Kinetic-Gated Enrichment of Microbial Metabolite Signaling

When a gut microbe releases a small molecule, that molecule does not knock politely on one door. Butyrate, at millimolar concentrations in the colon, simultaneously inhibits histone deacetylases (the enzymes that close down chromatin), activates at least three cell-surface G-protein coupled receptors, and shifts intracellular pH. Which of these engagements actually drives a downstream transcriptional program is almost impossible to tell from a single-endpoint RNA sequencing experiment, because every arm fires at once. Meanwhile, the biophysical community has spent decades measuring exactly the kind of information that would disentangle this — Surface Plasmon Resonance (SPR), an optical technique that immobilizes one binding partner on a gold film and tracks, in real time and without labels, how fast a partner binds and detaches — but that information is never carried into genome-wide pathway analysis. The closest existing integration, an anti-SARS-CoV-2 screen of Qingyan Dropping Pills, used SPR purely as a yes/no filter on binding before falling back on targeted endpoint assays, with no RNA-seq and no pathway statistics Elucidation of anti-SARS-CoV-2 and anti-inflammatory bioactives in Qingyan Dropping Pills via integrated in silico screening and bioactivity validation.

The claim of this proposal is that SPR binding kinetics can be used, quantitatively, as the input weights of a topology-aware functional enrichment analysis, converting a metabolite–receptor binding time-course into a predicted schedule of pathway activation that can be tested against time-course RNA-seq. Done well, this gives the microbial-metabolite-signaling field the causal disentangling tool it has repeatedly called for Gut microbial metabolites in inflammatory bowel disease: immunological mechanisms regulating Treg/Th17 balance and therapeutic potential, Causal discovery for the microbiome.

Core Claim

For a metabolite $m$ at concentration $C$ binding a receptor with SPR-measured rate constants $k_{on}$ and $k_{off}$, receptor occupancy follows a simple, well-known saturation curve,

$$o(t) = \frac{C\,k_{on}}{C\,k_{on}+k_{off}}\left(1 - e^{-(C\,k_{on}+k_{off})\,t}\right),$$

so the onset time constant $\tau = 1/(C\,k_{on}+k_{off})$ and the residence time $1/k_{off}$ predict when and how persistently a receptor's downstream transcriptional program should switch on. I propose to inject these predicted occupancy schedules as edge weights and perturbation magnitudes into a topology-based enrichment method of the NetGSA family — which already models pathway graphs as weighted adjacency matrices and tests pathway activity conditional on network structure netgsa: Fast computation and interactive visualization for topology-based pathway enrichment analysis — and to validate the resulting "kinetic-gated" enrichment against time-course RNA-seq in which the same receptors are genetically or pharmacologically removed. The testable hypothesis is two-part: first, that kinetic gating (using $k_{on}$, $k_{off}$, and occupancy timing) ranks receptor-dependent pathways above receptor-independent pleiotropic programs better than affinity-blind enrichment does; second, that across metabolite–receptor pairs, pathway persistence correlates with residence time $1/k_{off}$ more strongly than with equilibrium affinity $K_D = k_{off}/k_{on}$ Molecular Interactions in Biological Systems: Technological Applications and Innovations.

Research Program

Three components, each addressing a known obstacle. First, an SPR measurement panel: roughly ten canonical microbial metabolites (acetate, propionate, butyrate, indole, indole-3-propionic acid, indole-3-aldehyde, deoxycholic acid, lithocholic acid, 3-oxoLCA, isoalloLCA) against seven host targets (FFAR2/GPR43, FFAR3/GPR41, HCAR2/GPR109A, FXR, TGR5, the AhR–ARNT complex, and RORγt), all catalogued as signaling nodes in the IBD and intestinal-homeostasis reviews Gut microbial metabolites in inflammatory bowel disease: immunological mechanisms regulating Treg/Th17 balance and therapeutic potential, The role of complex interactions between the intestinal flora and host in regulating intestinal homeostasis and inflammatory bowel disease. Two SPR-specific problems are handled directly. The small-molecule mass problem — a 60 Da analyte barely bends the signal of a 40–100 kDa immobilized receptor — is handled with competitive displacement assays: the receptor is pre-bound to a high-molecular-weight RNA aptamer reporter (aptamers of this kind, ~15–30 kDa, are routine affinity reagents Generation of RNA aptamers against chikungunya virus E2 envelope protein), and the metabolite's displacement of the reporter produces a large signal proportional to its occupancy; direct measurement is retained for the stronger binders, following the precedent of the Qingyan Dropping Pills study, which resolved natural-product $K_D$ values from $4.515 \times 10^{-8}$ M to $7.718 \times 10^{-6}$ M by SPR Elucidation of anti-SARS-CoV-2 and anti-inflammatory bioactives in Qingyan Dropping Pills via integrated in silico screening and bioactivity validation. The receptor-immobilization problem — amine coupling distorts fragile GPCRs — is handled by capturing receptors reconstituted in lipid nanodiscs on membrane-optimized sensor surfaces, an established SPR format Optimization of Surface Plasmon Resonance Biosensor for Analysis of Lipid Molecules, with the endogenous-modulator study of the AT1 receptor providing the structural template for pairing purified-receptor SPR with signaling readouts Cartilage oligomeric matrix protein is an endogenous β-arrestin-2-selective allosteric modulator of AT1 receptor counteracting vascular injury.

Second, the transcriptomic arm: time-course bulk RNA-seq (0.5, 1, 2, 4, 8, 24 hours) of human colonic epithelial cells and monocyte-derived dendritic cells treated with each metabolite at physiological concentrations, with parallel arms in which FFAR2, HCAR2, AhR, or FXR/TGR5 is antagonized or knocked out. The knockout arms are not a competitor — they are the ground truth: pathways enriched at early time points in the wild-type arm but collapsing in the knockout arm are receptor-dependent by construction. Third, the computational arm: the kinetic-gated enrichment itself. Each KEGG pathway graph is augmented with a synthetic metabolite node connected to its SPR-confirmed targets with edge weight $-\log_{10} K_D$, and the perturbation magnitude at each target follows the occupancy curve above; pathway test statistics then follow the NetGSA machinery, whose recent Restricted Haseman-Elston overhaul makes per-time-point analysis of dozens of pathways computationally trivial (the package reports analyzing the 520-sample TCGA breast cancer benchmark of 2,784 genes across 114 KEGG pathways in 4.60 minutes with clustering versus 167.83 minutes for the old estimator netgsa: Fast computation and interactive visualization for topology-based pathway enrichment analysis). Evidence for decomposing multi-layer evidence, and for expanding pathway gene sets beyond enzyme-only lists, is borrowed from ActivePathways' Brown's-test fusion of non-independent omics layers Integrative pathway enrichment analysis of multivariate omics data and MOMENTA's PPI-expanded pathway sets, which produced a 25-fold increase in significantly enriched metabolic pathways over enzyme-only GSEA Multiomic Metabolic Enrichment Network Analysis Reveals Metabolite–Protein Physical Interaction Subnetworks Altered in Cancer.

Baselines

The naive reference point is unweighted over-representation analysis — Fisher's exact test on thresholded differentially expressed gene lists, the default in the field. It is the reference because it is what essentially every metabolite-treatment RNA-seq paper currently reports, including the closest transcriptomic integration in this space, the morusin aortic-dissection study, which ran standard GO and KEGG over-representation on docking-predicted targets Integrated Network Pharmacology, Single‐Cell Transcriptomics Unveil the Mechanistic Role of Morusin in Aortic Dissection. It will be run on the same time-course data with the same thresholds and the same KEGG annotation release, so the only difference is the statistic.

The strongest competitor is unweighted NetGSA with the REHE estimator, run on identical count matrices, identical pathway graphs, and the same compute, with tuning budgets (clustering parameters, significance thresholds) matched. It is the strongest because the systematic benchmark of nine topology-aware methods found NetGSA and DEGraph outperform over-representation methods in statistical power and Type I error control, particularly in the small, sparse metabolic pathways that dominate metabolite signaling A comparative study of topology-based pathway enrichment analysis methods — though I note the benchmark reports no numeric power values found on Lacuna; the only quantitative figure I can anchor to is the runtime above. Beating it means higher area under the precision-recall curve against the knockout-defined ground truth at early time points, with Type I error held at the nominal 5% under sample-label permutation.

The key ablation replaces SPR-measured kinetics with two cheaper surrogates on otherwise identical pipelines: static docking energies (AutoDock Vina, the approach used in the morusin study, which reported binding energies of −9.0 to −7.0 kcal/mol Integrated Network Pharmacology, Single‐Cell Transcriptomics Unveil the Mechanistic Role of Morusin in Aortic Dissection), and randomly permuted SPR weights. If the SPR-weighted pipeline does not beat both, the kinetic-gating ingredient — as opposed to merely adding a metabolite node or any target prior — is not doing the work.

Why not just…?

Why not just use molecular docking instead of SPR? Docking is cheap and scales to the whole exposome. But docking returns a static energy, not rates; it cannot predict the occupancy onset time $\tau$ or the residence time that this proposal's gating and persistence hypotheses depend on, and docking-only pipelines have already been shown to leave engagement kinetics entirely unvalidated — the morusin study never confirmed any of its predicted bindings experimentally Integrated Network Pharmacology, Single‐Cell Transcriptomics Unveil the Mechanistic Role of Morusin in Aortic Dissection, while the Qingyan Dropping Pills work used SPR precisely because docking and target databases were insufficient on their own Elucidation of anti-SARS-CoV-2 and anti-inflammatory bioactives in Qingyan Dropping Pills via integrated in silico screening and bioactivity validation. That said, the docking ablation is the honest test: if docking weights match SPR weights in predictive accuracy, the expensive SPR arm is not earning its keep, and the proposal should retreat to docking priors plus a small SPR validation subset.

Why not just run standard over-representation analysis per time point? This is what a skeptical reviewer would call sufficient. The comparative benchmark answers it directly: over-representation methods lose power precisely on sparse metabolic pathways A comparative study of topology-based pathway enrichment analysis methods, they ignore graph structure and gene correlation, and — the deeper problem from the signaling literature — a single metabolite's gene list conflates direct receptor targets with secondary inflammatory cascades, the exact causality bottleneck the IBD review names as the field's central methodological failure Gut microbial metabolites in inflammatory bowel disease: immunological mechanisms regulating Treg/Th17 balance and therapeutic potential. The decisive experiment settles it: if unweighted per-time-point over-representation recovers the knockout-defined receptor-dependent pathways as accurately as kinetic gating at early time points, the proposal's added machinery is unnecessary.

Why not just do receptor knockouts and overlap the gene lists? This is the wet-lab gold standard and, for a single metabolite–receptor pair, it genuinely works as well — I will not pretend otherwise. Its weakness is scaling: it costs one knockout series per receptor, tells you nothing about unanticipated off-targets (the pleiotropy problem the reviews document at length Gut microbial metabolites in inflammatory bowel disease: immunological mechanisms regulating Treg/Th17 balance and therapeutic potential), and cannot rank a whole exposome. The pipeline's value proposition is that the knockout arms, run for four receptors here, become training and validation data for a predictor that generalizes to the rest of the panel; if cross-receptor generalization fails, the proposal has built an expensive assay rather than a method.

Why not just impute metabolite engagement from RNA covariation, UnitedMet-style? UnitedMet is an impressive predictor — it imputed 48% to 67% of metabolites from RNA in cross-validation UnitedMet harnesses RNA–metabolite covariation to impute metabolite levels in clinical samples — but it models intracellular metabolite pools statistically, not physical binding to receptors, contains no kinetic parameters, and its own authors note a substantial fraction of the metabolome is unpredictable from RNA. It answers a different question and is not a baseline here.

Decisive Checks

The proposal stands or falls on four numbers. First, early-time-point precision-recall against knockout ground truth: kinetic gating must beat unweighted NetGSA and over-representation analysis, and the docking-weight ablation must underperform SPR weights. Second, the onset prediction: Spearman correlation between SPR-derived $\tau$ per metabolite–receptor pair and the empirically observed lag of that receptor's pathway activation in the RNA-seq time course; a near-zero correlation kills the gating story. Third, the persistence prediction: across the panel, pathway signal duration at 24 hours should track residence time $1/k_{off}$ more tightly than $K_D$; no reported result found on Lacuna for any existing residence-time-versus-transcriptomic-persistence comparison, so this would be a first estimate. Fourth, biological face validity: the pipeline must recover the three established axes — butyrate–GPR109A, indole–AhR/IL-22, and 3-oxoLCA–RORγt Gut microbial metabolites in inflammatory bowel disease: immunological mechanisms regulating Treg/Th17 balance and therapeutic potential — ranked above the pleiotropic HDAC program at early time points, and the HDAC/chromatin program above them at late ones. If the biophysics and the transcriptomics disagree on all four checks, the honest conclusion will be that binding kinetics, at least at SPR's resolution for sub-300 Da ligands, do not propagate to the transcriptome in a predictable way — itself a result the field currently has no way to obtain.