No persistent circadian oscillator at genome resolution: pseudo-coherence in gut microbiome dynamics

Technical Summary: No Persistent Circadian Oscillator at Genome Resolution

Problem Statement
The prevailing interpretation of diurnal rhythms in the gut microbiome relies on the conceptual framework of coupled phase oscillators. This view assumes that microbial communities contain intrinsic or host-entrained oscillators that phase-lock to a common clock, generating observed rhythmicity, anti-synchronized clusters, and spectral peaks. However, this interpretation rests on the strong physical assumption that the system operates near a Hopf bifurcation or possesses intrinsic oscillatory modes.

This paper challenges that assumption by asking whether the time-frequency structure of genome-resolved (Metagenome-Assembled Genome, or MAG-level) gut microbiome dynamics could instead originate from a stable, oscillator-free stochastic regime. Specifically, it investigates whether "non-normal transient amplification"—a geometric property of the interaction matrix where non-orthogonal eigenvectors allow perturbations to grow transiently before decaying—can organize collective behavior into pseudo-coherent patterns without an underlying oscillator.

Methodology
The authors reanalyzed an hourly, two-week time series of the mouse gut microbiome from two animals (Mouse A and Mouse B) [45]. The analysis pipeline was designed to test for four specific signatures of "pseudo-coherence," a regime where geometric amplification reshapes stochastic fluctuations onto a low-dimensional reaction subspace:

1. Local Jacobian Inference: Using four consecutive hourly observations, the authors estimated a local Jacobian matrix ($\hat{A}_k$) via least-squares update. Rather than analyzing the full high-dimensional matrix, they diagonalized the commutator $B_k = [\hat{A}_k, \hat{A}_k^\top]$ to extract a rank-two subspace. This subspace defines two modes: the non-normal mode (which absorbs stochastic forcing) and the reaction mode (into which fluctuations are transiently amplified).
2. Non-Normal Diagnostics: The study computed the non-normality index ($K$) relative to a geometric threshold ($K_c$) to determine if the system is in a reactive regime ($K/K_c > 1$). It also tracked the "support" ($s_r, s_n$) of these modes, measuring how broadly the amplification is distributed across the MAGs.
3. Phase and Cluster Analysis: The authors utilized a non-parametric phase determination (NPPD) method to assign phases to MAGs and cluster them. Crucially, they developed a phase-agnostic cluster recovery method based solely on the sign of the reaction-mode components ($r_i$) derived from the commutator, independent of any phase information.
4. Spectral and Covariance Tests:
   • Time-Frequency Analysis: Morlet wavelet scalograms were generated to search for persistent ridges (fixed time-frequency structures).
   • Surrogate Testing: Amplitude-Adjusted Fourier Transform (AAFT) surrogates were generated to test if observed low-frequency power exceeded what is expected from marginal spectra alone, controlling for false discovery.
   • Time-Reversal Symmetry: Lagged covariance ($C(\tau)$) was analyzed for asymmetry ($I(\tau)$), a signature of irreversibility and circulating probability currents.
   • Entropy Production: A local entropy production proxy ($\Sigma_{local}$) was derived from the slope of the lagged imbalance at $\tau \to 0$.

Key Results
The analysis yielded four consistent findings across both animals, jointly supporting the pseudo-coherence hypothesis and contradicting the persistent oscillator hypothesis:

1. Intermittent Phase Alignment Driven by Support: Macroscopic phase coherence (cluster order parameters) was found to be intermittent. Crucially, the timing of these coherence peaks tracked the spatial support of the reaction mode ($s_r$), not proximity to spectral instability. Wavelet coherence between support and order parameters was high ($0.55–0.70$) across all frequency bands, and Granger causality suggested support drives coherence.
2. Absence of Persistent Ridges: The time-frequency representation (scalograms) showed no persistent ridges at any frequency, including the circadian band. While the time-averaged spectrum showed a weak enhancement at low frequencies (peaking between 1/24 and 1/16 $h^{-1}$), this was not a fixed feature but a drifting concentration of power. Surrogate tests confirmed this low-frequency excess was statistically significant but lacked the stationary structure of an oscillator.
3. Broken Time-Reversal Symmetry: The lagged covariance was markedly asymmetric, with a global imbalance peak ($I(\tau)$) occurring at intermediate lags (tens of hours). This indicates a non-equilibrium steady state with circulating probability currents. The slope of this imbalance at the origin provided a strictly positive, quasi-stationary estimate of entropy production, consistent across both animals.
4. Biological Recovery of Trophic Guilds: The phase-agnostic clustering based on reaction-mode signs successfully recovered two distinct functional guilds that matched the original phase-based clusters with high agreement (>93%).
   • Cluster 1: Dominated by Bacteroidota (e.g., Muribaculaceae, Bacteroidaceae), identified as primary polysaccharide degraders.
   • Cluster 2: Dominated by Bacillota A (e.g., Lachnospiraceae, Butyricicoccaceae), identified as secondary short-chain fatty acid fermenters.
   • This recovery occurred without invoking any phase information, suggesting the "night/day" split previously observed is actually a reflection of the upstream/downstream trophic architecture of cross-feeding.

Significance and Claims
The paper claims that the genome-resolved dynamics of the gut microbiome are best described as a stable, strongly non-normal stochastic regime rather than a system of coupled oscillators.

• Reinterpretation of Rhythmicity: The observed "rhythms" are not evidence of intrinsic or host-entrained oscillators. Instead, they are "pseudo-coherent" episodes where geometric amplification reshapes stochastic noise into low-dimensional, intermittent, synchrony-like patterns.
• Mechanism: The system operates via a directed cross-feeding cascade (primary degraders $\to$ secondary fermenters). The non-normal architecture of this interaction network amplifies fluctuations, creating transient collective excursions and broken time-reversal symmetry without requiring a Hopf bifurcation.
• Null Hypothesis: The authors propose that the appropriate null hypothesis for genome-resolved microbiome dynamics is a stable, non-normal stochastic system, not a phase-locked oscillator.
• Falsifiable Prediction: The paper proposes a specific test using clock-gene knockout cohorts (e.g., Bmal1, Per knockouts). The non-normal model predicts that while host-clock-locked 24-hour components may weaken, the transient amplification structure, lead-lag asymmetry, and the two-cluster guild organization should persist. Conversely, an oscillator model would predict a collapse of the rhythmic structure entirely.

The authors conclude that life itself, in this context, can be viewed through the lens of non-normal amplification of fluxes, where the gut microbiome serves as an empirical realization of a non-equilibrium steady state sustained by directional cross-feeding rather than oscillator-mediated phase locking.