B10 — Protein Folding and Levinthal Paradox Resolution
| Field | Value |
|---|---|
| Domain | Biology |
| System | Globular protein on Ramachandran φ/ψ torus |
| Group | SO(2) × SO(2) (φ/ψ torus at H⁰/H¹); π₁(fold space) at H² |
| H^k tier | H¹ |
| ISA | Forge (β ≈ β*) |
| Status | Validated |
| Opcodes | ORBIT · TWIST · BIND · LABEL |
| Papers | Paper 515, Paper 510, Paper 490 |
Physical system
Levinthal’s paradox (1969): a 100-residue protein has ~3¹⁰⁰ ≈ 10⁴⁸ backbone conformations (Ramachandran-unrestricted). At 10¹³ conformational transitions per second, exhaustive search takes 10³⁵ seconds. Observed folding times: 10⁻⁶ to 10⁻³ s. The gap is ~10³⁸ — the protein cannot find its native state by random search, yet it does so reproducibly in milliseconds.
The ISA resolution is not “the energy landscape is a funnel” (true but incomplete). The ISA identifies three successive tier-specific mechanisms that eliminate the Levinthal states sequentially — each operating on a different cohomological level with a different group:
H⁰ — Ramachandran geometry (group: SO(2) × SO(2), the φ/ψ torus): The backbone dihedral angles (φ, ψ) per residue live on a 2-torus T² ≅ SO(2)×SO(2). The peptide bond planarity (partial double bond character) and van der Waals steric constraints restrict ~60% of T² as forbidden. The two dominant LABEL attractors are the α-helix basin (φ≈−57°, ψ≈−47°) and β-sheet basin (φ≈−120°, ψ≈+120°). The H⁰ LABEL constraint alone eliminates ~10⁸⁰ of Levinthal’s states — instantaneously, by geometry, not by search.
H¹ — Secondary structure cooperativity (group: U(1) helical phase): α-helix formation is an H¹ ORBIT on the Ramachandran torus with a cooperative attractor: once residues i, i+1, i+2 enter the α-basin, residue i+3 is geometrically forced in by the i→i+4 H-bond. β-sheet formation is a non-local ORBIT closure requiring two strand segments from distant sequence positions to simultaneously enter the β-basin. The H¹ cooperative ORBIT eliminates ~10⁵⁰ further conformations: the search space contracts from backbone conformations to secondary-structure arrangements.
H² — Tertiary topology (group: π₁(fold space), knot group for knotted proteins): Assembling secondary-structure elements into the native 3D fold is an H² ORBIT. The native fold topology (which helices pack against which, domain connectivity, knot type) is a global H² invariant that cannot be deduced from local H¹ information. The rate-limiting step of folding is H² obstruction resolution — consistent with the observed folding rate scaling:
τ_fold ∝ exp(c × RCO)
where RCO = relative contact order (a proxy for H² obstruction depth; x515a measured r = −0.978 for unknotted proteins, p = 8.75 × 10⁻¹²).
Target category
FoldCat(n) — the category of protein conformational states for an n-residue chain. Objects: conformations C = {(φᵢ, ψᵢ)}ᵢ₌₁ⁿ ∈ (T²)ⁿ subject to the Ramachandran constraint; the H⁰ subobject is the set of allowed Ramachandran basins. Morphisms: conformational transitions that pass through the H¹ ORBIT gates (H-bond formation) or the H² topology-changing moves (domain repacking, knot passage).
The native state is the terminal object of FoldCat(n) under the Anfinsen thermodynamic hypothesis — every folding pathway converges to it. The H^k hierarchy gives the structure of the morphisms: H⁰ (local geometry, instantaneous), H¹ (secondary structure, 10–100 ns), H² (tertiary topology, 10 μs–10 ms).
Interpretation functor
F: C → FoldCat(n) defined by:
| Opcode | F(opcode) |
|---|---|
| LABEL | Ramachandran basin assignment: φᵢ/ψᵢ → {α, β, coil}; eliminates H⁰ forbidden zones; discrete LABEL per residue; instantaneous (bond geometry constraint); eliminates ~10⁸⁰ of Levinthal’s 10¹⁴³ states |
| ORBIT | Secondary structure nucleation: cooperative H-bond closure; α-helix = i→i+4 ORBIT with 3-step cooperative lock; β-sheet = long-range ORBIT requiring two-strand simultaneous β-basin entry; timescale 10–100 ns; eliminates ~10⁵⁰ further states |
| TWIST | Hydrophobic collapse: H¹→H² boundary event; non-polar residues cluster to form a hydrophobic core; TWIST gate at the H¹/H² boundary (β > β*); timescale 1–10 μs |
| BIND | Tertiary topology closure: H² ORBIT must close consistently across the entire chain; long-range contacts between secondary-structure elements form the native topology; rate-limiting step ΔG_H²; knotted proteins = non-trivial element of H²(fold; ℤ), folding 100–1000× slower |
ISA programme
-- Phase 0: sequence (H0 register)
LABEL[sequence | mRNA -> amino acids; H0 LABEL per residue]
-- Phase 1: Ramachandran constraint (H0, instantaneous)
LABEL[phi_i, psi_i -> {alpha, beta, coil} | Ramachandran LABEL]
-- Eliminates ~10^80 of Levinthal's states by geometry
-- Phase 2: secondary structure (H1 ORBIT, 10-100 ns)
ORBIT[helix_nucleation | cooperative: residue i locks i+1,i+2,i+3]
ORBIT[sheet_pairing | two-strand simultaneous beta-basin entry]
-- Eliminates ~10^50 further states; search space -> secondary structure arrangements
-- Phase 3: hydrophobic collapse (H1/H2 TWIST, 1-10 us)
TWIST[hydrophobic_core | non-polar burial, beta > beta*]
-- Phase 4: tertiary topology (H2 ORBIT, 10 us - 10 ms)
BIND[domain_packing | H2 ORBIT close; rate proportional to exp(-c*RCO)]
-- Rate-limiting; knotted proteins need knot-passage BIND moves (100-1000x slower)
-- Phase 5: chaperone QEC (if needed)
BIND[GroEL_capture | wrong H2 attractor detected by hydrophobic exposure]
LABEL[ATP_hydrolysis | irreversible beta* commit; 7 ATP per GroEL cycle]
ORBIT[refolding_in_cage | H2 ORBIT attempt in isolated GroES environment]
BIND[release if H2 correct | else recapture]
Computable output
-
Folding rate correlation with RCO (x515a, Paper 515): for 26 unknotted two-state folders, r(RCO, ln k_f) = −0.978 (literature best: −0.72–0.85 from Plaxco 1998). Full model r² = 0.815, p = 8.75 × 10⁻¹². ISA interpretation: RCO is a proxy for H² obstruction depth — each 0.01 increase in RCO slows folding by ~1.6×. The correlation is the quantitative LABEL output of the H² obstruction theory.
-
Knotted protein slowdown (x515a): trefoil-knotted proteins (YibK, YbeA, UCH-L1) lie 2.3–2.9 log units below the unknotted regression line; figure-eight knotted DehI lies 4.2 log units below. ISA predicted slowdown for trefoil: 331× (observed: 100–1000×). The BIND moves needed to pass the knot through the chain during folding are categorical H² obstructions — not reducible to any sequence of H¹ ORBIT steps.
-
Amyloid prediction from H¹ orbit attractor analysis (x515b, Paper 515): the ISA amyloid score S = β-propensity × (1 + hydrophobic moment period-2) − 1.5 × charge density uses zero training data (parameters from physics only). On the Garbuzynskiy 2010 hexapeptide benchmark (n=80): AUC = 0.909, ranking 1/9 methods tested, including TANGO, AGGRESCAN, Waltz, PASTA. The ISA identifies amyloid-prone sequences as those whose H¹ β-ORBIT has a second H² attractor accessible from the β-basin.
-
GroEL substrate classification by H² complexity (x515c, Paper 515): using the Kerner/Hartl 2008 Class I/II/III E. coli chaperone substrate dataset (90 proteins): H² complexity (= RCO) increases monotonically I → II → III (ANOVA F=71.2, p = 4.88 × 10⁻¹⁹). Binary discrimination Class III vs I+II: AUC = 0.981. Class III substrates (obligate GroEL) are enriched 67% for TIM-barrel folds (the most topologically complex H² fold class); Class I (chaperone-independent) = 0% TIM-barrel.
The group structure: why SO(2) × SO(2) at H⁰ and π₁ at H²
H⁰ group: SO(2) × SO(2)
The Ramachandran torus T² = SO(2) × SO(2) is the natural configuration space per residue (two continuous rotation angles φ, ψ). The H⁰ Ramachandran constraint is a subset of T² — the allowed region — defined by the steric clash condition. The two main LABEL attractors (α and β basins) are connected components of the allowed region separated by the TWIST gate (the forbidden region boundary).
The SO(2) × SO(2) group is not a unification fiction: the backbone dihedral rotation about the N−Cα bond IS an SO(2) rotation (φ), and about the Cα−C bond IS an SO(2) rotation (ψ). These are physical rotational degrees of freedom. Their product structure T² = SO(2) × SO(2) is the correct mathematical object.
H¹ group: U(1) helical phase
The α-helix winding is a U(1) phase: each residue adds 100° of rotation and 1.5 Å of axial translation, making the helix a U(1) ORBIT with period 3.6 residues. The helical phase θ = 100° × i mod 360° is a U(1) coordinate on the helix ORBIT.
H² group: π₁(fold space) / knot group
For unknotted proteins, the native fold topology class is discrete (labelled by its SCOP/CATH fold class) and the relevant group is effectively trivial (ℤ₂ for fold handedness). For knotted proteins, the knot group is the fundamental group of the complement of the knot K in ℝ³: π₁(ℝ³ \ K). For the trefoil knot (most common knotted fold): π₁ = ⟨a, b | a² = b³⟩ = the trefoil knot group. The BIND move that threads the chain to create or destroy the knot is a generator of this group.
Summary: the group of protein folding is tier-dependent:
- H⁰: SO(2)×SO(2) (Ramachandran torus; continuous, local)
- H¹: U(1) (helical phase; periodic, cooperative)
- H²: π₁(fold space) (knot group or fold topology class; discrete, global)
This tier-dependence is the honest ISA answer — and it is more informative than any single Lie group assignment.
Chaperones as H² QEC (structural identity with Paper 510)
The ISA identifies chaperones as an H²-level error-correction hierarchy, structurally identical to kinetic proofreading (Paper 510) at the H¹ level:
| Chaperone tier | Chaperone | H^k | Action |
|---|---|---|---|
| Co-translational | Trigger factor, RAC | H¹ | Local ORBIT reset; single-cycle |
| Cytosolic early | Hsp70 (DnaK) | H¹ | Bind/release; local secondary structure reset |
| Late maturation | Hsp90 | H¹/H² | Near-native client; co-chaperone assisted |
| Barrel encapsulation | GroEL/GroES | H² | Global topology reset; 7 ATP/cycle; confirmed by x515c |
| Disaggregase | ClpB/Hsp104 | H²/H³ | Dissolves committed amyloid fibrils; speculative H³ |
The GroEL–mismatch repair identity (exact at ISA level):
| Step | GroEL/GroES (H²) | DNA mismatch repair (H¹) |
|---|---|---|
| Error detect | Hydrophobic patch (wrong H² fold) | MutS (misincorporated base) |
| Isolate | GroES cage closure | Strand nicking + MutL |
| Irreversible commit | 7× ATP hydrolysis | ATP hydrolysis MutL/MutS |
| Correct | Refolding in cage | Exonuclease + resynthesis |
| Release | Native H² topology | Correct H¹ sequence |
| Fidelity gain | 10–100× per cycle | 100–1000× per repair |
Both follow the programme: CAPTURE → ISOLATE → RESET(ATP) → REORBIT → MEASURE → RELEASE or RECAPTURE. The difference is only the H^k level of the error (H¹ for DNA sequence, H² for protein topology).
The ISA/AlphaFold boundary (Paper 515 §5)
The ISA and AlphaFold answer different questions:
| Question | ISA | AlphaFold |
|---|---|---|
| What does this protein compute? | Yes (ISA programme) | No |
| Will it misfold or form amyloid? | Yes (H¹ attractor analysis) | Partial |
| Does it need a chaperone? | Yes (H² complexity score) | No |
| What are the Cartesian coordinates? | No | Yes |
| What does the disordered loop look like? | No | Yes |
The ISA is zero-compute, zero-training-data for the questions it answers — it operates at the level of orbit theory (which fold class, which programme, which attractor), not at the level of atomic positions.
Connections to other entries
- B07 RNA polymerase / Paper 510 (Kinetic Proofreading): proofreading is H¹ error correction; GroEL/GroES is H² error correction; the ISA hierarchy H¹→H² maps to proofreading→chaperone in the same organism (E. coli Pol III → GroEL)
- C04 RNR radical transfer: metalloenzyme active sites are predicted from the ISA orbit vector (Paper 490 pipeline); the ISA pre-selects the active-site H¹ coordination geometry; B10 pre-selects the fold H² topology; together they fully specify the metalloenzyme programme without AlphaFold
- B06 Krebs cycle: the TIM-barrel fold (67% of GroEL Class III substrates) is the most common enzymatic fold; TIM-barrel enzymes include many Krebs-cycle enzymes (e.g., aconitase, succinate dehydrogenase subunit C); the H² folding complexity of TIM-barrels is why they depend on GroEL
Validation
- Levinthal (1969): paradox stated; folding time gap 10³⁸ s.
- Plaxco, Simons & Baker (1998), J. Mol. Biol. 277, 985: contact order correlation with folding rate (r = −0.72–0.85 literature); x515a improves to r = −0.978.
- Mallam & Jackson (2007), J. Mol. Biol. 366, 650: knotted proteins fold 100–1000× slower than unknotted; ISA predicted 331× from H² obstruction theory.
- Kerner et al. (2008), Cell 122, 209: GroEL substrate classification I/II/III; x515c confirmed H² complexity predicts class with AUC = 0.981.
- Garbuzynskiy et al. (2010): hexapeptide amyloid benchmark; x515b AUC = 0.909, rank 1/9, zero training data vs 8 trained methods.
- Jumper et al. (2021), Nature 596, 583: AlphaFold2; confirms ISA/AlphaFold boundary — ISA answers what it computes, AlphaFold answers where atoms are.