Conditioning as a route to stereotyped behavior in growing populations

Technical Summary: Conditioning as a Route to Stereotyped Behavior in Growing Populations

Problem Statement
Biological systems frequently execute complex, multi-step processes (e.g., genome replication, ribosome assembly, embryonic development) in a reproducible, stereotyped sequence despite underlying stochastic dynamics. The standard explanation posits "micromanagement" by molecular machinery that recognizes and corrects specific errors (e.g., mismatch repair enzymes). However, this approach faces a high-dimensional recognition problem: the machinery must distinguish correct from incorrect states at every step across a vast space of possible deviations. This paper explores an alternative, qualitatively different route to stereotyped behavior: conditioning. Instead of recognizing specific errors, conditioning relies on a coarse criterion (e.g., a time threshold) where attempts failing the criterion are destroyed and do not leave a physical record. The surviving ensemble is narrower and more ordered. The central question is whether such a mechanism can spontaneously evolve to produce ordered trajectories without direct selection for order, and under what conditions it is beneficial.

Methodology
The authors develop a theoretical framework combining stochastic processes, population dynamics, and information theory.

1. General Framework:

   • Fitness Definition: Fitness ($f_{growth}$) is defined not by the mean replication time but by the long-term exponential growth rate of a population, governed by the Euler-Lotka equation: $2\tilde{P}(-f_{growth}) = 1$, where $\tilde{P}$ is the moment generating function of the completion time distribution $P(T)$. This weights faster replication events more heavily.
   • Conditioning via Resets: The system is modeled as undergoing stochastic resets. Any replication attempt not completed by a threshold time $T_r$ is aborted and restarted. This creates a conditional distribution $P^{cond}[x(t)] = P[x(t) | T < T_r]$.
   • Fitness Metric: A combined fitness function is introduced: $\ln F = \ln f_{growth} + s \Delta S$. Here, $\Delta S$ represents the reduction in trajectory entropy (a measure of stereotypy/reproducibility), and $s$ is a selection coefficient for entropy reduction. The authors analyze the trade-off between the growth rate cost of resetting (wasted time) and the benefit of reduced entropy.

2. Mechanistic Model: "Socks-Before-Shoes":

   • A cell must complete $n$ actions to replicate. Ideally, actions occur in a canonical order ($1, 2, \dots, n$).
   • Dynamics: At each step, the system selects the canonical next action with probability $1-\epsilon$ or picks a random remaining action with probability $\epsilon$.
   • Time Costs: In-order actions take time $\tau_{fast}$. Out-of-order actions incur an additional delay, resulting in a mean time $\tau_{slow} = \eta_{stall} \tau_{fast}$, where $\eta_{stall}$ is the stalling factor.
   • Analysis: The model calculates the ordering score $\kappa$ (fraction of action pairs in canonical order) and the growth rate as a function of the reset time $T_r$.

3. Save Points:

   • To address the scaling cost of resets in large systems, the authors introduce "save points": positions in the sequence where verified progress is preserved across resets. If a trajectory reaches a save point, a reset returns the system to that point rather than the start.

4. Phenomenological Survey:

   • The authors analyze the relationship between the shape of the completion-time distribution $P(T)$ and the efficacy of conditioning. They survey five standard distribution families (Normal, Weibull, Log-normal, Fréchet, Gumbel) to determine how tail weight influences the transition from antagonistic (ordering costs speed) to synergistic (ordering improves speed) regimes.

5. Application to Templated Replication:

   • The framework is applied to nucleic acid replication, modeling exonuclease activity (partial reset/save points) versus full degradation (full reset). They map experimental parameters (error rates $\mu_0$ and stalling factors $\eta_{stall}$) from various systems (non-enzymatic copying, ribozymes, DNA polymerases) onto the model.

Key Results

• Spontaneous Emergence of Order: The paper identifies a regime where conditioning evolves spontaneously without direct selection for order ($s=0$). This occurs when the completion-time distribution is sufficiently heavy-tailed (large variance or large $\eta_{stall}$). In this "synergistic" regime, resetting slow, disordered trajectories actually increases the population growth rate because the time saved by discarding extremely slow outliers outweighs the cost of resetting. Consequently, the fastest-growing population is automatically the most ordered.
• The Role of Tail Weight: The transition between antagonistic (ordering reduces growth) and synergistic (ordering increases growth) regimes is determined by the tail weight of $P(T)$. For heavy-tailed distributions (e.g., Fréchet, Log-normal with high $\sigma$), resets are beneficial for speed alone. For light-tailed distributions (e.g., Normal), resets always incur a speed penalty, requiring direct selection for order ($s>0$) to evolve.
• Hierarchical Ordering: In the "socks-before-shoes" model, resets impose a hierarchical temporal ordering on actions without microscopic control over the sequence. The probability of an action occurring in the correct order increases after conditioning, particularly for intermediate actions.
• Save Points Enable Scalability: For large $n$ (complex processes), unconditional resets become antagonistic because the combinatorial space of permutations makes discarding all slow trajectories too costly. Introducing save points (partial resets) restores the synergistic regime, allowing conditioning to scale to complex multi-step processes by preserving verified progress.
• Biological Relevance: The authors map experimental data to their model. They find that for many biological systems, particularly those with high stalling factors (e.g., high-fidelity DNA polymerases where mismatches cause massive delays), the parameters fall within the synergistic regime. This suggests that mechanisms like exonuclease proofreading may have evolved primarily to speed up replication, with the reduction of trajectory entropy (error correction) being a beneficial byproduct.

Significance and Claims
The paper claims to provide a "minimal route to reliable behavior" that requires only a clock (a temporal signal) rather than complex molecular machinery capable of recognizing specific microscopic errors.

• Alternative to Error Correction: It challenges the view that stereotyped behavior must arise from specific error recognition. Instead, it proposes that "exploratory dynamics" where failures are destroyed can naturally select for ordered trajectories.
• Jaynesian Conditioning: The work connects to the ideas of E.T. Jaynes, suggesting that conditioning on a final outcome (survival/completion) makes the surviving trajectories appear directed, even if the underlying dynamics are unchanged.
• Evolutionary Implications: The authors argue that conditioning mechanisms can evolve spontaneously as a byproduct of selection for growth rate, provided the system exhibits sufficient stochasticity (heavy-tailed completion times). This offers a potential explanation for the origin of error-correction mechanisms in prebiotic or early evolutionary contexts before complex recognition machinery existed.
• Universality: The framework applies broadly to any system where trajectory quality correlates with completion time (e.g., stalling due to kinetic traps), suggesting that temporal conditioning is a general physical principle for achieving reproducibility in stochastic systems.

The paper concludes that for a specific class of processes characterized by heavy-tailed completion times, "it pays for itself," meaning the speed benefit of resetting is sufficient to drive the evolution of mechanisms that also produce stereotyped, low-entropy outcomes.