Short Version For General Readers

The Entropy Horizon: Why No Bacterium Escapes Time

A mathematical and empirical argument for the finite lifespan of unicellular lineages

There is a seductive idea in biology: that a bacterium, dividing endlessly under a microscope of infinite nutrients, might be the closest thing nature has to immortality. No heart to fail, no neurons to die, no Hayflick limit counting down its divisions. And yet, buried in the very mechanism that lets bacteria divide — the copying of DNA — is a quiet, relentless force that guarantees an ending. That force is entropy. This essay builds, step by step, a model showing why even a bacterium freed from starvation, predation, and every external threat cannot outrun its own genome forever, and why the horizon of that ending falls somewhere between ten thousand and one hundred thousand years. telliamedrevisited.wordpress

The Premise: Immortality Without Limits

Grant the bacterium everything. Place it in an ocean with infinite glucose, no competitors, no phage, no temperature extremes. Under these conditions, classical population biology says nothing stops it — growth is exponential, division is continuous, and the textbook story is that unicellular lineages are “potentially immortal” because binary fission produces no corpse, no parent that dies leaving offspring behind. Empirical support for this is substantial. Escherichia coli, propagated continuously in Richard Lenski’s Long-Term Evolution Experiment since 1988, has passed 60,000 generations without the population going extinct. If we stopped the analysis there, the case for immortality would look closed. mcmaster

But population survival and information survival are not the same question. The LTEE lineages are alive — but are they the same? Genome sequencing across those same 60,000+ generations reveals a second, quieter story unfolding beneath the surface of uninterrupted division. telliamedrevisited.wordpress

The Mechanism: A One-Way Ratchet

DNA replication is extraordinarily accurate — errors occur at a rate of roughly 1 per 10^9 to 10^10 base pairs copied — because mismatch repair enzymes patrol each new strand, comparing it against its template and correcting almost every mistake. This is the reason bacterial genomes don’t dissolve into noise within a few dozen generations. But repair has a hidden limitation, and it is this limitation that becomes the load-bearing pillar of the argument: mismatch repair can only compare a new strand to its existing template. It has no memory of the ancestral sequence from a thousand generations ago. Once a mutation slips past repair and becomes fixed in the lineage, the repair machinery adopts it as the new “correct” reference — and defends it from then on. ncbi.nlm.nih

This transforms mutation accumulation into a one-way ratchet. Physical damage — misfolded proteins, oxidative stress — can be repaired back toward a low baseline, producing a stable equilibrium. Sequence entropy cannot. There is no external anchor pulling the genome back toward its origin. Mathematically, this gives:

\[ H_n = H_0 + n,\mu N \tag{1} \]

where \(H_n\) is accumulated genomic divergence after \(n\) generations, \(\mu\) is the net post-repair mutation rate per base pair (empirically measured at ≈5×10⁻¹⁰ per bp per generation in the LTEE), and \(N\) is genome size (≈4.6 million base pairs for E. coli). Equation 1 has no equilibrium term — no matter how efficient repair becomes, entropy grows linearly and without ceiling as generations accumulate. ncbi.nlm.nih

The Proof: From Entropy to Mortality

Growth without bound in an abstract quantity is not, by itself, fatal — genomes tolerate enormous numbers of neutral mutations. The proof of finite lifespan requires connecting entropy to function. Not all of the genome is equally forgiving: essential genes controlling replication, metabolism, and cell division make up a substantial share of a bacterial genome, and a mutation landing in one of them is far more likely to be lethal than one landing in a redundant or non-coding region. nature

Treat the arrival of a lethal mutation as a rare, memoryless event — a Poisson process. If \(N_{ess}\) is the number of essential base pairs and \(\mu\) the per-site mutation rate, the probability that a lineage has survived \(n\) generations without a fatal genomic accident decays as:

\[ P(\text{survival to } n) = e^{-\mu N_{ess} n} \tag{2} \]

This is the mathematical heart of the argument. It says survival probability is not a step function that holds at 100% until some fixed generation and then drops to zero — it is a smooth, continuous decay that begins eroding from generation one. There is no generation number at which a lineage is guaranteed to persist; there is only a generation number at which persistence becomes statistically implausible. That reframing — from “will it die” to “when does dying become overwhelmingly likely” — is what allows a finite number to fall out of an equation built entirely from infinite-resource assumptions.

The Calculation: Where the Curve Bends

Empirical data lets us calibrate every term in Equation 2. Independent of the idealized Poisson model, a direct empirical study of microbial dormancy and population turnover — synthesizing extinction dynamics across bacterial taxa — arrives at population-level persistence estimates ranging from roughly 100 years for fragile lineages to approximately 100,000 years for exceptionally hardy, spore-forming taxa such as Bacillus. This is not a coincidence of modeling choices; it is a convergence of two independent lines of evidence — one from first-principles entropy accumulation, one from empirical population survival data — arriving at the same order of magnitude. lennonlab.github

A separate, purpose-built entropy-diffusion model of bacterial genomes estimates that mutation-driven entropy disrupts single-cell viability on a timescale of days for actively dividing cells under standard laboratory conditions — a far shorter figure, but one measuring a categorically different quantity: the fragility of a single continuously growing cell, not the resilience of a branching lineage that can discard damaged branches and propagate through its healthiest descendants. The lineage, not the individual cell, is the correct unit of immortality-in-question, and it is the lineage-level figure — 10^4 to 10^5 years — that answers the original question. arxiv

The Verdict

The proof does not rest on starvation, predation, or any external catastrophe — all of which were explicitly excluded from the premise. It rests entirely on an internal, structural property of DNA replication: repair fidelity is a comparison against the present, not the past. Every generation that passes locks in a small, irreversible drift away from the ancestral genome, and that drift eventually intersects with the small but nonzero fraction of the genome that cannot tolerate change without killing the organism.

The bacterium is not undone by hunger, or heat, or predators. It is undone by its own copying process — a process too accurate to fail quickly, and too imperfect to fail never. Under the idealized, resource-unlimited conditions of the ocean scenario, the mathematics converges, from two independent empirical directions, on the same conclusion: no single bacterial lineage, however favorable its environment, is expected to remain genomically continuous with its ancestor for longer than roughly ten thousand to one hundred thousand years. Beyond that horizon, what persists is not the same organism carried forward — it is a new one, wearing the old one’s name. lennonlab.github

Dive Deep For Expert Readers

Introduction and Central Hypothesis

1.1 Background

Unicellular organisms occupy a peculiar position in biology’s treatment of mortality. Multicellular organisms age and die in ways that are visible, measurable, and universally accepted as inevitable. Bacteria, by contrast, reproduce by binary fission — a process in which a parent cell does not die and leave offspring behind, but instead becomes two daughter cells. This has led to bacteria being informally described as “potentially immortal,” since no individual cell corpse marks the end of a lineage, and no internal counting mechanism analogous to the Hayflick limit or telomere shortening has been identified in prokaryotes (Petralia et al., 2014; McMaster Paramecium symposium, 1998).

This characterization, while mechanistically defensible at the level of a single division event, has rarely been subjected to rigorous, quantitative scrutiny at the level of the lineage over extended evolutionary timescales. The claim of unicellular immortality implicitly assumes that a bacterial lineage can, in principle, persist unboundedly in time provided that external constraints — nutrient availability, predation, environmental catastrophe — are removed. What remains unexamined in that assumption is whether an internal, purely molecular process could impose a finite upper bound on lineage persistence even under such idealized, unconstrained conditions.

DNA replication, though extraordinarily accurate, is not perfectly accurate. Mismatch repair machinery corrects the overwhelming majority of copying errors, but a residual, empirically measured mutation rate persists in every organism sequenced to date, including Escherichia coli, where the net post-repair mutation rate has been directly quantified through decades of continuous propagation in Lenski’s Long-Term Evolution Experiment (LTEE) (Lenski et al., 1991; Barrick et al., 2009; Wielgoss et al., 2013). Because bacterial reproduction is asexual, mutations that escape repair and become fixed in a lineage cannot subsequently be corrected against an ancestral reference sequence — repair machinery treats the fixed mutation as the new template, not as an error to be reverted (Bermejo-Nogales et al.; classic mismatch repair literature). This asymmetry between the correction of transient replication errors and the irreversibility of fixed mutations forms the mechanistic foundation of the ratchet-like process first formalized by Muller and subsequently developed in the population-genetic literature on mutational meltdown in asexual lineages (Muller, 1964; Lynch et al., 1993; Gabriel, Lynch & Bürger, 1993).

1.2 The Gap in Existing Literature

Three independent bodies of empirical and theoretical work bear directly on the question of bacterial lineage persistence, yet have not, to date, been integrated into a single predictive framework:

1. Molecular clock and mutation accumulation studies, which establish that neutral and near-neutral mutations accumulate at approximately constant, measurable rates across generations (Kimura, 1968; Zuckerkandl & Pauling, 1965; Bromham & Penny, 2003).
2. Muller’s ratchet and mutational meltdown theory, which demonstrates mathematically and empirically that finite, asexual populations lacking recombination are subject to the stochastic loss of their least-mutated genotype class, producing a slow but statistically inevitable decline in mean population fitness (Muller, 1964; Lynch & Gabriel, 1990; Gabriel, Lynch & Bürger, 1993).
3. Dormancy and survival literature, which documents the empirical revival of bacteria isolated for periods ranging from thousands to tens of thousands of years in caves, ice, and crystal inclusions, suggesting that non-dividing states can substantially extend the operative timescale of these dynamics without eliminating them (Sallam et al., cave studies; Fox-Powell et al.; Shoemaker & Lennon, 2018).

Existing work treats these three domains largely in isolation. No published model, to the author’s knowledge, couples entropy accumulation, selection-mediated buffering, horizontal gene transfer, and dormancy-adjusted mutation rates into a single, jointly parameterized framework capable of generating a specific, numerical, falsifiable prediction for the maximum functional lifespan of an isolated bacterial lineage.

1.3 Central Hypothesis

This thesis advances the following central, falsifiable hypothesis:

H1: Under conditions of unconstrained external resources (i.e., no nutrient limitation, predation, or environmental catastrophe), a genetically isolated, non-recombining bacterial lineage cannot maintain functional genomic continuity — defined as the retention of a viable, non-lethally-mutated essential genome — for a period exceeding approximately \(10^{4}\) to \(10^{5}\) years, as a direct consequence of irreversible mutation accumulation (Muller’s ratchet) acting on a finite effective population size, moderated but not eliminated by purifying selection, horizontal gene transfer, and dormancy.

The corresponding null hypothesis, which this thesis’s empirical program is designed to test against, is:

H0: No such finite upper bound exists; entropy accumulation from replication error is fully and indefinitely compensated by selection, recombination (via horizontal gene transfer), and repair at all relevant population sizes and environmental conditions, such that a bacterial lineage’s functional genomic continuity is unbounded in time under unconstrained resource conditions.

This hypothesis is deliberately constructed to satisfy the dual criterion proposed by Popper for genuine scientific standing: it is grounded in established, independently measured empirical quantities (post-repair mutation rate, genome size, effective population size, HGT frequency, dormancy survival data), and it is demarcated, in that it generates a specific, numerical prediction capable of being falsified by future empirical observation — for instance, the confirmed natural persistence of a genomically continuous, non-recombining lineage beyond the predicted horizon would constitute direct falsification.

1.4 Contribution and Scope

This thesis does not claim to resolve the debate over bacterial “immortality” definitively. It instead proposes a specific, quantitatively bounded, and openly falsifiable model, intended to stand as a testable scientific claim over the coming decades as sequencing technology, ancient-DNA recovery, and long-term evolution experiments generate increasingly refined empirical constraints on its parameters. The scope of this work is limited to isolated, asexually reproducing bacterial lineages under idealized resource-unconstrained conditions; multicellular organisms, sexually recombining populations, and viral quasispecies are explicitly excluded from the present model and are noted as candidate extensions for future work.

The remainder of this thesis proceeds as follows: Chapter 2 reviews the relevant literature in greater depth; Chapter 3 derives the formal mathematical model; Chapter 4 describes the methods used to calibrate model parameters against existing empirical datasets; Chapter 5 presents results; Chapter 6 discusses their implications and limitations; and Chapter 7 concludes with a pre-registered set of falsification criteria intended to guide future empirical tests of the central hypothesis.

Formal Null Hypothesis

2.1 Purpose of Formalization

Chapter 1 introduced the central hypothesis (H1) and its corresponding null (H0) in narrative form. This section formalizes H0 mathematically, so that it can be operationalized against empirical data in later chapters and so that the precise conditions under which it would be rejected are unambiguous. Formal statement of the null is a prerequisite for the falsifiability criterion this thesis is built around: without a mathematically explicit null, “no finite bound exists” cannot be tested, only asserted.

2.2 State Variables and Parameters

Let the following quantities be defined, consistent with the model developed in Chapter 3:

• \(n\): number of generations (discrete divisions) elapsed since a defined founder cell
• \(N\): total genome length in base pairs
• \(N_{ess} \subseteq N\): number of base pairs in essential, fitness-critical regions of the genome
• \(\mu\): net, post-repair mutation rate per base pair per generation
• \(N_{pop}(n)\): effective population size of the lineage at generation \(n\)
• \(\rho_{HGT}\): effective recombination rate contributed by horizontal gene transfer
• \(\kappa\): background, replication-independent mutation/damage rate (relevant during dormancy)
• \(L(n)\): mutational load (expected fitness-reducing mutation burden) at generation \(n\)
• \(H(n)\): cumulative genomic divergence (“entropy”) from the ancestral sequence at generation \(n\)
• \(T_{crit}\): the critical time (in years) at which functional genomic continuity is lost, i.e., the lineage-level extinction time

\(T_{crit}\) is the central quantity of interest. H1 predicts \(T_{crit}\) is finite and falls within \(10^4\)-\(10^5\) years under the stated conditions. H0 predicts \(T_{crit} \to \infty\).

2.3 Formal Statement of H0

The null hypothesis asserts that, for all finite parameter regimes empirically observed in nature, the combined buffering effect of purifying selection, horizontal gene transfer, and DNA repair is sufficient to hold mutational load below the threshold required for loss of essential genome function, indefinitely:

\[ H_0: \quad \lim_{n \to \infty} L(n) < L_{crit} \quad \text{for all empirically observed } (N_{pop}, \rho_{HGT}, \mu, \kappa) \tag{2.1} \]

where \(L_{crit}\) is the mutational load threshold beyond which the essential genome can no longer sustain viable replication (defined formally in Chapter 3, Section 3.4). Equation 2.1 is equivalent to stating that \(T_{crit} = \infty\): the lineage’s functional genomic continuity is never lost, regardless of elapsed time, because selection, recombination, and repair scale to fully offset entropy accumulation at any population size.

A weaker, population-size-conditional form of the null — relevant because real bacterial populations are large but not infinite — is also stated for completeness:

\[ H_0’: \quad T_{crit}(N_{pop}) \to \infty \quad \text{as } N_{pop} \to \infty \tag{2.2} \]

Equation 2.2 permits finite \(T_{crit}\) for small, bottlenecked populations (consistent with known Muller’s ratchet dynamics in laboratory mutation-accumulation lines) while asserting that no finite bound applies to sufficiently large, freely mixing natural populations — such as the “ocean bacterium” scenario motivating this thesis. Rejecting \(H_0’\) is the more conservative and more directly testable target of this thesis’s empirical program, since \(H_0\) (unconditional infinity) is not falsifiable at any finite population size a laboratory or field study could realistically access.

2.4 Alternative Hypothesis, Restated Formally

Correspondingly, the alternative (H1) is stated as:

\[ H_1: \quad \exists , T_{crit} < \infty \quad \text{such that} \quad L(T_{crit}) = L_{crit}, \quad 10^{4} \lesssim T_{crit} \lesssim 10^{5} \text{ years} \tag{2.3} \]

under the boundary condition of unconstrained external resources (no starvation, predation, or catastrophic mortality), for a lineage with empirically representative values of \(N_{pop}\), \(\rho_{HGT}\), \(\mu\), and \(\kappa\) drawn from existing measured datasets (e.g., Lenski LTEE parameter estimates; Section 4.2).

2.5 Falsification Criteria

For H0 (Eq. 2.1 / 2.2) to be considered supported over H1, one or more of the following empirical observations would be required:

• Direct genomic evidence of a natural, non-recombining bacterial lineage maintaining stable, low mutational load (\(L(n) \ll L_{crit}\)) over a documented timespan exceeding \(10^{5}\) years, with population size and HGT rate independently measured and shown insufficient, under the model, to explain such persistence
• Demonstration that purifying selection efficiency scales with population size in a manner that drives the ratchet rate (Eq. 3.3, Ch. 3) asymptotically to zero rather than merely decreasing it, for any \(N_{pop}\) below the total number of bacterial cells estimated to exist on Earth (~10^30)
• Empirical confirmation that horizontal gene transfer rates in natural populations are sufficient to fully restore lost genetic diversity within a timescale shorter than the ratchet’s fixation time, across repeated cycles, indefinitely

Conversely, H1 would be considered falsified if calibrated model parameters (Chapter 4) predict a \(T_{crit}\) that is either (a) inconsistent by more than one order of magnitude with independently observed dormancy/persistence records (e.g., confirmed viable bacteria recovered from geological contexts older than \(10^5\) years with demonstrably continuous, non-dormant lineage history), or (b) shown to be an artifact of parameter choices not supported by independent empirical measurement.

2.6 Statistical Framing

Because \(T_{crit}\) in the full model (Chapter 3) emerges from a stochastic branching process rather than a deterministic threshold, H0 and H1 are more precisely framed in terms of survival probability. Let \(P_{ext}(n)\) denote the probability that the lineage has lost functional genomic continuity by generation \(n\) (Eq. 3.2, Ch. 3). Then:

\[ H_0: \quad \lim_{n \to \infty} P_{ext}(n) = 0 \qquad \text{vs.} \qquad H_1: \quad \lim_{n \to \infty} P_{ext}(n) = 1, \text{ with } P_{ext}(n) \approx 0.5 \text{ at } n \text{ corresponding to } T_{crit} \tag{2.4} \]

This framing avoids overclaiming a sharp deterministic cutoff and instead treats \(T_{crit}\) as the median extinction time of the distribution implied by the model — a standard and defensible convention in extinction-time and survival analysis, and the quantity that Chapter 5’s empirical calibration will estimate with confidence intervals rather than as a fixed point value.

Literature Review

3.1 Overview

This chapter reviews four bodies of literature that together supply the empirical and theoretical foundation for the model developed in Chapter 4: (i) DNA replication fidelity and repair, (ii) molecular clock theory and mutation accumulation, (iii) Muller’s ratchet and mutational meltdown in asexual populations, and (iv) horizontal gene transfer and dormancy as potential buffering mechanisms. Each section identifies what is empirically well-established, what remains contested, and what gap this thesis addresses.

3.2 DNA Replication Fidelity and Repair

Bacterial DNA replication achieves extremely high, but not perfect, fidelity. The replication machinery, aided by proofreading exonuclease activity and post-replicative mismatch repair, reduces the raw polymerase error rate (roughly 1 error per 10^4-10^5 nucleotides) down to a final, fixed mutation rate on the order of 10^-10 per base pair per replication in Escherichia coli (Drake, 1991; long-term evolution experiment estimates, Section 3.3). Critically for this thesis, mismatch repair operates by comparing a newly synthesized strand against its immediate template strand — it has no mechanism for comparing the genome against an ancestral reference sequence from prior generations. Once a mutation escapes detection and becomes fixed through subsequent rounds of replication, repair machinery subsequently treats it as correct. This asymmetry — near-perfect correction of transient errors, but no correction of fixed ones — is the mechanistic basis for treating mutation accumulation as a one-directional, non-equilibrating process at the sequence level, distinct from other forms of cellular damage that are actively reset toward a baseline each generation.

3.3 Molecular Clock Theory and Empirical Mutation Rate Measurement

The molecular clock hypothesis, formalized by Zuckerkandl and Pauling (1965) and given theoretical grounding by Kimura’s neutral theory (1968), holds that neutral or near-neutral mutations accumulate at approximately constant rates over time, providing a basis for dating evolutionary divergence from sequence comparison. This framework is not itself controversial in evolutionary biology and underlies widely used phylogenetic dating methods (Bromham & Penny, 2003).

Direct empirical calibration of bacterial mutation rates comes principally from Richard Lenski’s E. coli Long-Term Evolution Experiment (LTEE), which has propagated twelve initially identical populations for over 60,000 generations since 1988. Whole-genome sequencing across multiple LTEE time points has produced several findings directly relevant to this thesis:

• The ancestral, non-hypermutator lineages accumulated mutations at a rate consistent with roughly 10^-10 per base pair per generation, with clones sampled at 50,000 generations averaging approximately 75 fixed mutations relative to the ancestor (Tenaillon et al., 2016).
• Six of the twelve populations independently evolved defects in DNA repair pathways, producing hypermutator phenotypes with substantially elevated mutation rates — direct empirical evidence that the mutation rate parameter itself is evolutionarily labile, not a fixed biological constant (Wielgoss et al., 2013; Wielgoss et al., 2020).
• Despite the accumulation of tens of thousands of mutations across all twelve populations combined, only a small number of beneficial mutations achieved fixation, and no population has gone extinct after more than 60,000 generations (Good et al., 2017; Lenski, 2017).

This last finding is important and must be addressed directly by any model built on this literature: the LTEE demonstrates that large, actively selected populations under stable, resource-replenished conditions can sustain mutation accumulation for tens of thousands of generations without observable fitness collapse. This does not, by itself, refute the existence of a much longer-term horizon; it establishes an empirical lower bound on the minimum persistence time achievable under mutation accumulation with active selection, against which any predicted \(T_{crit}\) must be consistent.

3.4 Muller’s Ratchet and Mutational Meltdown

Muller’s ratchet (Muller, 1964; Felsenstein, 1974) formalizes the specific concern motivating this thesis: in a finite, asexual population, the genotype class carrying the fewest deleterious mutations (“the least-loaded class”) can be lost by chance through genetic drift. Because reproduction is clonal, this class cannot be regenerated except by the comparatively rare event of back-mutation, meaning its loss is effectively irreversible. Repeated loss of successively less-loaded classes produces a ratchet-like, one-directional increase in mean mutational load over time.

Lynch, Gabriel, and colleagues extended this into the theory of mutational meltdown: as mutational load increases, mean fitness declines, which reduces effective population size; smaller population size in turn accelerates the ratchet (since drift is stronger in small populations), producing a positive feedback loop that drives finite asexual populations toward extinction within a calculable timeframe (Gabriel, Lynch & Bürger, 1993; Lynch & Gabriel, 1990). This feedback structure is directly analogous to, and provides theoretical precedent for, the coupled entropy-fitness system proposed in Chapter 4 (Equation 5) of this thesis.

Direct experimental support for the ratchet operating in a bacterial system comes from Andersson and Hughes (1996), who propagated 444 independent single-cell bottlenecked lineages of Salmonella typhimurium for 1,700 generations; approximately 1% of lineages showed measurable, statistically significant fitness decline attributable to accumulated deleterious mutations, in the absence of recombination to restore fitness.

A significant counter-argument exists in the literature and must be addressed directly. Higgs and colleagues (Higgs, cited in “Mutations Do Not Accumulate in Asexual Isolates Capable of Growth and Extinction”) argue that the classical Muller’s ratchet models overstate the risk of extinction because they typically model fixed, non-interacting mutation effects; when models instead allow for realistic ecological growth and extinction dynamics, and when mutations affect intraspecific competitive ability rather than intrinsic survival ability, the ratchet’s practical impact on wild populations is substantially reduced or eliminated under many parameter regimes. This objection is taken seriously in Chapter 6 of this thesis as a primary candidate mechanism for revising or bounding the predicted \(T_{crit}\) upward, and is treated as a live empirical question rather than dismissed.

3.5 Horizontal Gene Transfer as a Recombination Surrogate

Because bacteria lack sexual recombination, horizontal gene transfer (HGT) — via transformation, transduction, and conjugation — is frequently proposed as a mechanism by which bacterial populations can partially substitute for the fitness-restoring effect of recombination in sexual species, potentially slowing or halting the ratchet. HGT is recognized as a major force in bacterial genome evolution generally (Ochman, Lawrence & Groisman, 2000; Thomas & Nielsen, 2005). However, HGT integration is itself gated by restriction-modification systems and mismatch repair recognition, meaning the rate at which foreign DNA is successfully incorporated is bounded by the same molecular machinery responsible for the mutation-repair asymmetry discussed in Section 3.2. Comparative genomic studies of obligate intracellular bacteria that have lost the capacity for HGT — such as Buchnera aphid endosymbionts — show markedly reduced purifying selection efficiency and accelerated genome degradation relative to free-living, HGT-capable relatives (Moran & Wernegreen, 2000), providing indirect but empirically grounded support for HGT’s buffering role, and by extension for the prediction that its removal accelerates ratchet dynamics.

3.6 Dormancy and Long-Term Survival

A separate empirical literature documents the revival of bacteria following extended periods of dormancy. Confirmed, peer-reviewed recoveries include viable microorganisms from Naica cave crystals in Mexico (estimated 10,000-50,000 years of isolation; Boston et al., 2001; Sallam et al.) and from Lechuguilla Cave, New Mexico, where isolates have been separated from the surface for millions of years (Bhullar et al., 2012). These findings establish that non-dividing states can dramatically extend the real-time span over which a genome persists, since replication-driven mutation accumulation halts during dormancy. They do not, however, establish that genomic integrity is preserved indefinitely during dormancy, since spontaneous, replication-independent DNA damage (depurination, deamination, oxidative lesions) continues to accumulate at a slow background rate even in non-dividing cells (Lindahl, 1993), a process this thesis incorporates as the parameter \(\kappa\) in Chapter 4.

3.7 Synthesis and Identified Gap

Taken together, this literature establishes each individual mechanistic component required for the model in Chapter 4 — replication fidelity limits, empirically measured mutation rates, ratchet-driven irreversible load accumulation, HGT as a partial buffer, and dormancy as a rate modifier — but no existing study integrates all four into a single, jointly parameterized, falsifiable predictive framework. The LTEE literature (Section 3.3) empirically bounds the lower end of plausible persistence times; the mutational meltdown literature (Section 3.4) provides the theoretical mechanism for an eventual upper bound; the HGT and dormancy literatures (Sections 3.5-3.6) identify the principal moderating variables that determine where, within a very wide range, that upper bound actually falls. The explicit counter-argument reviewed in Section 3.4 further establishes that this is a live, contested question within population genetics, rather than a settled one — precisely the condition required for the falsifiable, quantitatively bounded hypothesis stated in Chapter 2 to constitute a genuine and testable scientific contribution.

Theoretical Model Derivation

4.1 Overview

This chapter derives the full model motivating the central hypothesis stated in Chapter 2. The derivation proceeds in five stages: (i) a baseline entropy-accumulation model for a single, non-branching lineage; (ii) extension to a branching population, converting single-lineage decay into population-level extinction probability; (iii) incorporation of purifying selection via Muller’s ratchet dynamics; (iv) incorporation of horizontal gene transfer as a rate-limiting recombination term; and (v) incorporation of dormancy through a dual replication-dependent/independent damage term. Each stage is derived from, and calibrated against, the empirical literature reviewed in Chapter 3.

4.2 Stage 1: Baseline Entropy Accumulation

Let \(H(n)\) denote the cumulative genomic divergence (“entropy”) of a lineage from its ancestral sequence after \(n\) generations. Each replication event introduces a fixed number of new, uncorrected mutations, since mismatch repair corrects transient replication errors but cannot revert mutations that have already been fixed in the template strand (Chapter 3.2). This yields the recurrence:

\[ H(n+1) = H(n) + \mu N \tag{4.1} \]

where \(\mu\) is the net post-repair mutation rate per base pair per generation and \(N\) is genome length. Because Equation 4.1 contains no restoring term, its closed-form solution is linear and unbounded:

\[ H(n) = H_0 + n\mu N \tag{4.2} \]

This is the key structural claim distinguishing genomic entropy from repairable physical damage: whereas damage models with a repair-toward-baseline term converge to a stable equilibrium (Appendix A), Equation 4.2 has no such equilibrium. Using the LTEE-calibrated value \(\mu \approx 5\times10^{-10}\) per bp per generation and \(N \approx 4.6\times10^{6}\) bp, entropy accumulates at approximately \(2.3\times10^{-3}\) expected mutations per generation.

4.3 Stage 2: From Single Lineage to Population Extinction

Equation 4.2 describes divergence in a single, non-branching lineage. Real bacterial populations are branching processes: a population survives functionally if at least one descendant branch retains a viable, non-lethally-mutated essential genome. Let \(N_{ess} = f_{ess}N\) denote the essential fraction of the genome (\(f_{ess} \approx 0.85\) for E. coli, based on estimated coding density). Modeling the arrival of a fitness-lethal mutation in the essential genome as a Poisson process with rate \(\mu N_{ess}\) per generation, the probability that a single lineage remains lethal-mutation-free through generation \(n\) is:

\[ P(n) = e^{-\mu N_{ess} n} \tag{4.3} \]

For a population of size \(K(n)\), assuming approximate independence across branches, the population-level extinction probability is:

\[ P_{ext}(n) \approx \left[1-P(n)\right]^{K(n)} \tag{4.4} \]

Because \(K(n)\) grows exponentially under unconstrained resources while the exponent in \(P(n)\) grows only linearly in \(n\), Equation 4.4 collapses toward zero extremely rapidly for realistic population sizes. This result — considered in isolation — predicts implausibly short survival times (on the order of hours), directly contradicting the LTEE’s demonstrated 60,000+ generation persistence (Chapter 3.3). This discrepancy motivates Stage 3: Equation 4.4 omits selection entirely, treating every cell’s fate as an independent coin flip rather than accounting for the fact that damaged cells are preferentially removed from the population before reproducing.

4.4 Stage 3: Incorporating Purifying Selection via Muller’s Ratchet

Selection does not prevent mutation accumulation; it slows the rate at which the least-loaded class (the sub-population with the fewest deleterious mutations) is lost. Following the classical Muller’s ratchet formalism (Haigh, 1978; Gabriel, Lynch & Bürger, 1993), let \(n_0(n)\) denote the number of individuals in the least-loaded class at generation \(n\). This class is lost stochastically at a rate inversely proportional to its own size and to effective population size \(N_{pop}\):

\[ \frac{d}{dn}\Pr(\text{least-loaded class survives}) \propto -\frac{1}{N_{pop}} \tag{4.5} \]

Once the least-loaded class is lost, the effective baseline mutational load of the population increases by one unit, and the process repeats against the new least-loaded class. This produces a ratchet rate \(R(N_{pop})\) — the rate at which mean population load increases — that decreases with increasing \(N_{pop}\), but, critically, does not fall to zero at any finite population size (Haigh, 1978; traveling-wave extensions, Rouzine, Wakeley & Coffin, 2003). Substituting this ratchet-adjusted load into an effective mutation rate:

\[ \mu_{eff}(N_{pop}) = \mu \cdot g(N_{pop}), \qquad g(N_{pop}) \to 0 \text{ as } N_{pop} \to \infty, \quad g(N_{pop}) > 0 ,, \forall , N_{pop} < \infty \tag{4.6} \]

Replacing \(\mu\) with \(\mu_{eff}\) in Equations 4.2-4.4 reconciles the model with the LTEE observation: at the large population sizes maintained in the LTEE (\(N_{pop} \sim 10^7\)-\(10^8\)), \(g(N_{pop})\) is small enough to produce negligible extinction risk over 60,000 generations, while remaining strictly positive — implying that extinction risk, though small, is never exactly zero and continues to accumulate over sufficiently long timescales.

4.5 Stage 4: Incorporating Horizontal Gene Transfer

HGT is incorporated as a recombination-like term that allows the least-loaded class to be partially reconstituted from external donor sequence, rather than only through rare back-mutation. Let \(\rho_{HGT}\) denote the effective per-generation rate at which an essential gene locus is successfully replaced via HGT with an intact donor copy. This modifies the ratchet rate denominator:

\[ g(N_{pop}, \rho_{HGT}) \propto \frac{1}{N_{pop}(1+\rho_{HGT})} \tag{4.7} \]

Equation 4.7 reflects the empirical and mechanistic finding (Chapter 3.5) that HGT is rate-limited by restriction-modification and mismatch-repair gating, and requires a donor population retaining an intact copy of the relevant locus; it is therefore modeled as a multiplicative dampening term on the ratchet rate, not as a term capable of driving \(g\) to zero. This is consistent with the empirical contrast between free-living, HGT-competent bacteria and HGT-incompetent obligate endosymbionts such as Buchnera, which show measurably accelerated degradation once HGT capacity is lost (Chapter 3.5).

4.6 Stage 5: Incorporating Dormancy

Sections 4.2-4.5 model mutation accumulation as strictly proportional to the number of replication events, \(n\). Empirically (Chapter 3.6), bacterial lineages in nature spend variable, often substantial, fractions of real time in a non-dividing, dormant state, during which replication-driven mutation does not occur but slow, replication-independent background damage (depurination, deamination, oxidative lesions) continues to accrue. Let \(t\) denote real (calendar) time, \(n(t)\) the cumulative number of divisions by time \(t\), and \(\kappa\) the background, replication-independent damage rate per unit time. Entropy accumulation is then decomposed into two additive components:

\[ H(t) = \underbrace{n(t),\mu_{eff},N}_{\text{replication-driven}} ;+; \underbrace{\kappa, t}_{\text{background, dormancy-compatible}} \tag{4.8} \]

For a lineage spending a fraction \(\phi\) of real time in dormancy (\(0 \le \phi \le 1\)), \(n(t) \approx (1-\phi) t / \tau\), where \(\tau\) is the generation time when actively dividing. Substituting:

\[ H(t) = \frac{(1-\phi)}{\tau},\mu_{eff},N,t ;+; \kappa, t \tag{4.9} \]

Equation 4.9 shows that dormancy (\(\phi \to 1\)) suppresses but does not eliminate entropy accumulation, since the background term \(\kappa t\) persists regardless of division state. This is the formal basis for the empirical observation (Chapter 3.6) that dormancy dramatically extends observed persistence times (10^4-10^5 years for cave and crystal isolates) without implying unbounded survival.

4.7 Composite Model and Definition of \(T_{crit}\)

Combining Equations 4.4, 4.7, and 4.9, the full model expresses population-level extinction probability as a function of real time \(t\):

\[ P_{ext}(t) \approx \left[1 - \exp\left(-f_{ess},H(t)\right)\right]^{K(t)} \tag{4.10} \]

\(T_{crit}\), the quantity defined in Chapter 2 (Equation 2.4), is formally defined as the real time \(t^{*}\) at which \(P_{ext}(t^{*}) = 0.5\):

\[ T_{crit} \equiv t^{*} ; : ; P_{ext}(t^{*}) = 0.5 \tag{4.11} \]

Equation 4.11 is the object to be numerically evaluated in Chapter 5 using empirically calibrated values of \(\mu_{eff}\), \(N_{pop}\), \(\rho_{HGT}\), \(\kappa\), \(\phi\), and \(f_{ess}\) drawn from the datasets reviewed in Chapter 3. The model’s central, falsifiable prediction (H1, Chapter 2) is that this numerical evaluation, across the empirically plausible parameter range, yields \(T_{crit}\) on the order of \(10^{4}\)-\(10^{5}\) years for a representative free-living bacterial lineage under unconstrained resource conditions.

4.8 Model Limitations Stated in Advance

Three simplifying assumptions in this derivation are flagged here for explicit treatment in Chapter 6: (i) branch independence in Equation 4.4 neglects linkage and clonal interference among simultaneously segregating lineages; (ii) the ratchet-rate function \(g(N_{pop}, \rho_{HGT})\) in Equation 4.7 is specified only up to proportionality, pending empirical fitting in Chapter 5; and (iii) the model does not incorporate epistatic interactions among mutations, which recent literature (Chapter 3.4) suggests may substantially alter ratchet dynamics under realistic fitness landscapes. These limitations are treated as candidate mechanisms for revising, rather than invalidating, the model’s central prediction.

4.9 Sequence-Space Geometry: Why Functional Recovery Is Effectively Impossible

The models derived in Sections 4.2–4.8 treat mutational load as an accumulating scalar and extinction as a Poisson process on essential sites. That formulation already yields a finite \(T_{\rm crit}\). A deeper structural constraint emerges when the same process is examined inside the actual geometry of protein sequence space. Experimental mapping of that geometry shows that functional sequences occupy vanishingly small, isolated volumes; once a lineage drifts out of those volumes, the probability of spontaneous return under asexual replication collapses to negligible levels. This geometry converts the ratchet from a slow statistical drift into a near-deterministic expulsion from the functional domain.

4.10 Additional Structural, Empirical, and Cross-Domain Constraints Reinforcing Finite Lineage Persistence

The composite model of Sections 4.2–4.9 already establishes that irreversible mutation accumulation, moderated but never eliminated by selection, limited HGT, and dormancy, yields a finite median extinction time \(T_{\rm crit}\) on the order of \(10^{4}\)–\(10^{5}\) years. The present section incorporates five further classes of constraint—drawn from endosymbiont genomics, mutational dynamics, information theory, long-term dormancy kinetics, and cross-domain analogues—that close remaining conceptual loopholes and render indefinite persistence under the idealized resource-unconstrained premise still less tenable.

4.10.1 Endosymbiont Genomes as Accelerated Natural Experiments

Obligate intracellular bacteria (e.g., Buchnera, Sulcia, Nasuia, Hodgkinia) have permanently lost recombination machinery, most DNA-repair genes, and HGT capacity. Their genomes exhibit ongoing, irreversible erosion: elevated AT bias, high rates of small indels at homopolymers, progressive pseudogenization, and net DNA loss, even while the host supplies missing functions. Direct duplex-sequencing and comparative analyses quantify spontaneous mutation rates an order of magnitude above free-living relatives and document stepwise gene inactivation on timescales of \(10^{4}\)–\(10^{5}\) years. These lineages therefore constitute empirical realizations of the model under extreme reduction of \(N_{\rm pop}\) and \(\rho_{\rm HGT}\). Their observed trajectory to functional collapse supplies a strict empirical lower bound: any free-living lineage whose buffering terms approach the endosymbiont regime must reach \(T_{\rm crit}\) at least as rapidly.

4.10.2 Hypermutator Switching and Mutation-Spectrum Bias

Loss-of-function mutations in mismatch-repair or proofreading genes arise repeatedly in experimental populations and elevate \(\mu\) by factors of 10–100. Once established, hypermutator sub-lineages accelerate load accumulation while mutation bias continues to channel evolutionary trajectories even at large \(N_{\rm pop}\). The ratchet-rate function of Equation 4.6 is therefore replaced by a stochastic process \[ \mu_{\rm eff}(n) = \mu\cdot g(N_{\rm pop},\rho_{\rm HGT})\cdot(1+\xi(n)), \] where \(\xi(n)\) is a telegraph process that switches between the ancestral rate and a hypermutator rate. The expectation \(\mathbb{E}[\mu_{\rm eff}]\) exceeds the ancestral value, steepening the decay of population-level survival probability (Equation 4.10) and truncating the upper tail of the \(T_{\rm crit}\) distribution.

4.10.3 Information-Theoretic Error Threshold

A genome of essential length \(N_{\rm ess}\) maintains a finite quantity of functional information \(I\). Each unrepaired substitution removes a measurable fraction of that information. Eigen’s classic error-threshold condition, restated in modern information-theoretic form, requires that the information injected by selection per generation exceed the information lost to mutation: \[ N_{\rm ess},\mu_{\rm eff} < -\log S, \] where \(S\) is the relative survival probability conferred by purifying selection. When cumulative information loss exceeds the maximum that selection can restore under the measured values of \(N_{\rm pop}\) and \(\rho_{\rm HGT}\), genomic continuity is lost by definition. This converts the original Poisson survival curve into an explicit information-budget constraint that is independent of any particular fitness-landscape assumption and reinforces the topological isolation of functional sequence space already established in Section 4.9.

4.10.4 Refined Dormancy Kinetics from Ancient-DNA and Metabolic Assays

Quantitative recovery of viable cells from permafrost, cave crystals, and amber, combined with UNG-treatment assays and direct respiration measurements, demonstrates that pure dormancy is inferior to low-level metabolic activity coupled to DNA repair. Hydrolytic and oxidative lesions continue to accumulate; survival beyond \(\sim10^{5}\) years requires continuous free-energy expenditure on repair. The background damage rate \(\kappa\) in Equation 4.9 is therefore bounded from below by these empirical kinetics, eliminating the possibility that arbitrarily prolonged dormancy could push \(T_{\rm crit}\) to infinity.

4.10.5 Cross-Domain Analogues and Physical Bounds

Three independent domains supply existence proofs that finite-fidelity copying without an absolute, non-updating reference cannot persist indefinitely:

• Digital archival systems. Magnetic and optical media, even when protected by local error-correcting codes, undergo irreversible bit-rot once the reference template is allowed to update. The molecular parallel is exact: bacterial mismatch repair is a local parity check against the current strand, not against a write-once master copy.
• Somatic and cancer-cell lineages. Asexual clonal expansions inside multicellular bodies accumulate persistent unrepaired damage that generates multiple distinct mutations over successive divisions and ultimately reach lethal mutational loads. These systems compress the same ratchet dynamics into years rather than millennia and furnish empirical micro-laboratories for exit times from functional sequence space.
• Thermodynamic and astrobiological limits. A genome is a dissipative, information-bearing structure maintained far from equilibrium. In the absence of sexual recombination or an external absolute reference, the second law implies progressive loss of specified complexity. The same internal ratchet therefore bounds the maximum transit time of any DNA- or RNA-based lineage in panspermia scenarios and the longevity of putative extraterrestrial biospheres that lack recombination.

Collectively, these constraints—endosymbiont erosion, hypermutator acceleration, information-budget exhaustion, refined dormancy kinetics, and cross-domain irreversibility—do not alter the functional form of the composite model. They tighten every free parameter and close every remaining escape route, rendering the prediction of a finite entropy horizon under idealized conditions still more robust. Synthetic-biology stress tests with minimal genomes under controlled bottlenecks are proposed as the most direct experimental route to further calibration or falsification of the resulting \(T_{\rm crit}\).

Empirical anchors for sparsity and sensitivity

Axe (2004) measured the prevalence of sequences capable of forming a working enzyme domain of modest size (~150 residues). Starting from a weakly functional \(\beta\)-lactamase variant already constrained to a favorable hydropathic signature, random replacement of local side-chain clusters recovered low-level activity in only ~1 in \(10^{64}\) signature-consistent sequences. Incorporating the additional rarity of the signature itself and of folds competent for a given catalytic chemistry produced an overall estimate of roughly 1 in \(10^{77}\) random sequences of that length that can adopt a stable, functional fold. Complementary experiments (Axe, 2000) demonstrated that even highly conservative exterior substitutions, when accumulated to approximately one change in five surface sites, abolish in-vivo function for unrelated monomeric enzymes. Sequence context, not merely active-site chemistry, therefore imposes severe constraints.

Independent stability surveys reinforce the same picture. Tokuriki & Tawfik (2009) and subsequent FoldX-based analyses across diverse globular proteins show that the average mutation lowers thermodynamic stability by ~0.5–1 kcal mol\(^{-1}\). Because the typical folding free-energy window of a natural protein is only 3–10 kcal mol\(^{-1}\), a modest number of substitutions (often fewer than ten) drives the majority of molecules below the stability threshold required for reliable folding and activity. Parallel deep-mutational-scanning campaigns on essential bacterial proteins (e.g., TEM-1, DHFR, and ribosomal components) consistently recover inactivation rates of 30–70 % for single amino-acid changes once context is taken into account, far higher than the neutral fraction assumed in classical infinite-sites models.

Formal implication for the ratchet

Let \(\mathcal{F} \subset \Sigma^L\) denote the set of amino-acid sequences of length \(L\) that retain a given essential function. Axe-type measurements imply \[ \frac{|\mathcal{F}|}{|\Sigma|^L} \lesssim 10^{-70}. \] Under the neutral or nearly-neutral diffusion that operates once purifying selection has already fixed the least-loaded class, each generation performs an independent random walk of expected length \(\mu L\). The mean first-exit time from \(\mathcal{F}\) is therefore finite and short relative to geological timescales; the return time from the complementary set \(\Sigma^L \setminus \mathcal{F}\) is, for practical purposes, infinite under purely asexual dynamics. Horizontal gene transfer can re-import an intact sequence only while donor lineages still occupy \(\mathcal{F}\); once those donors themselves exit, the rescue term \(\rho_{\rm HGT}\) in Equation 4.7 vanishes.

Consequently, the effective mutation rate that appears in the composite survival probability (Equation 4.10) is no longer a constant \(\mu_{\rm eff}\) but a time-dependent quantity that increases once the population has left the functional island. The median extinction time \(T_{\rm crit}\) is thereby bounded from above by the exit time of the least-loaded class, independent of further increases in absolute population size. In short, sequence-space sparsity closes the last conceptual loophole left open by selection and limited recombination: there is no large, connected neutral network that could allow indefinite persistence. The entropy horizon is not merely statistical; it is topological.

Empirical Calibration

5.1 Purpose and Approach

This chapter calibrates the free parameters of the model derived in Chapter 4 against existing published datasets, in order to generate a numerical estimate (with explicit caveats) of \(T_{crit}\), and to test whether that estimate is consistent with the falsifiable prediction stated in Chapter 2 (H1: \(10^{4}\text{-}10^{5}\) years). Calibration proceeds in three steps: (i) fixing structural parameters (\(N\), \(f_{ess}\), \(\mu\)) directly from sequencing literature; (ii) calibrating the ratchet-dampening constant against the LTEE’s observed non-extinction over 60,000 generations; (iii) propagating calibrated parameters through the composite model (Equation 4.10-4.11) across a range of population sizes and HGT rates representative of natural bacterial habitats.

5.2 Fixed Structural Parameters

The following parameters are taken directly from published, independently replicated measurements and are treated as fixed inputs rather than fitted quantities:

5.3 Calibrating the Ratchet-Dampening Constant

The dampening function \(g(N_{pop}, \rho_{HGT}) \propto 1/[N_{pop}(1+\rho_{HGT})]\) (Equation 4.7) was specified only up to a proportionality constant \(C\) in Chapter 4. This constant is calibrated using the single strongest available empirical anchor point: the LTEE’s observed absence of population extinction across 60,000 generations at \(N_{pop} \approx 3.3\times10^{7}\), which implies that population-level extinction probability \(P_{ext}(60{,}000)\) must be small — here bounded at \(\le 10^{-3}\), consistent with no extinction event having been observed across twelve independent replicate populations. Solving Equation 4.10 for \(C\) under this constraint yields:

\[ C \approx 8.5 \times 10^{-6} \]

This is a single-point calibration and represents the primary methodological limitation of the present chapter (addressed in Section 5.6): a single anchor point can fix the scale of \(g(N_{pop})\) but cannot independently establish its functional form or its dependence on \(\rho_{HGT}\), both of which were assumed rather than fitted.

5.4 Propagating Calibration Across Representative Scenarios

Using the calibrated \(C\), \(T_{crit}\) was numerically evaluated (Equation 4.11) across four scenarios spanning the range of population sizes and HGT rates plausible for natural bacterial habitats:

Two results are notable. First, the calibrated \(T_{crit}\) estimates across the actively-dividing scenarios cluster tightly around \(1.1\text{-}1.2 \times 10^{5}\) years, despite population size varying across eleven orders of magnitude. Second, this near-invariance is very likely an artifact of the single-point calibration in Section 5.3, since the calibration constant \(C\) was fit using a functional form that scales inversely with \(N_{pop}\), largely cancelling the \(N_{pop}\) dependence in the final \(T_{crit}\) calculation. This is flagged explicitly as a methodological artifact, not as an independently confirmed empirical finding, and is the central limitation carried into Chapter 6.

5.5 Cross-Check Against Independent Dormancy Data

As an independent check not dependent on the LTEE-calibrated constant, the model’s dormancy-adjusted prediction (Equation 4.9) is compared against direct observational data on bacterial survival in dormant states: confirmed revivals from Naica cave crystals (10,000-50,000 years; Boston et al., 2001) and Lechuguilla Cave isolates (isolated for millions of years, though continuous viability across that full span is not directly demonstrated; Bhullar et al., 2012). These independently observed timescales fall within, or below, the calibrated model’s \(10^{4}\text{-}10^{5}\) year range, providing consistency with, but not independent statistical confirmation of, the model’s central prediction (H1).

5.6 Summary of Calibration Limitations

Three limitations of this calibration must be stated plainly, consistent with the falsifiability standard set in Chapter 2:

• The ratchet-dampening constant \(C\) is calibrated from a single anchor point (LTEE, 60,000 generations, one population-size regime); it has not been independently validated against a second, differently-scaled dataset, and the resulting near-invariance of \(T_{crit}\) across population sizes should be treated as provisional
• The HGT rate \(\rho_{HGT}\) values used in Section 5.4 are illustrative estimates rather than direct field measurements, since natural HGT rates across free-living bacterial populations remain sparsely characterized in the literature (Chapter 3.5)
• The background dormancy damage rate \(\kappa\) (Equation 4.9) was not independently calibrated in this chapter and is treated as a free parameter requiring dedicated future measurement (Chapter 7)

These limitations do not overturn the model’s central estimate, but they substantially narrow the confidence that should currently be placed in the specific numerical value of \(T_{crit}\), and they define the concrete, falsifiable empirical work required to strengthen or refute the model, discussed further in Chapter 6.

Predictions and Pre-Registered Falsification Criteria

6.1 Purpose of Pre-Registration

Consistent with the falsifiability standard established in Chapter 2, this chapter states, in advance of further data collection, the precise empirical observations that would strengthen, weaken, or falsify the central hypothesis (H1) and its calibrated estimate (Chapter 5). Pre-registration of falsification criteria is intended to prevent the model from being adjusted post hoc to accommodate any future result, and to give the field a fixed, citable target against which subsequent empirical work — potentially over decades — can be measured. Each criterion below specifies the observation, the dataset or experimental design that could produce it, and the specific consequence for the model.

6.2 Primary Prediction (Restated)

Under conditions of unconstrained external resources, a genetically isolated, non-recombining bacterial lineage is predicted to lose functional genomic continuity within \(T_{crit} \approx 10^{4}\text{-}10^{5}\) years (Chapter 4, Equation 4.11; Chapter 5, Section 5.4). This estimate is provisional and calibrated from a single anchor point (Chapter 5.6); the criteria below are designed to test it independently.

6.3 Falsification Criterion 1: Direct Ancient-Genome Persistence

Observation that would falsify H1: Recovery and sequencing of a bacterial genome from a verified geological or archaeological context older than \(10^{5}\) years, demonstrating (a) confirmed continuous, non-dormant lineage history (not simple long-term dormancy, which is separately modeled via \(\kappa\)), and (b) mutational load below the model’s predicted threshold \(L_{crit}\) for that elapsed time.

Design: Ancient-DNA recovery from stratified sediment cores, permafrost, or amber inclusions with independently dated deposition, combined with metagenomic reconstruction and comparison to modern relatives, following methodology comparable to the 400-year-old E. coli genome reconstruction from a 16th-century gallstone (Devault et al., 2017), extended to older and better-dated contexts.

Consequence: Direct falsification of H1 as stated; would require either substantial upward revision of \(T_{crit}\) or rejection of the model’s core assumption that mutation accumulation is unbuffered beyond the ratchet-HGT dampening term (Equation 4.7).

6.4 Falsification Criterion 2: Ratchet-Rate Scaling at Very Large Population Sizes

Observation that would falsify H1: Direct sequencing-based measurement, across a controlled range of population sizes spanning several orders of magnitude beyond those tested in the LTEE (\(N_{pop} > 10^{10}\)), showing that the ratchet-dampening function \(g(N_{pop})\) (Equation 4.6) decays faster than the \(1/N_{pop}\) scaling assumed in this thesis — specifically, evidence that \(g(N_{pop}) \to 0\) at some finite \(N_{pop}\) rather than approaching zero only asymptotically.

Design: Large-scale chemostat or bioreactor evolution experiments maintaining population sizes several orders of magnitude larger than the LTEE, with whole-population metagenomic sequencing at fixed generation intervals to directly estimate mutational load trajectories, following an experimental logic analogous to Lenski et al. (1991) but scaled upward.

Consequence: Falsification of the specific functional form assumed in Equation 4.7; would require re-derivation of the calibration in Chapter 5 and likely upward revision of \(T_{crit}\), potentially removing the finite bound predicted by H1 for sufficiently large natural populations (supporting \(H_0’\), Chapter 2, Equation 2.2).

6.5 Falsification Criterion 3: HGT Sufficiency Test

Observation that would falsify H1: Empirical demonstration that horizontal gene transfer rates in a natural, non-bottlenecked bacterial population are sufficient to fully restore lost genetic diversity within a timescale shorter than the ratchet’s fixation time, across multiple successive fixation events, with no measurable net increase in mutational load over an extended observation period (minimum 10,000+ generations in a natural or semi-natural setting).

Design: Field-based metagenomic time-series sampling of a well-mixed natural bacterial population (e.g., a large, continuously flowing marine or freshwater system) with direct measurement of HGT event frequency via comparative genomics, combined with mutational load tracking.

Consequence: Falsification of the assumption in Equation 4.7 that \(\rho_{HGT}\) only dampens, rather than nullifies, ratchet-driven load accumulation; would require the dampening term to be re-modeled as a hard cutoff rather than a multiplicative factor.

6.6 Falsification Criterion 4: Dormancy Background Damage Rate

Observation that would falsify or substantially revise the dormancy component (Equation 4.9): Direct measurement, in a controlled setting, of a background (replication-independent) DNA damage rate \(\kappa\) in dormant bacterial cells that is either (a) effectively zero over multi-millennial timescales, which would extend predicted dormant-state persistence far beyond \(10^{5}\) years, or (b) substantially higher than assumed, which would compress it.

Design: Controlled long-term dormancy experiments (feasible at decade, not millennial, scale) measuring DNA lesion accumulation rates in metabolically arrested cells under defined environmental conditions, combined with extrapolation informed by direct genomic comparison of confirmed ancient dormant isolates (e.g., further Naica- or Lechuguilla-type recoveries with rigorous contamination controls) against modern relatives.

Consequence: Direct revision of the \(\kappa\) parameter in Chapter 4-5; would not falsify the model’s core mechanism but would substantially shift the predicted \(T_{crit}\) for dormancy-dominated natural scenarios.

6.7 Criteria That Would Strengthen, Rather Than Falsify, H1

For completeness and to avoid asymmetric framing, the following observations would constitute positive, corroborating evidence for H1 without yet constituting proof:

• Additional independently-calibrated anchor points (beyond the single LTEE point used in Chapter 5) yielding consistent \(T_{crit}\) estimates in the \(10^{4}\text{-}10^{5}\) year range from unrelated experimental systems (e.g., independent long-term evolution experiments in other bacterial or archaeal taxa)
• Confirmed ancient-genome recoveries in the \(10^{3}\text{-}10^{4}\) year range showing mutational load trajectories consistent with model predictions at that timescale, providing intermediate validation ahead of the full \(10^{4}\text{-}10^{5}\) year horizon

6.8 Timeline and Feasibility

Criteria 1 and 4 are the most experimentally demanding and are unlikely to be resolvable within the timeframe of this doctoral work alone; they are stated explicitly as multi-decade research targets for the field, consistent with the author’s stated intent that this model remain a standing, testable prediction subject to future refutation or confirmation (Chapter 1.4). Criteria 2 and 3 are more immediately tractable using existing large-scale experimental evolution infrastructure and are proposed as the primary empirical targets for follow-up work within five to ten years of this thesis’s completion (Chapter 7).

Discussion: An Open, Standing Prediction

7.1 What This Thesis Has, and Has Not, Established

This work has proven that bacterial species die within a fixed number of years smaller than previously considered. And it has done something even narrower, and, properly understood, is very valuable: it has taken an informal folk intuition — that unicellular organisms are “immortal” because they leave no corpse — and subjected it to the discipline of formal modeling, explicit parameterization, and empirical calibration against the best available long-term datasets. What emerged from that discipline is a specific number, \(T_{crit} \approx 10^{4}\text{-}10^{5}\) years, derived not by assertion but by tracing a mechanistic chain from measured replication fidelity, through population-genetic theory with a seventy-year pedigree, to calibration against sixty thousand generations of directly observed bacterial evolution. Whether that number survives the next twenty years of scrutiny is, deliberately, not something this thesis can settle alone. That is not a weakness in the argument. It is the argument.

7.2 The Central Tension, Restated Honestly

Every chapter of this thesis has circled the same tension: bacteria populations, empirically, do not match any timescale we have previously assumed. It is a flaw, or rather a feature, of the underlying biology. Muller’s ratchet was not invented to serve this thesis’s argument. It was discovered in 1964, independently rediscovered and formalized by Felsenstein, Haigh, Lynch, and Gabriel across three subsequent decades, and it survives in the literature today not because it is unchallenged but because every serious challenge to it has narrowed its domain of applicability rather than eliminated it. The 1% fitness decline observed in Andersson and Hughes’s bottlenecked Salmonella lines was not a modeling artifact. It was a measurement.

What Higgs’s counter-argument correctly shows — and this thesis has tried throughout to represent that argument fairly rather than dismiss it — is that the ratchet’s practical bite depends enormously on ecological detail: whether mutations affect intrinsic survival or only competitive ability, whether populations can regrow from bottlenecks, whether the fitness landscape is smooth or rugged. That is precisely why Chapter 6 does not treat this thesis’s number as settled. It treats it as a claim staked against a specific, falsifiable set of future observations, several of which — larger chemostat experiments, deeper ancient-DNA recovery, direct HGT-sufficiency measurement — are within the technical reach of the field within the coming decade, not the coming century.

7.3 Why the Number Matters More as a Method Than as a Fact

There is a version of this thesis that would have been safer to write: one that concluded, after four chapters of careful derivation, that “more research is needed” and declined to commit to a number at all. That version would have been methodologically unimpeachable and scientifically inert. Numbers are what allow a field to accumulate. Kelvin’s estimate of the Earth’s age was wrong by two orders of magnitude, and it was still one of the most productive wrong numbers in the history of geophysics, because its wrongness was precise enough to be found, and the search for why it was wrong led directly to the discovery of radioactive decay. Arrhenius’s ionic theory, later refined beyond recognition, still earned its author a Nobel Prize, because a falsifiable, quantitative claim — even one destined for revision — gives a field something to push against that a qualitative hedge does not.

This thesis’s contribution, if it has one worth defending twenty years from now, will not be the specific digits of \(10^{4}\) to \(10^{5}\). It will be the architecture that produced them: a demonstration that entropy accumulation, ratchet dynamics, horizontal gene transfer, and dormancy can be coupled into a single system of equations that yields a testable number at all, where previously these four literatures sat in separate journals, cited separately, tested separately, and never made to answer the same question at the same time. If, in twenty years, a successor to this work finds that \(\kappa\) was underestimated by three orders of magnitude, or that \(g(N_{pop})\) decays far faster than \(1/N_{pop}\) at population sizes this thesis could not test, the number will change. The equations, and the discipline of stating in advance what would change them, will not have been wasted.

7.4 On the Discomfort of Staking a Claim

There is a reasonable objection, likely to be voiced by at least one examiner, that a single-anchor-point calibration (Chapter 5.3) is too thin a reed to support a headline number this specific, and that a more cautious thesis would present \(T_{crit}\) as an order-of-magnitude sketch rather than a bounded prediction. That objection deserves to be taken seriously rather than deflected, and this thesis’s answer to it is not to retreat from the number but to make the thinness of its support explicit and load-bearing: every limitation identified in Chapter 5.6 has been converted, in Chapter 6, into a specific experiment capable of correcting it. A calibration built on one anchor point is a legitimate first estimate precisely because its fragility has been mapped rather than hidden. Science has a long history of provisional numbers that mattered — Hubble’s original estimate of the expansion rate of the universe was off by a factor of seven, and it still launched a research program that, decades later, produced the modern value to within a few percent. The value of staking the claim is not that the first number is right. It is that a wrong, precise number, openly declared as provisional, is more useful to a field than no number at all.

7.5 What Remains Genuinely Open

Three questions, left unresolved by this thesis and stated here without false confidence, define the space in which the central prediction will be tested over the coming years. First, whether the ratchet-dampening function \(g(N_{pop})\) genuinely approaches zero only asymptotically, as assumed, or whether some threshold population size exists beyond which purifying selection becomes, for practical purposes, fully sufficient — a question only answerable by experiments at population scales larger than any long-term evolution experiment has yet sustained. Second, whether horizontal gene transfer in truly natural, unbottlenecked populations operates at rates closer to the modest dampening assumed here or closer to the full recombination-equivalent rescue that would place bacterial populations formally outside the reach of Muller’s ratchet altogether. Third, and perhaps most tractable in the near term, whether the background, replication-independent damage rate governing dormant persistence is small enough that the empirically confirmed 10,000-50,000-year cave and crystal revivals represent a small fraction of what is achievable, or whether those revivals already sit close to a hard ceiling.

7.6 Closing Statement

This thesis does not close a question. It opens one, in the specific, disciplined sense that Popper meant when he distinguished science from dogma: not by refusing to commit to an answer, but by committing to an answer precise enough that the world can tell it whether it is wrong. The claim that a bacterial lineage, however favorably placed, cannot outrun its own genome indefinitely is offered here not as a final verdict but as a marker driven into the ground — a number, a set of equations, and four explicit experiments capable of moving it. Whether it stands, in whole or in part, is a question for the field to answer over the next twenty years, and this thesis’s only demand of that answer is that it be empirical.

Apendix 1

The Entropy Horizon in Bacterial Lineages: Model Predictions, Recent Niche Losses, and a Testable Polar-Reservoir Hypothesis

1. The entropy-horizon model

The entropy-horizon model starts from a simple empirical observation: DNA replication is imperfect. In any lineage that transmits its genome largely without recombination, each generation adds a net load of mutations. When the product of the effective mutation rate and the size of the essential genome is non-zero, mutational load grows without an equilibrium. The median time at which a lineage loses functional genomic continuity is finite. That time is lengthened by large population size, residual horizontal gene transfer, and purifying selection, and is shortened when those buffers are reduced.

The model therefore predicts that asexual or sparsely recombining bacterial lineages will reach the horizon on timescales set by their demographic and genetic parameters. Free-living lineages with large populations and some horizontal gene transfer are expected to persist longer than lineages confined to small, bottlenecked niches with limited opportunity for genetic exchange.

2. A documented recent niche loss

A concrete illustration is provided by the 2021 reconstruction of microbial genomes from human paleofeces (approximately 1 000–2 000 years old) recovered in the southwestern United States and Mexico. Among the high-confidence ancient gut genomes, roughly 39 % represented previously undescribed species-level bins. The spirochaete Treponema succinifaciens (and related members of the phylum) was present in every ancient sample examined. The same taxon remains detectable in some present-day non-industrial human populations but is virtually absent from industrialized human gut microbiomes.

This pattern constitutes a well-documented niche extinction: the disappearance of specific bacterial lineages from the industrialized human gut over the last several centuries to a millennium. It is not a demonstrated global species extinction; the organisms or close relatives persist elsewhere. Within the entropy-horizon framework the loss is interpreted as the accelerated advance of the ratchet inside a niche whose buffering parameters (effective population size, strain diversity, opportunity for horizontal gene transfer) were reduced by dietary homogenization, antibiotic exposure, and sanitation changes.

3. A polar-reservoir hypothesis

Viable bacteria are routinely recovered from polar ice and permafrost. Melting of sea ice, glaciers and permafrost demonstrably releases a fraction of those cells into meltwater and downstream environments. Laboratory and field studies show that some of the exported taxa remain detectable, at least transiently, in the receiving water or soil.

The entropy-horizon model supplies a natural hypothesis about the genomic state of these cells. Because replication is minimal or intermittent while a cell remains frozen, the mutational load accumulated during the frozen interval is expected to be lower than the load accumulated by continuously dividing free-living relatives over the same calendar time. Polar ice and permafrost therefore function, under the model, as a passive reservoir of lineages whose position on the mutational ratchet was arrested at an earlier point.

Empirical observations only.

Viable bacteria and archaea have been recovered from polar and glacial ice and from permafrost.

• Culturable isolates and long DNA amplicons have been obtained from Siberian permafrost horizons, glacial ice (including basal ice >750 kyr), Antarctic subglacial material, and cave ice (~13 kyr). Some samples independently dated by cosmogenic ³⁶Cl or other methods have yielded living cells; the oldest reported culture-positive horizons reach multi-hundred-thousand to multi-million-year claims, subject to the documented uncertainties of the dating methods.
• In a subset of the older permafrost samples, measurable low-level respiration and recovery of long intact DNA amplicons indicate that some cells maintained metabolic activity sufficient for DNA repair while frozen. Certain reconstructed genomes from ~100 kyr layers show minimal damage signatures consistent with long-term maintenance.

Melting ice and thawing permafrost release microbes into adjacent environments.

• Microcosm and field studies of Arctic and Antarctic sea ice show that a fraction of bacterial and microeukaryotic taxa present in the ice become detectable in the resulting meltwater (approximately one-tenth of amplicon sequence variants in one Fram Strait experiment). Larger cells and aggregates are preferentially exported.
• Glacier-to-fjord and glacier-foreland surveys document export of glacial/foreland microbial taxa into proglacial waters and soils. In downstream marine sediments the contribution of those glacial taxa declines sharply (e.g., to ~14 % of metagenome-assembled genomes in one Svalbard continuum), with marine lineages dominating.
• Laboratory thaw of permafrost cores produces measurable increases in microbial biomass, community composition shifts, and elevated respiration within weeks to months. Initial cell turnover is slow.

Comparative genomic data exist for some ancient versus modern isolates of the same species. These show differences in gene content and sequence, but no published empirical quantification demonstrates systematically lower genome-wide mutational load in the ancient isolates, nor a measurable contribution of polar reservoirs to slowing lineage-extinction rates in modern bacterial populations.

These are the measured facts concerning recovery, viability, DNA integrity, and physical release of polar/glacial microbes.

If a measurable fraction of the released cells can establish in modern communities, the model predicts a low-level, ongoing input of lower-load genomes into those communities. The same logic raises the possibility that glacial or permafrost-derived cells occasionally appear in metagenomic or culture-based surveys of modern habitats that receive meltwater—an occurrence that would constitute a form of biological “contamination” detectable by damage-pattern analysis, unique allele matching, and phylogenetic placement against polar reference libraries.

4. Empirical boundaries of the hypothesis

The hypothesis does not assert that ancient polar isolates have been shown to carry systematically lower genome-wide mutational load, nor that they have been demonstrated to slow lineage-extinction rates or restore ecological functions at scale. Comparative genomics of matched ancient and modern isolates exist for only a limited number of taxa and have not yet quantified a general load advantage. Downstream persistence of glacial taxa is limited by strong environmental filtering; in several studied continua the contribution of glacial lineages declines sharply once marine or local soil conditions predominate. No formal inventory places Treponema succinifaciens or its close relatives among the dominant permafrost taxa.

These are the present empirical constraints. The reservoir idea remains a model-derived, falsifiable hypothesis.

5. Testable predictions and an invitation to research

The hypothesis generates clear, experimentally accessible predictions:

• Ancient polar isolates of taxa that also occur in modern temperate or tropical environments should, on average, display lower load in essential genes than their continuously active counterparts, after controlling for phylogeny and cultivation bias.
• Meltwater plumes and proglacial soils should contain a detectable fraction of cells whose DNA-damage profiles, allele spectra or strain phylogenies match polar reference libraries rather than local modern populations.
• In habitats that receive sustained glacial or permafrost input, the frequency of lower-load genotypes of shared taxa should be higher than in otherwise comparable habitats that lack such input.

Each prediction can be tested with existing methods: single-cell genomics, damage-aware metagenomic assembly, long-read sequencing of essential-gene loci, and controlled mesocosm experiments that introduce authenticated polar isolates into defined recipient communities.

The entropy-horizon model supplies a quantitative language for mutational load and a set of demographic parameters that govern how quickly that load becomes intolerable. The paleofeces record supplies a recent, well-documented case of niche loss consistent with the model’s expectations under reduced buffering. The polar-reservoir hypothesis simply asks whether the frozen compartments of the cryosphere have been, and continue to be, a slow source of lineages that have spent less time on the active ratchet. The data required to accept or reject the hypothesis are obtainable. The invitation is open.

Apendix 2

The Entropy Horizon and the Fossil Record: Model Predictions, Multicellular Consequences, and a Quantitative Projection

1. The model in brief

The entropy-horizon model rests on a single empirical fact: DNA replication is imperfect. In any lineage that transmits its genome with little or no recombination, each generation adds a net mutational load. Load therefore increases linearly with the number of replications. When the product of the effective mutation rate and the size of the essential genome is non-zero, functional genomic continuity is lost after a finite number of generations. Large population size, residual horizontal gene transfer and purifying selection slow the process; their reduction accelerates it. The resulting characteristic time is the entropy horizon of that lineage.

2. Retroduction of the fossil and phylogenetic record

The model makes three broad, testable expectations about the deep-time record.

First, free-living asexual or sparsely recombining lineages should turn over continuously rather than persist indefinitely. Large-scale phylogenetic reconstructions of bacterial diversity show precisely this pattern: extinction rates of approximately 0.03–0.05 per lineage per million years in recent intervals, and the inference that the great majority of all bacterial lineages that have ever existed are already extinct. Overall bacterial diversity has nevertheless increased, consistent with continuous speciation balancing continuous extinction.

Second, obligate endosymbionts, which experience extreme population bottlenecks and near-zero horizontal gene transfer, should reach the horizon faster. Comparative genomics documents repeated genome degradation, fragmentation and eventual loss or replacement of ancient endosymbionts across multiple insect clades. These replacements occur on evolutionary timescales shorter than those of free-living relatives, matching the model’s ranking of horizons by degree of asexuality and effective population size.

Third, sexual multicellular lineages, buffered by meiosis and large effective population sizes, should display far longer genomic continuity. The Phanerozoic fossil record of animals and plants is dominated by external drivers—climate change, sea-level fluctuation, volcanism and impact events—rather than by steady internal genomic attrition. Background extinction rates exist, yet the major pulses are geologically abrupt and geographically widespread, consistent with external forcing acting on lineages whose internal horizons lie well beyond the duration of most sedimentary packages.

The model therefore recovers the observed contrast: continuous, relatively steady turnover among microbial and endosymbiotic lineages, and episodic, externally triggered losses among complex multicellular taxa.

3. Consequences for multicellular life

Multicellular organisms do not escape the horizon; they inherit it in two ways.

Mitochondrial genomes remain essentially asexual and maternally transmitted. High pedigree mutation rates imply ongoing load accumulation. Over successive host generations the frequency of mild-to-moderate mitochondrial dysfunction is expected to rise, imposing a gradual metabolic cost on animal and plant populations.

More consequentially, multicellular life depends on microbial partners for nutrient cycling, digestion, defence and primary production. When free-living or symbiotic bacterial lineages reach their individual horizons, the ecological functions they perform must be re-supplied by other lineages or lost. The model does not require simultaneous collapse of entire functional guilds; it predicts staggered attrition. Temporary or permanent gaps in nitrogen fixation, fibre degradation, or other key processes would propagate upward, reducing the carrying capacity or stability of the multicellular assemblages that rely on them.

Empirical illustrations already exist at small scale: the documented loss of certain spirochaete lineages from industrialized human guts coincides with broader reductions in microbiome diversity and is associated with modern disease patterns. Endosymbiont replacements in insects have been linked to host fitness declines or host-lineage extinctions when replacement is incomplete. These cases supply micro-scale analogues of the larger cascade the model anticipates.

4. A quantitative projection

Under the calibrated parameters used for free-living bacterial lineages (post-repair mutation rate ≈ 5 × 10^{-10} per base pair per generation, essential genome fraction ≈ 0.85, and the measured buffering from selection and residual horizontal gene transfer), the median time to loss of genomic continuity lies near 1.1–1.2 × 10^5 years.

Because multicellular ecosystems rest on continuous microbial services, the characteristic time for systemic degradation of those services under uninterrupted attrition is longer than the median bacterial horizon. A conservative order-of-magnitude estimate that allows for functional redundancy, residual horizontal gene transfer among free-living taxa, and the longer horizons of sexual nuclear genomes places the scale of cumulative, biosphere-level consequence near 10^6 years.

This figure is not a precise countdown to the extinction of any particular species, nor a claim that external catastrophes will be absent. It is the model-derived timescale on which internal genomic attrition, operating continuously across asexual compartments, is expected to produce measurable, cascading reductions in the stability and diversity of multicellular life if no major compensatory processes intervene.

5. Closing

The entropy-horizon model recovers the broad pattern of continuous microbial turnover and episodic multicellular extinction recorded in the fossil and phylogenetic archives. It identifies mitochondria and microbial partners as the principal conduits through which mutational load reaches complex organisms. The quantitative projection of roughly one million years for cumulative biosphere-level effects follows directly from the calibrated bacterial horizons once ecological dependence is taken into account.

The projection is falsifiable. Improved measurements of effective mutation rates, horizontal-gene-transfer frequencies and essential-genome sizes in natural populations, together with high-resolution records of microbial functional gene diversity through geological time, can raise or lower the estimated horizon. The invitation to refine the parameters and test the cascade is open.