n = 30, though they carry different statistical weight). The framework evaluates functional claims; it does not assess safety, which requires independent toxicological and microbiological evaluation.

5. Relationship to MS1

This framework is complementary to the Compartmentalized Co-Metabolism Network Model (MS1). MS1 provides a biological model of bidirectional host–microbiome interactions across three gut compartments, with a four-dimensional matrix rating system for diagnostic evidence mapping at the metabolite-node level. MS2 provides a hierarchical evidence assessment framework for evaluating translational readiness of specific probiotic candidates at the strain level. The two frameworks are layered: MS1 diagnoses where evidence gaps exist across a co-metabolism network; MS2 diagnoses how close a specific candidate is to translational readiness within that network. Together they form the methodological foundation for the Gut Microbiome Co-Metabolism Evidence Monitor on bioTriage-Open.

6. Methodology References

The six-tier framework is anchored to established evidence-grading and probiotic-evaluation standards rather than to ad hoc criteria.

Anchored to established standards. Level assignment follows the hierarchy-of-evidence tradition (Evans 2003; GRADE — Guyatt 2008, Balshem 2011); the strain-specificity and target-host requirements follow the ISAPP probiotic consensus (Hill 2014; Binda 2020); the evidence-mapping protocol follows PRISMA-ScR (Tricco 2018).

The causal-inference and review-quality references that underpin the coupling and measurement-depth dimensions (Bradford Hill, commensal Koch’s postulates, AMSTAR 2, PRISMA 2020) are listed in MS1, section 7 (Methodology References).