LTEE Extensions

Date: March 1, 2026 Status: Proposal with quantitative predictions. Anderson anomaly catalog, critical disorder threshold, and dilution amplification validated. 🧬👁️ helixVision (structure prediction) primitives validated (154 checks). Agricultural time-series test bed established. Domain: Evolutionary biology, microbial genomics, structural biology Novelty: Specific, falsifiable Anderson-QS predictions for LTEE populations; integration of constrained evolution signatures across lab, field, and agricultural sample archives; self-hosted structure prediction for tracking protein fold evolution across 75,000+ generations

Abstract

The constrained evolution framework generates specific, testable predictions for how microbial populations evolve under environmental constraint. We propose applying these predictions to three complementary sample archives: the Lenski Long-Term Evolution Experiment (LTEE, 75,000+ generations at MSU), permafrost thaw microbial communities (deep-time natural experiment), and agricultural soil time series (contemporary managed constraint). Each archive tests a different facet of the theory: the LTEE tests convergence and contingency in a controlled setting; permafrost tests constraint release after millennia of stasis; agricultural time series test constraint engineering through human management.

1. The LTEE as the Gold Standard

1.1 What Exists

The Lenski LTEE (Michigan State University, started 1988) maintains 12 replicate populations of Escherichia coli in glucose minimal medium with daily serial transfer. The frozen fossil record preserves samples from every 500 generations — a time machine for evolution.

Key published findings (Lenski et al. 1991, Wiser et al. 2013, Blount et al. 2008, 2012, Tenaillon et al. 2016):

• Fitness increases follow a power law, not asymptote
• Parallel evolution across replicates (convergent pathways)
• Historical contingency (Ara-3 citrate innovation required potentiating mutations)
• Genome streamlining in late generations
• Substantial neutral/hitchhiker fraction in fixed mutations

1.2 What the Constrained Evolution Framework Predicts

From thesis Chapter 14, we predict:

1.3 The Anderson-QS Layer

A novel prediction from Sub-thesis 01 (Anderson localization):

If LTEE populations are grown in biofilm format (3D structure) vs planktonic format (shaken flask), QS regulon expression should differ according to Anderson’s dimensional prediction.

E. coli has a partial QS system (sdiA receptor, no synthase — the eavesdropper strategy). In biofilm, sdiA should respond to any added AHL signal; in planktonic, the same signal should fail to coordinate due to dilution-amplified Anderson disorder (Exp137: W_eff = W_base / occupancy).

This is a testable bench experiment at MSU using existing LTEE populations.

2. Permafrost Thaw Communities

2.1 Rationale

Permafrost preserves microbial communities under absolute constraint (frozen, no metabolism, no evolution) for 10,000-100,000+ years. When thawed, these communities resume evolution under modern conditions — a natural constraint- release experiment.

2.2 Constrained Evolution Predictions

• Immediate diversity crash: frozen community meets modern competitors it has never co-evolved with → rapid selection under novel constraint
• QS re-emergence timing: if thawed community forms biofilm (3D), Anderson predicts QS within the community structure regardless of diversity. If community remains dispersed in meltwater (planktonic), QS fails.
• Convergent adaptation: thawed populations should show accelerated convergence toward the same adaptive peaks occupied by modern analogs (the fitness landscape is shaped by constraint, not starting genotype)

2.3 Sample Sources

• Arctic permafrost cores: active layer microbial ecology well-characterized (Mackelprang et al. 2011, Jansson & Taş 2014)
• Antarctic dry valley soils: extremely low diversity (J near 0 → low Anderson disorder → QS possible even in 2D mats)
• Rika Anderson’s deep-sea vent archives (Carleton College): Sulfurovum populations under continuous extreme constraint

3. Agricultural Soil Time Series

3.1 Rationale

Agriculture is managed constraint: tillage, irrigation, crop rotation, and agrochemical application reshape the soil microbiome annually. Long-term agricultural experiments (LTAR, Broadbalk at Rothamsted) maintain archived soil samples spanning decades.

3.2 Anderson-QS Predictions for Agriculture

Testable with archived soil DNA: compare QS gene prevalence (luxI/luxR, lasI/lasR) between no-till and conventional-till plots from the same LTAR site and year.

🌬️♨️ airSpring contributes complementary agricultural time series: a 60-year water balance (Wooster OH, Triplett-Van Doren dataset) parallel to LTEE as a long-term archive; cover crop dual Kc + no-till validation (40/40 checks) providing agronomic context for QS predictions in tilled vs no-till soil; and the Michigan Crop Water Atlas (100 stations, 80 years simulated) as a massive temporal dataset for agricultural time series analysis.

3.3 Pivot Bio Connection

The Pivot Bio model (engineering N-fixation via seed-coat inoculant) relies on the inoculant reaching, colonizing, and maintaining QS-mediated gene regulation in the root zone. The Anderson model predicts this works because:

1. Root surface → 3D biofilm structure → QS-active (Exp127-130)
2. Inoculant is a monoculture → J near 0 → W near 0.5 → deep in extended regime
3. As soil diversity invades the biofilm → W increases → but stays below W_c in 3D

Failure mode: if the inoculant disperses into bulk soil without forming biofilm → planktonic dilution → Anderson localization → QS regulation fails → N-fixation gene expression drops. This predicts that biofilm-forming ability of the inoculant strain is the critical success factor.

4. Integration: Three Archives, One Framework

     LTEE                Permafrost            Agricultural
  (controlled)          (natural)              (managed)
      │                     │                      │
  75K gens              10K-100K yrs           50-150 yrs
  12 replicates         1 thaw event           annual cycles
      │                     │                      │
      └─────────┬───────────┴──────────────────────┘
                │
    Constrained Evolution Predictions:
    • Convergent pathways (not identical mutations)
    • Power-law temporal dynamics
    • Hitchhiker burden proportional to constraint strength
    • Anderson geometry determines QS activity
    • Historical contingency for innovations

If all three archives show the same signatures, the constrained evolution principle is established as domain-general, independent of timescale, control level, or environmental specifics.

5. Practical Requirements

5.1 Computational Pipeline

The 💧♨️ wetSpring sovereign pipeline (16S, DADA2, chimera, taxonomy — all validated, Exp001-070) can process all three sample types. Additionally:

• WGS assembly (MAGs from metagenomes) → pangenomics for convergence detection
• QS gene annotation via the HMM profiles built for Exp140-142
• Anderson geometry assignment from sample metadata (biofilm/planktonic/mat)
• Level spacing ratio computation via 🐸🍄 ToadStool primitives (anderson_3d, lanczos)

5.2 neuralSpring Integration

🧠♨️ neuralSpring adds ML primitives for LTEE analysis (S135: 966 lib tests, 232 binaries, 220/220 validate_all, 3,034+ checks, 5 WDM surrogates complete):

• HMM / PhyloNet-HMM (Liu Papers 016-018): Introgression detection applied to LTEE genomes — identify horizontal transfer events and adaptive introgression across 75,000 generations
• Transfer learning (Exp 004, nW-04): Cross-environment adaptation models. Training on one LTEE population, testing on others, mirrors the constrained-evolution prediction that convergent pathways (not identical mutations) recur across replicates. nW-04 demonstrates classical→WDM transfer learning — the same framework applies to training on one LTEE replicate and transferring to another
• LSTM time series (Study 004, NSE=0.849; nW-03 LSTM reservoir, R²=0.98): Predict temporal dynamics of mutation accumulation and fitness plateaus across the 12 LTEE populations. nW-03’s pooled-readout LSTM (mean + std + last hidden state after washout) validates the sequence processing pipeline for extracting temporal features from biological time series
• ESN regime classifier (nW-05, 96.5% accuracy): Classify evolutionary regimes (e.g., pre-citrate vs post-citrate Ara-3) from population genomic features using reservoir computing. Fixed-weight ESN + ridge readout requires no backpropagation — suitable for rapid regime detection on streaming data

5.3 groundSpring Integration

⛰️♨️ groundSpring contributes uncertainty quantification and stochastic modeling directly relevant to LTEE analysis:

• Exp 014 — Drift vs selection (R. Anderson 2022): Wright-Fisher fixation probability and Kimura neutral theory. Quantifies when stochastic drift dominates deterministic selection — the central question for interpreting LTEE mutation fixation trajectories. Predicts the ~30-50% hitchhiker fraction in §1.2 above. 7/7 Py, 7/7 Rust checks
• Exp 017 — Quasispecies threshold (Dolson 2023): Eigen’s error threshold predicts when mutation rate destroys genetic information. Directly applicable to LTEE’s observed power-law fitness dynamics — the mutation-selection balance determines whether adaptive trajectories are signal (selection) or noise (drift). 6/6 Rust checks
• Exp 016 — Rare biosphere signal detection (R. Anderson 2015): Sequencing depth determines the boundary between real biological signal and sampling artifact. Critical for LTEE frozen fossil analysis — detecting rare mutant lineages that will later become dominant (pre-potentiating mutations for the citrate innovation). 10/10 Rust checks
• Exp 019 — Jackknife estimation (Bazavov 2025): Subpercent precision error bars via delete-one and block jackknife. The standard method for quantifying uncertainty in population genomic measurements from LTEE samples. 9/9 Rust checks
• Exp 004 — Sequencing noise rarefaction: Genus saturation at 5,000 reads; phyla robust at 100 reads. Establishes the sampling floor for LTEE community-level analysis. 15/15 Rust checks

Cross-spring pipeline for LTEE: 💧♨️ wetSpring (16S/WGS pipeline) → ⛰️♨️ groundSpring (sampling noise floor + jackknife error bars + drift vs selection classification) → 🧠♨️ neuralSpring (HMM introgression + ESN regime detection). This three-spring pipeline covers the full LTEE analysis workflow from raw sequencing to evolutionary inference.

5.4 Wet Lab (MSU Resources)

• MSU Genomics Core: Illumina sequencing for LTEE frozen samples
• MSU RTSF: high-throughput sequencing for soil/environmental DNA
• LTEE access: Lenski Lab, MSU Department of Microbiology and Molecular Genetics
• Soil archives: MSU LTAR (Kellogg Biological Station)

6. Falsification Criteria

This sub-thesis is falsifiable:

• If LTEE populations show identical mutations (not pathway convergence) → falsified
• If no-till and tilled soil have the same QS gene prevalence → Anderson prediction falsified
• If permafrost thaw communities skip the diversity crash → constraint-release model falsified
• If biofilm vs planktonic LTEE shows no difference in sdiA expression → Anderson-QS model falsified

The predictions are specific, quantitative, and testable with existing infrastructure at MSU.