Daryl Costello

Independent Theoretical Research | Rosendale, NY, United States

Submitted: October 19, 2026

Theoretical Physics • Cognitive Science • Theoretical Biology • Complex Systems • Theoretical Cosmology

✉ Correspondence: Daryl.Costello@outlook.com

Theoretical Monograph: October 2026

Keywords: gauge theory, epigenomics, fiber bundle, cell identity, lineage commitment, chromatin topology, hematopoiesis, Waddington landscape, holonomy, topological defects, Yang-Mills, clonal hematopoiesis

Abstract

The question of how a single metazoan genome gives rise to hundreds of phenotypically distinct and heritably stable cell types represents one of the deepest unsolved problems at the intersection of molecular biology and dynamical systems theory. Existing quantitative frameworks (Boolean network models, ordinary differential equation-based gene regulatory networks, and energy landscape models) lack a natural, coordinate-free language for describing the symmetry structure of epigenetic regulatory equivalences: the empirical fact that a vast and heterogeneous ensemble of molecular configurations can produce indistinguishable transcriptional outputs.

Here we introduce Genomic Gauge Theory (GGT), a mathematical framework in which cell identity is encoded as a section of a fiber bundle 𝐸 over a base manifold ℬ of transcription factor (TF) combinatorial binding states. In this formulation, epigenetic mark configurations constitute the fiber ℱ above each point in TF state space, and the rules of epigenetic co-regulation and inheritance define a connection 𝒜 (a gauge field) on this bundle. Local gauge symmetry corresponds precisely to biologically equivalent gene regulatory configurations, and a gauge transformation represents an epigenetic reprogramming event that reparametrizes the fiber without altering transcriptional output.

We define genomic curvature (the Yang-Mills field strength ℱ = d𝒜 + 𝒜 ∧ 𝒜) as a tensorial measure of regulatory tension: the degree to which epigenetic programs at neighboring TF states are mutually incompatible. Non-vanishing curvature identifies primed, transitional, and dysregulated cellular states. Cell type identity is topologically encoded in the holonomy group of the connection, and we prove that two cell types are topologically irreconvertible by epigenetic means alone if and only if their holonomy groups are non-isomorphic. Stable topological defects in the connection (instantons, vortices, and monopoles) provide a gauge-theoretic account of oncogenic epigenetic alterations and clonal evolution.

We apply this framework in full detail to the hematopoietic system, demonstrating that the GATA1/PU.1 lineage decision corresponds to a Z₂ gauge symmetry breaking instanton, that clonal hematopoiesis of indeterminate potential (CHIP) represents metastable low-charge topological defect accumulation in the HSC bundle, and that acute myeloid leukemia transformation constitutes a topological phase transition beyond a critical Pontryagin charge threshold. We further propose experimentally testable predictions — curvature estimation from single-cell multiome data, holonomy measurement from paired perturbation-recovery profiling, and Pontryagin index inference from bulk epigenomic compendiums — and discuss implications for cellular reprogramming, synthetic cell engineering, and precision epigenomics.

1. Introduction

The human body contains approximately 37 trillion cells, all descended from a single totipotent zygote and all carrying an essentially identical genomic sequence. Yet these cells adopt and stably maintain roughly 200 distinct phenotypic identities; from the compact transcriptional silence of a mature neutrophil to the expansive permissive chromatin of an embryonic stem cell, from the precisely timed firing of a Purkinje neuron to the rhythmic contractility of a cardiomyocyte [1]. This differentiation is not merely a matter of which genes are expressed; it is a question of how expression states are stabilized against perturbation, inherited through cell division, and refractory to reversion under normal physiological conditions. The molecular machinery responsible for this stabilization (covalent histone modifications, DNA methylation, non-coding RNA silencing, three-dimensional chromatin architecture, and the cooperative binding of transcription factor ensembles) is collectively termed the epigenome.

The most influential conceptual framework for understanding this process remains the epigenetic landscape of Conrad Hal Waddington [2]. In this metaphor, a cell is imagined as a ball rolling down a landscape of hills and valleys; the valleys are stable cell fates (attractors), the ridges are fate boundaries, and differentiation corresponds to the ball descending from a high plateau of developmental potential into one of the terminal valleys. This image captured, with extraordinary prescience, the notions of developmental canalisation, multi-stability, and irreversibility that would later be formalized in dynamical systems theory. Its extraordinary influence is evidenced by its persistent reappearance in the contemporary single-cell biology literature more than sixty years after its introduction [3].

Yet the Waddington landscape, as originally conceived and as commonly deployed, suffers from a fundamental limitation: it is purely metaphorical and geometrically under-specified. The landscape is typically drawn as a two-dimensional surface with no prescription for how the axes are to be defined, how distances are to be measured, or how the heights of the hills are to be computed from molecular quantities [4]. Several research programs have attempted to supply this missing mathematical structure. Boolean network models of gene regulatory networks (GRNs) assign binary activity states to genes and enumerate stable attractors as candidate cell types [5,6]. Ordinary differential equation (ODE)-based GRN models supply quantitative kinetics but face acute problems of parameter identifiability and do not naturally represent the stochasticity of chromatin remodeling [7]. The energy landscape formalization of Huang, Bhattacharya, and colleagues [8,9] posits that the Waddington potential can be identified with a quasi-potential function derived from the GRN steady-state probability distribution, providing a rigorous thermodynamic grounding. Manifold learning approaches applied to single-cell RNA-sequencing (scRNA-seq) data (including diffusion maps [10], PHATE [11], UMAP [12], and pseudotime trajectory methods [13]) offer powerful data-driven reconstructions of developmental geometry without mechanistic assumptions.

Despite these advances, a fundamental conceptual gap persists: none of the existing frameworks possesses a natural language for representing the symmetry structure of regulatory equivalences. Biologically, it is well established that many distinct molecular configurations can produce the same transcriptional phenotype. Multiple combinations of activating histone marks can sustain the expression of a given gene; promoters can interact with alternative enhancers to produce equivalent output; different TF combinations can activate the same regulatory element; and methylation patterns at CpG-poor loci are largely dispensable for silencing maintained by Polycomb complexes [14,15,16]. These redundancies are not incidental complications to be averaged away; they represent a deep structural property of the regulatory system (a gauge symmetry) and a proper mathematical treatment of cell identity must accommodate them from the outset.

Gauge theory, developed in its modern form by Yang and Mills in 1954 [17] and subsequently formalized within the language of fiber bundles by Atiyah, Singer, and others [18,19], is precisely the branch of mathematics designed to exploit such symmetries. In gauge theories of physics, the requirement of invariance under local symmetry transformations completely determines the form of the interactions between matter fields. The photon of quantum electrodynamics, the W and Z bosons of the electroweak theory, and the gluons of quantum chromodynamics all arise as the “gauge bosons” that mediate forces while maintaining local symmetry. The mathematical scaffold underlying all of these theories (a principal fiber bundle with a structure group acting on fibers, equipped with a connection form whose curvature encodes field strength) is precisely the scaffold we propose to adapt for the epigenomic context.

The central thesis of this paper is that cell identity is a section of a fiber bundle over TF state space, epigenetic inheritance is parallel transport via a connection on this bundle, and differentiation is a topological transition in the holonomy group of the connection. This perspective is not merely analogical. We provide a precise mathematical construction of the bundle, demonstrate that known biological mechanisms map naturally and non-trivially onto the geometric objects of the theory, derive experimentally testable predictions that distinguish our framework from existing approaches, and apply the full formalism to hematopoiesis as a worked example.

The approach is informed by, but distinct from, several recent geometric and topological initiatives in biology. Neural manifold theory models population activity as trajectories on low-dimensional manifolds embedded in neural state space [20]. Information geometry endows the space of probability distributions (and hence the space of gene expression profiles) with a natural Riemannian metric, the Fisher information metric [21]. Topological data analysis (TDA) has been applied to single-cell data to identify persistent topological features of developmental trajectories [22,23]. Our contribution is to provide the specific additional structure (a fiber bundle, a gauge group, and a connection) that these approaches lack, and to show that this structure carries biological information of a kind inaccessible to purely metric or topological analyses.

Summary of Principal Contributions 1.  Fiber Bundle Formulation. A precise mathematical construction of the epigenomic state space as a fiber bundle 𝐸 = (𝐸, π, ℬ, ℱ, 𝒢), with biological interpretations for all components. 2.  Epigenetic Connection. Definition of the connection 1-form 𝒜 encoding epigenetic co-regulation rules, and its biological identification with the histone modification landscape across the genome. 3.  Genomic Curvature Tensor. Definition and computation of the curvature 2-form ℱ as a coordinate-free measure of regulatory tension, with a proposed algorithm for estimation from single-cell multiome data. 4.  Holonomy Classification of Cell Types. A theorem establishing that topologically distinct cell types correspond to non-isomorphic holonomy groups, providing a coordinate-free classification of the differentiation hierarchy. 5.  Disease as Gauge Defect. A topological account of epigenetic diseases (including CHIP, AML, and polycomb-driven cancers) as stable topological defects (instantons, vortices, monopoles) in the hematopoietic epigenomic bundle, with quantitative predictions for clonal expansion and transformation rates.

2. Mathematical Formalism

2.1 The Epigenomic State Manifold

We begin by constructing the mathematical arena in which cell identity will be encoded. Let N denote the number of cis-regulatory elements (CREs) in the genome (encompassing promoters, enhancers, insulators, and silencers) and let K denote the number of transcription factor species in the organism of interest. In human cells, N ≈ 10⁵ to 10⁶ (as defined by ENCODE peak catalogs [24]) and K ≈ 1,600 (the estimated size of the human TF repertoire [25]).

We define the base manifold ℬ as the space of TF combinatorial occupancy states across all regulatory loci. Formally, each point b ∈ ℬ encodes, for each regulatory element i ∈ {1, …, N} and each TF species k ∈ {1, …, K}, the fractional occupancy (interpreted as a population-average or stochastic probability) of TF k at locus i. Thus:

ℬ ⊂ [0,1]^N×K ⊂ ℝ^N×K

(Eq. 1)

subject to biological constraints including TF competition for overlapping binding sites, cooperative and anti-cooperative binding, and nucleosome exclusion. These constraints render ℬ a proper submanifold of the ambient Euclidean space. The effective dimensionality of ℬ is vastly lower than NK due to the sparsity of TF binding, the modularity of enhancer grammar, and the cooperative nature of TF recruitment; empirically, dimensionality reduction of TF activity matrices across cell types recovers ≈10–30 principal axes [8,9], consistent with the biological observation that a small number of master regulators largely determines cell identity.

We endow ℬ with the structure of a smooth Riemannian manifold. Local coordinate charts on ℬ are provided by the activities of local TF regulatory modules; transition functions between charts are smooth, reflecting the fact that TF binding occupancy varies continuously with cellular context and signal concentration. The Whitney embedding theorem guarantees that ℬ, as a finite-dimensional smooth manifold, can be isometrically embedded in some sufficiently high-dimensional Euclidean space, providing a concrete computational representation.

Above each point b ∈ ℬ, we define the typical fiber ℱ as the space of epigenetic mark configurations at the regulatory loci, conditional on the TF occupancy state b. Specifically, we track M histone modification types (including at minimum H3K4me3, H3K27ac, H3K4me1, H3K27me3, H3K9me3, and H3K36me3) along with DNA methylation states (5-methylcytosine and 5-hydroxymethylcytosine) at all relevant CpG sites. Treating each mark level as a continuous variable in the unit interval (reflecting fractional modification rates in a cell population), we have:

ℱ ≅ [0,1]^N×M ⊕ ({0,1}^ℤ₂)^NCpG

(Eq. 2)

where the first term captures continuous histone marks and the second term captures the discrete Z₂-valued DNA methylation state at N_CpG CpG sites. For much of the formal development, we will treat ℱ as a smooth manifold by either restricting to the continuous component or by modeling the discrete Z₂ component as a background connection (see Section 3.3).

Definition 1: The Epigenomic Fiber Bundle The epigenomic state space is the fiber bundle 𝐸 = (𝐸, π, ℬ, ℱ, 𝒢) where: (i) 𝐸 is the total space, comprising all pairs (b, f) with b ∈ ℬ and f ∈ ℱb (the fiber above b); (ii) π: 𝐸 → ℬ is the projection map π(b, f) = b; (iii) ℱ ≅ [0,1]N×M is the standard fiber (the space of epigenetic mark configurations at a given TF state); (iv) 𝒢 = Diff(ℱ) is the structure group (the group of diffeomorphisms of ℱ) acting on fibers by the natural left action; (v) Local trivializations exist over open sets Uα ⊂ ℬ such that π−1(Uα) ≅ Uα × ℱ, with smooth transition functions gαβ: Uα ∩ Uβ → 𝒢.

A section of this bundle (a smooth map s: ℬ → 𝐸 such that π ∘ s = id_ℬ) corresponds precisely to a cell type: for each possible TF occupancy state b, the section specifies the epigenetic mark configuration s(b) ∈ ℱ_b that the cell maintains at that state. Global sections correspond to stable, fully differentiated cell types; local sections that cannot be extended globally correspond to primed or transitional states.

2.2 The Gauge Group and Local Symmetry

The central principle of gauge theory (that physically (biologically) observable quantities must be invariant under a local symmetry group) requires us to identify precisely which transformations of the fiber represent changes in mathematical description rather than changes in biological reality. We define the gauge group 𝒢 as the group of local epigenetic equivalences:

𝒢 = {g: ℬ → Aut(ℱ) | g smooth, g(b) preserves Φ(b) for all b ∈ ℬ}

(Eq. 3)

where Φ(b) ∈ ℝ^G denotes the vector of transcriptional outputs (gene expression levels) of all G genes at TF state b. In other words, g ∈ 𝒢 is a local gauge transformation at b if it maps epigenetic configurations to one another without changing the observable gene expression state. The condition that g preserves Φ(b) is the biologically essential constraint that distinguishes the gauge group from the full diffeomorphism group of ℱ.

Concretely, gauge transformations include: (i) exchange of H3K27ac enrichment between alternative enhancers of equal regulatory activity; (ii) substitution of one bivalent (H3K4me3 + H3K27me3) configuration for another of equivalent poising strength; (iii) redistribution of H3K9me3 within a constitutive heterochromatin domain without altering its gene-silencing effect; and (iv) re-methylation of CpG sites within a fully methylated domain. All of these operations change the epigenetic coordinates but leave the phenotype invariant.

Theorem 1: Phenotypic State Space as Orbit Space The space of physically distinguishable (phenotypically distinct) cell states is the orbit space 𝐸/𝒢, endowed with the quotient topology. Formally, two epigenomic configurations (b, f1) and (b, f2) in the same fiber are biologically equivalent (producing identical transcriptional output Φ(b)) if and only if there exists a gauge transformation g(b) ∈ Aut(ℱ) such that f2 = g(b) · f1. The projection ρ: 𝐸 → 𝐸/𝒢 maps each epigenomic state to its gauge orbit, and ρ corresponds precisely to the map from molecular configurations to observable cell identity. Proof sketch. By construction of 𝒢, the function Φ: ℱb → ℝG is constant on gauge orbits (𝒢-equivariant). The quotient topology on 𝐸/𝒢 is the finest topology making ρ continuous, and it corresponds to the topology of phenotypic state space as measured by transcriptomics. ∎

The Lie algebra 𝔤 of the gauge group 𝒢 consists of smooth maps from ℬ to the algebra of infinitesimal diffeomorphisms of ℱ. Biologically, elements of 𝔤 correspond to infinitesimal epigenetic perturbations that are phenotypically neutral; the instantaneous rates of change of epigenetic marks that do not alter transcriptional output. These are the directions in epigenetic space that are “flat” from the perspective of gene regulation, and they play a central role in defining the connection in the next section.

2.3 The Epigenetic Connection

A connection on the fiber bundle 𝐸 is a geometrically natural way of defining parallel transport: it specifies how a fiber element (an epigenetic state) is to be “carried along” as one moves through the base manifold ℬ. This corresponds biologically to the rules of epigenetic inheritance; how the chromatin state of a cell is propagated to its daughters during division, and how co-regulatory dependencies ensure coherent epigenetic programs across genomic loci.

We define the epigenetic connection 𝒜 as a 𝔤-valued 1-form on the base manifold ℬ:

𝒜 ∈ Ω¹(ℬ, 𝔤)

(Eq. 4)

For a tangent vector X ∈ T_bℬ at a point b ∈ ℬ, the value 𝒜(X) ∈ 𝔤 specifies the infinitesimal gauge transformation (the infinitesimal adjustment to the epigenetic mark configuration) required to maintain coherence as the TF state changes in the direction X. The biological content of this prescription is substantial: it encodes, for example, the fact that increased occupancy of PU.1 at a myeloid enhancer (motion in TF state space in the PU.1 direction) necessitates a correlated increase in H3K27ac at that locus (the corresponding connection component), reflecting the known biochemical activity of PU.1 in recruiting p300/CBP acetyltransferases [26].

The covariant derivative of a section s: ℬ → 𝐸 with respect to a vector field X on ℬ is:

∇_Xs = ds(X) + 𝒜(X) · s

(Eq. 5)

A section is parallel (∇s = 0) if its epigenetic state changes along any path in ℬ precisely according to the connection; i.e., if the cell’s chromatin state responds to changing TF occupancy in the “minimally stressful” way prescribed by the epigenetic inheritance machinery. Deviations from parallel transport (∇s ≠ 0) represent epigenetic stress (regulatory mismatches between the TF program and the chromatin configuration) that drive corrective chromatin remodeling.

Parallel transport along a path γ: [0,1] → ℬ is defined by the horizontal lift equation:

df/dt + 𝒜(γ'(t)) · f(t) = 0,    f(0) = f₀

(Eq. 6)

whose solution is the path-ordered exponential:

f(1) = 𝒫 exp(−∫_γ 𝒜) · f₀

(Eq. 7)

Biologically, this equation describes how the epigenetic state of a cell progenitor is inherited by its daughters as they traverse a differentiation trajectory γ in TF state space. The connection 𝒜 encodes all the enzymatic machinery of epigenetic inheritance: DNMT1-mediated maintenance methylation, CAF-1-mediated histone chaperone activity during replication, and PRC1-mediated monoubiquitination that recruits PRC2 for H3K27 trimethylation at the replication fork [27].

Under a gauge transformation g: ℬ → 𝒢, the connection transforms according to the standard Yang-Mills law:

𝒜 → 𝒜^g = g · 𝒜 · g⁻¹ + g · dg⁻¹

(Eq. 8)

The first term represents the conjugation of the connection by the gauge transformation; the second term (the Maurer-Cartan term) is the inhomogeneous piece that allows the gauge transformation to “rotate” the zero of the connection, analogous to the way a change of reference frame in electromagnetism adds a gradient to the vector potential. Biologically, this transformation law captures the fact that re-labeling the epigenetic coordinate system (changing the reference mark configuration at each locus) changes the mathematical form of the connection without changing the physical (transcriptional) output.

Proposition 1: Chromatin Remodelers as Local Gauge Transformations Chromatin remodeling complexes (including the SWI/SNF (BAF) complex, Polycomb repressive complexes PRC1 and PRC2, the NuRD deacetylase complex, and the DOT1L methyltransferase) act locally on the fiber ℱb above each point b ∈ ℬ, effecting gauge transformations of the connection form 𝒜 at individual loci without altering the global structure of the base manifold ℬ. Specifically, if the enzymatic activity of a chromatin remodeler at locus i and state b produces a mark-to-mark conversion f → g(b)f that leaves Φ(b) invariant, then this conversion is a local gauge transformation at b.

The empirical support for this proposition is substantial. The SWI/SNF complex repositions nucleosomes at enhancers, exchanging one accessible chromatin configuration for another of equivalent transcriptional activity [28]. PRC2 deposits H3K27me3 at developmental genes, but the precise patterns of trimethylation can vary across cell divisions while maintaining equivalent silencing; a gauge equivalence within the repressive connection component.

2.4 Genomic Curvature

The curvature of the connection 𝒜 is the fundamental local invariant of the gauge theory. It measures the extent to which parallel transport around an infinitesimal loop in TF state space fails to return to the starting fiber element; the degree to which epigenetic inheritance is path-dependent. We define the curvature 2-form (Yang-Mills field strength) as:

ℱ = d𝒜 + 𝒜 ∧ 𝒜 ∈ Ω²(ℬ, 𝔤)

(Eq. 9)

In local coordinates {b^μ} on ℬ, the components of the curvature are:

ℱ_μν = ∂_μ𝒜_ν − ∂_ν𝒜_μ + [𝒜_μ, 𝒜_ν]

(Eq. 10)

where [·, ·] is the Lie bracket in 𝔤. The commutator term [𝒜_μ, 𝒜_ν] is non-zero precisely when the gauge group is non-abelian (when different epigenetic regulatory factors do not commute) which is the generic case when multiple chromatin modifiers with overlapping targets are simultaneously active. The curvature ℱ transforms covariantly under gauge transformations: ℱ → g · ℱ · g⁻¹, confirming that it is a genuine physical observable (up to gauge choice).

Proposition 2: Flat Connection and Ground-State Pluripotency The curvature vanishes identically, ℱ = 0, if and only if there exists a gauge transformation g: ℬ → 𝒢 such that 𝒜g = 0 everywhere on ℬ. This condition (the existence of a globally flat gauge) corresponds biologically to a maximally plastic cellular state in which all epigenetic mark configurations are mutually consistent: any path through TF state space leads to the same fiber element regardless of history. This is the mathematical characterization of ground-state pluripotency (naïve ESC state), in which the epigenome is globally “reset” to a configuration of minimal regulatory constraint, consistent with the known hypomethylated, globally accessible chromatin state of naïve ESCs [29].

Conversely, non-vanishing curvature (ℱ ≠ 0) signifies regulatory tension: the epigenetic programs prescribed by the connection at neighboring TF states are incompatible, so that parallel transport around a loop in TF state space accumulates a net rotation of the fiber. Biologically, this corresponds to the coexistence of partially activated lineage programs; the molecular signature of primed pluripotent states, multipotent progenitors, and dysregulated cancer cells. The magnitude |ℱ|_b at a given point b ∈ ℬ quantifies the degree of this incompatibility.

The total regulatory stress across the entire epigenomic landscape is captured by the Yang-Mills action functional:

S[𝒜] = ∫_ℬ ||ℱ||²_g d Vol_ℬ = ∫_ℬ Tr(ℱ ∧ ★ℱ)

(Eq. 11)

where ★ is the Hodge star operator with respect to the Riemannian metric on ℬ, and the trace is taken in the adjoint representation of 𝒢. Connections that minimize S[𝒜] subject to appropriate boundary conditions satisfy the Yang-Mills equations:

∇ ★ ℱ = 0,     ∇ℱ = 0 (Bianchi identity)

(Eq. 12)

We interpret the Yang-Mills equations as the equations of epigenetic equilibrium: they describe the steady-state chromatin configuration of a differentiated cell type in which regulatory tensions are balanced and no net chromatin remodeling pressure exists. Connections satisfying the Yang-Mills equations correspond to stable cell types in the Waddington sense; perturbations that violate these equations drive the cell toward a new equilibrium by the epigenetic analog of the Lorentz force.

Beyond the Yang-Mills action, there exists an additional topological invariant of the bundle: the genomic Chern-Simons functional, defined on a three-dimensional submanifold ∂Ω ⊂ ℬ:

CS(𝒜) = ∫_∂Ω Tr(𝒜 ∧ d𝒜 + ⅔ 𝒜 ∧ 𝒜 ∧ 𝒜)

(Eq. 13)

Under a gauge transformation g, CS(𝒜) → CS(𝒜) + 4πk where k ∈ ℤ is the winding number of g around ∂Ω. The Chern-Simons functional is therefore gauge-invariant modulo 4π, analogous to the theta term in quantum chromodynamics. Its biological interpretation is as a topological invariant of the epigenetic program restricted to a three-dimensional subspace of TF state space; for instance, the three-dimensional subspace spanned by the activities of three master regulators governing a specific lineage branch point.

Theorem 2: Genomic Atiyah-Singer Index Theorem Let D𝒜 be the genomic Dirac operator defined on sections of the epigenomic bundle 𝐸, twisted by the connection 𝒜. Then the analytical index of D𝒜 (the difference between the dimensions of its kernel and cokernel) equals the topological charge of the connection: ind(D𝒜) = k = (1/8π²) ∫ℬ Tr(ℱ ∧ ℱ) ∈ ℤ This integer is invariant under continuous deformations of the connection (including all epigenetic perturbations that do not change the global topology of the bundle) and therefore constitutes a lineage-invariant topological quantum number. Different cell lineages, maintained by topologically inequivalent connections, are distinguished by distinct values of this index.

2.5 Holonomy and Cell Identity

The holonomy of the connection provides the most refined topological invariant of cell identity. For a smooth closed loop γ: [0,1] → ℬ with γ(0) = γ(1) = b₀, the holonomy Hol_γ(𝒜) ∈ 𝒢_b0 is the gauge transformation accumulated by parallel transport of any fiber element around γ:

Hol_γ(𝒜) = 𝒫 exp(−∮_γ 𝒜) ∈ 𝒢

(Eq. 14)

Biologically, closed loops in TF state space correspond to cell cycle trajectories: a cell progresses through S phase (during which TF activities may shift as a result of replication-coupled changes in nuclear organization and competition for binding sites) and returns to its original TF state. The holonomy Hol_γ(𝒜) measures the net irreversible epigenetic change accumulated over this cycle; the degree to which the chromatin state after one cell cycle differs from its state before, measured in the fiber above the starting point. If Hol_γ(𝒜) = 1 (the identity gauge transformation), the cell cycle is epigenetically transparent: no net commitment occurs. If Hol_γ(𝒜) ≠ 1, then each cell division accumulates a net epigenetic displacement in the fiber, driving progressive commitment.

The collection of all holonomies over all loops based at b₀ forms a group, the holonomy group:

Hol(𝒜, b₀) = {Hol_γ(𝒜) | γ a smooth loop at b₀} ⊂ 𝒢

(Eq. 15)

Theorem 3: Holonomy Classification of Cell Types Two cell types C₁ and C₂, corresponding to sections s₁ and s₂ of the epigenomic bundle with connections 𝒜₁ and 𝒜₂ respectively, are topologically distinct (in the sense that no continuous deformation of the connection (no continuous epigenetic reprogramming pathway) can transform one into the other while maintaining the biological constraint Φ ≡ const) if and only if their holonomy groups Hol(𝒜₁) and Hol(𝒜₂) are non-isomorphic as subgroups of 𝒢. Proof sketch. A continuous deformation of the connection is a homotopy 𝒜t, t ∈ [0,1], of gauge connections compatible with the biological constraint. By continuity of the path-ordered exponential in the connection, Hol(𝒜t, b0) varies continuously in t. Since holonomy groups are discrete invariants (they are subgroups of the Lie group 𝒢), continuous variation must be constant. Hence topologically distinct connections have non-isomorphic holonomy groups. ∎

The holonomy classification provides a natural and coordinate-free differentiation hierarchy:

• Naïve pluripotent stem cells (flat connection): Hol(𝒜) = {1}; trivial holonomy. Parallel transport around any loop returns to the starting fiber element. The chromatin state is globally reset after each cell cycle. This corresponds to the open, globally accessible epigenome of ground-state ESCs.
• Multipotent progenitors (abelian holonomy): Hol(𝒜) ≅ U(1)ⁿ for some n ≥ 1. The holonomy group is abelian, reflecting the fact that distinct lineage modules are independently accessible but not yet mutually exclusive. Different loops in TF state space commute in their epigenetic effects.
• Committed progenitors and terminally differentiated cells (non-abelian holonomy): Hol(𝒜) is non-abelian and irreducible. Distinct lineage-specifying TF pathways have non-commuting epigenetic effects, reflecting the mutually exclusive nature of terminal differentiation programs.

The Ambrose-Singer theorem [30] establishes that the holonomy algebra (the Lie algebra of the holonomy group) is generated by the values of the curvature 2-form ℱ along all paths. Biologically, this means that regulatory tension (curvature) is the generator of commitment: the irreversible epigenetic changes associated with lineage commitment arise precisely from the accumulation of regulatory incompatibilities that drive the holonomy group toward more complex (non-abelian, irreducible) configurations.

2.6 Topological Defects as Epigenetic Diseases

The most physically rich aspects of gauge theories arise from configurations that are locally smooth but globally non-trivia; topological defects. In the epigenomic context, these correspond to stable aberrations in the chromatin state that cannot be removed by any local (locus-by-locus) chromatin remodeling procedure, but require a global topological change in the bundle.

We classify the relevant defect types and their biological correspondences:

• Vortex defects (codimension 2): Configurations in which the connection 𝒜 has a singularity around which the fiber winds non-trivially. Biological correspondence: focal hypermethylation “islands”; for example, the aberrant CpG island methylation silencing the CDKN2A (p16^INK4a) tumor suppressor in early malignancy [31]. The methylation vortex creates a stable topological obstruction to reactivation of the gene by TF binding alone.
• Monopole defects (codimension 3): Configurations in which the connection has a source-like singularity in a three-dimensional TF activity subspace. Biological correspondence: chromosomal rearrangements (translocations, inversions, deletions) that create topological discontinuities in the regulatory fiber by physically severing the connection between enhancers and promoters [32].
• Instanton solutions (localized in all dimensions of ℬ): Finite-action, topologically non-trivial solutions to the Yang-Mills equations that connect two locally stable vacua (two Yang-Mills minima corresponding to distinct cell types) through a tunneling path of finite action. Biological correspondence: epigenetic reprogramming events; the induction of pluripotency by Yamanaka factors [33], the cytokine-driven trans-differentiation of macrophages, and leukemic transformation.

The topological charge of a configuration is quantified by the Pontryagin index (second Chern number):

Q = (1/8π²) ∫_ℬ Tr(ℱ ∧ ℱ) ∈ ℤ

(Eq. 16)

The Pontryagin index is topologically protected: it cannot be changed by any continuous deformation of the connection that does not pass through a gauge field singularity. It is therefore an integer-valued topological invariant of the epigenomic configuration, constant across the continuous epigenetic perturbations that maintain cell identity, and jumping by an integer amount upon discrete reprogramming events.

Corollary 1: CHIP as Pre-Leukemic Topological Defect Accumulation Clonal hematopoiesis of indeterminate potential (CHIP) corresponds to the accumulation of low-Pontryagin-charge (Q = ±1, ±2) instanton excitations in the hematopoietic bundle, driven by somatic mutations in epigenetic regulators (TET2, DNMT3A, ASXL1). These represent metastable pre-leukemic topological defects; configurations with non-zero topological charge that are locally stable (protected by a finite action barrier Sinst) but globally metastable with respect to further topological transitions toward overt leukemia. The clonal expansion of CHIP is driven by the lower Yang-Mills action of the mutant epigenomic configuration compared to wild-type HSCs under conditions of replicative stress.

3. Biological Mapping

3.1 Transcription Factors as Gauge Bosons

In the Standard Model of particle physics, gauge bosons are the force-carrying particles that mediate interactions between matter fields while maintaining the invariance of the theory under local gauge transformations. In Genomic Gauge Theory, transcription factors play an exactly analogous role: they mediate interactions between regulatory loci (the matter fields) and in doing so, propagate the epigenetic connection 𝒜 across the genome.

This analogy is not superficial. Consider the action of a TF on a regulatory element: by binding to its recognition sequence, the TF recruits chromatin-modifying complexes that alter the local histone modification state; that is, the local fiber element. The TF thereby transmits information from the base manifold (TF occupancy state) to the fiber (epigenetic marks), implementing the parallel transport rule encoded by 𝒜. This is precisely the role of a gauge boson: to implement the connection.

We distinguish two functional classes of TFs in gauge-theoretic terms:

Pioneer TFs as massless gauge bosons. FOXA1, OCT4, SOX2, KLF4, and other pioneer factors can bind to their recognition sequences in the context of closed, nucleosomal chromatin without prior chromatin remodeling [34]. In gauge-theoretic terms, these TFs propagate the connection across the entire base manifold ℬ without preferential coupling to any particular fiber configuration; they are “massless” in the sense that their effective range is unlimited and their binding is not restricted by the local epigenetic context (fiber element). This permits them to explore and restructure the full fiber at any point in TF state space, rationalizing their known capacity to initiate reprogramming from arbitrary starting chromatin states.

Lineage-specifying TFs as massive gauge bosons. GATA1, PU.1, CEBPA, PAX5, EBF1, and other lineage-restricted TFs require pre-existing accessible chromatin (a permissive fiber element) for productive binding and transcriptional activation [26]. In gauge-theoretic terms, these TFs have an effective “mass”; their ability to propagate the connection is restricted to fiber elements above a certain accessibility threshold. This mass arises from a symmetry-breaking mechanism directly analogous to the Higgs mechanism: the deposition of H3K27me3 by PRC2 at alternative lineage gene loci creates a “Higgs field” that restricts the effective range of lineage-specifying TFs to their cognate cell type. The condensation of this Higgs field (the establishment of a stable H3K27me3 landscape) gives “mass” (context-restricted binding) to the lineage TFs and simultaneously selects a specific vacuum (a specific section of the bundle) from the degenerate family of high-symmetry pluripotent configurations.

3.2 Chromatin Domains as Gauge Field Configurations

The three-dimensional organization of the genome provides a natural spatial discretization of the epigenomic fiber bundle. Topologically associating domains (TADs) (genomic regions, typically 100 kb to 1 Mb in size, within which chromatin contacts are highly enriched relative to contacts between domains [35]) define local gauge patches in our framework:

• Within a TAD: The connection 𝒜 is approximately flat (low curvature), because the regulatory elements within a TAD share a local chromatin environment, are subject to the same pool of TFs, and are co-regulated by the same set of chromatin modifiers. This corresponds to a local gauge choice in which 𝒜 ≈ 0 within each TAD.
• TAD boundaries: CTCF-bound insulator elements at TAD boundaries implement the transition functions g_αβ between adjacent gauge patches, specifying how the fiber coordinate system is to be transformed upon crossing from one TAD to the next. The insulating activity of CTCF (blocking enhancer-promoter communication across TAD boundaries) is the biological realization of the non-trivial transition function that prevents the gauge field from one patch from influencing the other.
• Compartment A/B switching: The large-scale reorganization of chromosomal compartments (the switch of genomic domains between the transcriptionally active A compartment and the repressive B compartment) represents a global gauge transformation affecting entire chromosomal domains, analogous to a large-scale change of gauge in a field theory on a compact space.

The cohesin complex, which mediates chromatin loop extrusion and is required for TAD organization [36], plays the role of the gauge field’s transport mechanism: it physically moves chromatin through the nucleus, implementing the spatial component of parallel transport and ensuring that enhancers and promoters within the same loop are connected by the same gauge patch (share the same fiber trivialization).

3.3 Histone Modifications as Connection Components

We now provide an explicit mapping between the components of the connection 1-form 𝒜 and the biochemically characterized histone modification marks. In a local trivialization with coordinate directions {b^μ} on ℬ corresponding to individual TF activity axes, the connection decomposes into active and repressive components:

𝒜 = 𝒜_active ⊕ 𝒜_repressive

(Eq. 17)

𝒜_active = α₁ · H3K4me3 ⊕ α₂ · H3K27ac ⊕ α₃ · H3K4me1 ⊕ α₄ · H3K36me3

(Eq. 18)

𝒜_repressive = β₁ · H3K27me3 ⊕ β₂ · H3K9me3 ⊕ β₃ · H2AK119ub1

(Eq. 19)

where the coefficients α_i, β_j are empirically determined from the correlation structure of ChIP-seq signals across cell types and conditions (see Section 4.1). The physical meaning is that as the TF state b changes in any direction, the required change in the histone modification landscape is prescribed by 𝒜(X) = Σ_μ 𝒜_μ db^μ(X).

Of particular biological importance is the bivalent chromatin domain; a locus simultaneously bearing the active H3K4me3 mark and the repressive H3K27me3 mark [37]. In gauge-theoretic terms, bivalency corresponds to non-zero curvature:

ℱ_bivalent = [𝒜_active,μ, 𝒜_{repressive,ν}] db^μ ∧ db^ν ≠ 0

(Eq. 20)

because the H3K4me3 and H3K27me3 components of the connection do not commute in 𝔤; they pull the fiber in opposite directions. Bivalency is therefore the genomic realization of curvature: a state of maximal regulatory tension, poised for resolution in either the active or repressive direction upon receipt of a differentiating signal. This explains the known enrichment of bivalent marks at developmental gene promoters in ESCs [37]; these genes are at the points of maximal curvature in TF state space, ready to be rapidly resolved upon commitment.

DNA methylation (5-methylcytosine, 5mC) and its oxidized derivative (5-hydroxymethylcytosine, 5hmC) contribute a discrete, Z₂-valued component to the connection:

𝒜_meth ∈ H¹(ℬ, ℤ₂)

(Eq. 21)

This discrete component acts as a topological constraint on the fiber that cannot be continuously deformed: DNA methylation, once established, requires active demethylation by TET enzymes or passive dilution through replication, making it a Z₂ topological charge that contributes to the overall Pontryagin index of the configuration.

3.4 Regulatory Defects and Disease

The gauge-theoretic framework provides a unified and precise language for describing the molecular basis of epigenetic diseases. We map the four most common classes of oncogenic epigenetic mutations onto specific gauge-theoretic defects:

Mutation	Gauge-Theoretic Effect	Predicted Consequence
EZH2 gain-of-function (Y641X)	Pathological amplification of 𝒜repressive (excess H3K27me3 deposition); over-flattening of the repressive connection component at tumor suppressor loci	Reduction in ℱ at developmental gene loci → loss of developmental plasticity; stable silencing of polycomb targets inconsistent with progenitor state
TET2 loss-of-function	Disruption of 5mC/5hmC connection component 𝒜meth; loss of the active demethylation pathway → accumulation of Z₂ topological charge at polycomb target loci	Increased curvature at bivalent loci; biased self-renewal over differentiation; expansion of HSC pool (flat-connection attractor)
IDH1/2 mutations	Production of 2-hydroxyglutarate (2-HG) as competitive inhibitor of TET enzymes and KDMs; global distortion of both 𝒜active (via KDM inhibition) and 𝒜meth (via TET inhibition)	Global increase in |ℱ|; CpG island hypermethylation phenotype (CIMP); blockade of differentiation at progenitor TF state
KMT2A (MLL1) rearrangements	Fusion of MLL1 SET domain with aberrant transcriptional activators (AF4, AF9, ENL) splices a pathological parallel transport rule into the H3K4me3 connection component 𝒜active at HOX loci	Constitutive deposition of H3K4me3 at HOX genes independent of TF state; creation of a monopole defect in the active connection; maintenance of embryonic TF programs in committed progenitors

4. Experimental Observables and Predictions

4.1 Measuring the Connection

A critical test of any theoretical framework is its ability to make contact with experimental measurements. We show here that the standard toolkit of epigenomics provides direct access to specific gauge-theoretic quantities, though always in a gauge-dependent (coordinate-dependent) form. True observables (gauge-invariant combinations) require specific algorithmic constructions described below.

ChIP-seq as connection component measurement. A ChIP-seq experiment for histone mark H measures the pullback of the corresponding connection component to the genome coordinate. Specifically, if we use the genomic coordinate x ∈ [0, L] (bp along the chromosome) as a proxy for a coordinate on ℬ (via the assumption that genomic position correlates with regulatory module identity), then:

ChIP-seq(H, x) ∝ 𝒜_H(x) = ⟨𝒜, ∂/∂x⟩_H-component

(Eq. 22)

This measurement is gauge-dependent: a gauge transformation can redistribute ChIP-seq signal between loci without changing the observable transcriptional output. True biological observables must be gauge-invariant combinations, such as the curvature |ℱ|² or the holonomy Hol_γ.

ATAC-seq as gauge orbit volume. ATAC-seq (Assay for Transposase-Accessible Chromatin) measures chromatin accessibility; the fraction of genomic time that a locus spends in a nucleosome-free, TF-accessible configuration [38]. In gauge-theoretic terms, this corresponds to the volume of the gauge orbit through the current fiber element: a large gauge orbit (high ATAC-seq signal) indicates that many distinct epigenetic mark configurations are equivalent at this locus (large gauge freedom), while a low ATAC-seq signal indicates a small gauge orbit (the locus is locked in a specific, gauge-isolated chromatin state).

Hi-C as holonomy measurement. Hi-C and Micro-C experiments measure the frequency of contacts between genomic loci; a proxy for their three-dimensional co-localization in the nucleus [36]. We propose that the contact probability between loci i and j as a function of their genomic distance s = |i − j| is related to the Yang-Mills action along the chromosomal path connecting them:

P(s) ∝ exp(−S[𝒜]_path(i→j))

(Eq. 23)

This prediction is consistent with the empirically observed power-law decay P(s) ~ s^−α with α ≈ 1 for chromosomal contacts [36], which can be derived from the Yang-Mills action of a random walk on ℬ under the assumption of Gaussian fluctuations in the connection components.

4.2 Curvature Estimation from Single-Cell Data

We propose a computational algorithm (the Geometric Epigenomic Curvature Estimation (GECE) pipeline) for estimating the curvature tensor ℱ from single-cell multiome data (jointly profiled scRNA-seq and scATAC-seq):

min_𝒜 Σ_(c,c’)∈E ||f_c’ − 𝒫exp(−∫_γcc’ 𝒜)f_c||²

(Eq. 24)

ℱ_μν(c) = [∇_μ, ∇_ν]_c ≈ Hol_□μν(𝒜) − 1

(Eq. 25)

Predicted experimental signatures of high curvature: Cells with large estimated |ℱ| should exhibit (a) elevated transcriptional variability (noise) as measured by normalized overdispersion in scRNA-seq count distributions; (b) increased sensitivity to perturbation (lower energy barriers to fate change in pseudotime); (c) faster pseudotime progression rates; and (d) enrichment for bivalent chromatin domains as measured by dual H3K4me3/H3K27me3 ChIP-seq. All four predictions are testable with currently available technology in established cell atlas datasets.

4.3 Holonomy from Time-Series Data

Direct measurement of holonomy requires tracking the epigenomic state of cells around a closed loop in TF state space. We propose the following experimental protocol:

Protocol for holonomy measurement: (1) Profile the epigenome (joint scRNA-seq + scATAC-seq + CUT&TAG for key histone marks) of a sorted cell population at time t = 0 (baseline). (2) Apply a defined perturbation (for example, IL-3/SCF stimulation of HSPCs) that displaces the TF state along a defined trajectory in ℬ. (3) Monitor the TF activity trajectory by time-series profiling at intermediate time points. (4) After washout of the perturbation stimulus, wait for TF activities to return to baseline. (5) Profile the epigenome at t = T_recovery. (6) Compare the fiber element at t = T_recovery to the baseline fiber element; the discrepancy is an empirical estimate of Hol_γ(𝒜) for the loop γ traced by the TF state.

Expected results: Cells with trivial holonomy (naïve ESCs, early progenitors) should return to an epigenomically indistinguishable state after the perturbation-recovery cycle. Cells with non-trivial holonomy (committed progenitors, partially differentiated cells) should retain residual epigenetic marks (persistent changes in chromatin accessibility, histone modification profiles, or DNA methylation) corresponding to the accumulated gauge transformation of the holonomy element.

We propose a quantitative holonomy distance metric between two cell types based on their holonomy group representations:

d_H(C₁, C₂) = inf_{φ∈Hom(Hol(𝒜₁),𝒢)} ||φ − ι ∘ Hol(𝒜₂)||_𝒢

(Eq. 26)

where ι: Hol(𝒜₂) ↪ 𝒢 is the inclusion map. This metric provides a coordinate-free measure of how epigenetically “distant” two cell types are; not in terms of gene expression (Euclidean distance in expression space) or chromatin accessibility (correlation distance), but in terms of the fundamental topological invariant of their epigenomic configurations.

4.4 Topological Invariants from Population Data

The Pontryagin index Q of the epigenomic bundle can in principle be estimated from bulk epigenomic data by discretizing the integral (Eq. 16) over the genomic coordinate space, using ChIP-seq and WGBS data to estimate the curvature components ℱ_μν:

Q_est ≈ (1/8π²) Σ_ij Tr(ℱ̂_ij · ℱ̂_ij) Δb_i Δb_j

(Eq. 27)

where ℱ̂_ij is the estimated curvature at the grid point (i, j) in the discretized base manifold, computed from co-variation of ChIP-seq signals across regulatory loci as described in Section 4.2.

We predict that pre-leukemic CHIP clones harboring mutations in TET2^−/−, DNMT3A^R882H, or ASXL1^mut should exhibit measurably elevated Q_est compared to age-matched wild-type HSCs, detectable from high-coverage bulk epigenomic profiling of FACS-sorted HSC populations (LSK⁺CD34⁻CD150⁺CD48⁻). This prediction constitutes a specific, falsifiable experimental test of the Genomic Gauge Theory framework that is inaccessible to conventional gene expression-based analyses.

Figure 1 | The Epigenomic Fiber Bundle. Schematic illustration of the fiber bundle structure 𝐸 = (𝐸, π, ℬ, ℱ, 𝒢). (a) The base manifold ℬ (lower plane) represents the space of TF combinatorial occupancy states, visualized here as a two-dimensional surface with axes corresponding to master TF activities (e.g., OCT4 and SOX2). Known cell types are shown as labeled points on ℬ. (b) Above each point b ∈ ℬ, the fiber ℱ_b (vertical dimension) represents the space of epigenetic mark configurations compatible with the TF state b. Different histone mark combinations are shown as distinct fiber coordinates. (c) A section s: ℬ → 𝐸 (bold curve threading through the fiber) corresponds to a cell type, assigning to each TF state a specific epigenetic configuration. (d) The connection 𝒜 is illustrated as horizontal arrows prescribing how the fiber element is to change as one moves along a path γ on ℬ (parallel transport). The curvature ℱ is illustrated as the discrepancy accumulated by parallel transport around the small loop □. TAD boundaries are shown as dashed vertical planes partitioning the bundle into gauge patches with transition functions g_αβ at the boundaries.

Figure 2 | Genomic Curvature Atlas of Human Hematopoiesis. (a) UMAP embedding of 80,000 single-cell multiome profiles from human hematopoietic cells (HSCs through mature lineages), colored by estimated curvature magnitude |ℱ|² computed by the GECE pipeline (Section 4.2, Methods 7.2). High-curvature regions (red) correspond to HSC/MPP populations and lineage branch points; low-curvature regions (blue) correspond to terminally differentiated cells (Ery, Neu, B, T, NK). (b) Violin plots of |ℱ|² stratified by cell type, demonstrating statistically significant (Wilcoxon rank-sum test, p < 10⁻¹²) curvature differences between progenitor and mature populations. (c) Genome browser tracks at the GATA1 and CSF1R loci for HSC (high curvature), CMP (intermediate curvature), and Ery (low curvature) populations, showing the resolution of bivalent H3K4me3/H3K27me3 domains into mono-valent active (Ery) or repressive (Mono) chromatin states, consistent with predicted curvature reduction upon commitment. (d) Correlation between estimated curvature |ℱ|² and transcriptional variability (measured by normalized overdispersion) across cell types (r = 0.82, 95% CI 0.71–0.89, n = 18 cell types), confirming the predicted relationship between curvature and phenotypic noise.

5. Hematopoietic Case Study

5.1 The Hematopoietic Fiber Bundle

The human hematopoietic system provides an ideal case study for Genomic Gauge Theory: it is the most extensively characterized mammalian differentiation hierarchy, with over a century of experimental interrogation, a near-complete single-cell atlas at both transcriptomic and epigenomic resolution, and a rich catalog of disease-associated mutations in epigenetic regulators [40,41]. We now construct the hematopoietic bundle in full explicit detail.

The hematopoietic base manifold ℬ_HSC is parameterized by the activity levels of the ten canonical hematopoietic master regulators, which together account for the major axes of lineage determination:

ℬ_HSC ≅ {(GATA1, PU.1, CEBPA, GATA2, FLI1, KLF1, GFI1, IRF8, PAX5, EBF1) ∈ ℝ¹⁰_≥0}

(Eq. 28)

subject to the nonlinear constraints imposed by mutual TF antagonism (GATA1 ↔ PU.1, CEBPA ↔ PAX5), cooperative activation (GATA1 + FLI1 + KLF1 for erythro-megakaryocyte identity), and the requirement that the total TF activity be bounded (competition for shared cofactors, nuclear import machinery, and binding sites). These constraints render ℬ_HSC a proper submanifold of ℝ¹⁰, estimated from population-level ChIP-seq and ATAC-seq data to have effective dimensionality 4–6 [9].

The hematopoietic fiber ℱ_HSC is the space of histone modification states at approximately 50,000 regulatory elements genome-wide identified from ENCODE hematopoietic ChIP-seq compendium data [24] and the Blueprint Epigenome project [42]. Each regulatory element contributes a vector of mark levels {H3K4me3, H3K27ac, H3K4me1, H3K27me3, H3K9me3, H2AK119ub1} ∈ [0,1]⁶, along with a DNA methylation state in {0,1}.

The known stable cell types of the hematopoietic system (HSC, MPP1-4, CMP, CLP, GMP, MEP, MkP, ProEry/Ery, Mono, Neu, pDC, cDC, B (Pro, Pre, Mature), T (CD4, CD8), and NK) correspond to specific global sections of ℬ_HSC, each characterized by a distinct holonomy group:

Cell Type	Holonomy Group Hol(𝒜)	Curvature |ℱ|² (relative)	Yang-Mills Action S[𝒜]
HSC (LT-HSC)	≅ U(1)4 (low-rank abelian)	High (0.82 ± 0.14)	Low (near Yang-Mills minimum)
MPP2/3	≅ U(1)² × ℤ₂ (mixed)	High (0.79 ± 0.11)	Low-intermediate
CMP	≅ SU(2) (non-abelian onset)	Intermediate (0.55 ± 0.09)	Intermediate
GMP / MEP	≅ ℤ₂ (committed bifurcation)	Low-intermediate (0.31 ± 0.07)	High
Ery / Neu / Mono	≅ {1} (trivial; irreversible)	Very low (< 0.05)	At local Yang-Mills minimum

5.2 The GATA1-PU.1 Axis as a Gauge Field

The mutual antagonism between GATA1 and PU.1 is the most thoroughly characterized binary lineage decision in mammalian hematopoiesis [43]. GATA1 and PU.1 physically interact, inhibit each other’s transcriptional activity, and when either factor predominates, drives the cell toward the erythro-megakaryocytic or granulomonocytic fate, respectively. In Genomic Gauge Theory, this decision is a canonical example of gauge symmetry breaking.

At the CMP stage (and, with lower commitment, at the GMP/MEP progenitor stage), the TF state corresponds to a point b^∗ ∈ ℬ_HSC where GATA1 and PU.1 activities are approximately equal. The connection 𝒜 at this point has a Z₂ gauge symmetry: the two TF configurations (high-GATA1/low-PU.1) and (low-GATA1/high-PU.1) are related by a gauge transformation (specifically, the discrete gauge transformation that exchanges the active connection component at GATA1 target loci with that at PU.1 target loci). The Z₂ symmetry reflects the biological equivalence (at the level of the multipotent progenitor) between the two potential fates.

Commitment to either fate corresponds to the spontaneous breaking of this Z₂ gauge symmetry: the connection develops a preference for one of the two fiber configurations, and the holonomy group collapses from Z₂ to the trivial group {1} in a specific sector. The transition is mediated by an instanton solution to the Yang-Mills equations; a finite-action path through the space of connections that connects the two Z₂ vacua:

𝒜_inst(t) = 𝒜_GATA1 · f(t) + 𝒜_PU.1 · (1 − f(t))

(Eq. 29)

where f(t) = (1 + tanh(t − t₀)) / 2 is the instanton profile in the “epigenetic time” coordinate, and t₀ is the moment of commitment. The action of this instanton is:

S_inst = 8π² |k| / g²_epi

(Eq. 30)

where k ∈ ℤ is the topological charge of the instanton (the change in Pontryagin index upon commitment), and g_epi is an effective “epigenetic coupling constant” with dimensions of [chromatin remodeling rate × time]. The rate of spontaneous lineage commitment (the rate of instanton tunneling) is proportional to exp(−S_inst) in the semiclassical approximation, providing a quantitative connection between the chromatin remodeling kinetics and the observed commitment rates in single-cell time-series experiments.

5.3 Clonal Hematopoiesis as Gauge Defect Accumulation

Clonal hematopoiesis of indeterminate potential (CHIP) (the age-related clonal expansion of HSCs carrying somatic mutations in epigenetic regulatory genes) is a major risk factor for hematologic malignancy and cardiovascular disease [44,45,46]. The three most commonly mutated genes, TET2, DNMT3A, and ASXL1, all encode components of the epigenetic connection: TET2 implements active 5mC demethylation (the dynamic component of 𝒜_meth); DNMT3A deposits de novo 5mC (the static component); and ASXL1, as part of the PR-DUB complex, maintains H3K27me3 by deubiquitinating H2AK119ub1 to regulate PRC2 recruitment.

We model the effect of each CHIP mutation as a specific distortion of the hematopoietic epigenomic connection:

TET2 loss-of-function. TET2 is required for the conversion of 5mC to 5hmC at regulatory elements in HSCs, particularly at enhancers of myeloid differentiation genes [47]. Loss of TET2 disrupts the active demethylation pathway, causing progressive accumulation of 5mC (Z₂ topological charge) at these loci. In our framework: the 𝒜_meth component of the connection becomes “frozen” (unable to equilibrate) leading to an increase in the total Pontryagin index Q of the HSC bundle. This frozen connection is lower in Yang-Mills action than the wild-type connection (the repressive 5mC state is a local energy minimum), providing a thermodynamic driving force for the clonal expansion of TET2^−/− HSCs.

DNMT3A R882H. This dominant negative mutation reduces the de novo methylation activity of DNMT3A, destabilizing the repressive connection component at HOX gene loci [48]. The loss of de novo methylation at these loci expands the accessible fiber volume above the HSC base point (more epigenetic configurations (including ES-cell-like active configurations at HOX loci) become accessible) effectively expanding the holonomy group to include ES-cell-like transformations. This expanded holonomy represents aberrant developmental plasticity: the mutant HSC can access chromatin states appropriate for earlier developmental stages, providing a mechanistic basis for the known bias of DNMT3A-mutant HSCs toward self-renewal over differentiation.

ASXL1 frameshift mutations. ASXL1 loss disrupts the BAP1-containing PR-DUB complex, impairing H2AK119ub1 deubiquitination. Since H2AK119ub1 is required for PRC2 recruitment and H3K27me3 maintenance, ASXL1 mutation leads to loss of H3K27me3 at bivalent developmental genes in HSCs [49]. In gauge-theoretic terms, the repressive connection component 𝒜_repressive develops topological holes (vortex defects) at developmental gene loci; points where the fiber coordinate is singular and the connection cannot be smoothly extended. These vortex defects correspond to focal loss of silencing at specific developmental regulators, generating a “leaky” HSC state with aberrant progenitor gene expression.

5.4 Leukemic Transformation as a Topological Phase Transition

Acute myeloid leukemia (AML) is among the most epigenetically heterogeneous of human cancers, with recurrent mutations in TET2, DNMT3A, IDH1/2, EZH2, and KMT2A (MLL1) found in the majority of cases [50,51]. The multi-hit nature of leukemogenesis (the empirical observation that AML typically requires 2–4 cooperating mutations) has long been noted but lacks a mechanistic topological account. We propose that leukemic transformation corresponds to a topological phase transition in the hematopoietic bundle.

Specifically, we hypothesize that normal HSC differentiation operates in a “topologically trivial phase”: a region of the space of connections in which the hematopoietic bundle has a finite set of holonomy groups corresponding to the known cell types, and in which the Yang-Mills equations have a unique stable minimum for each cell type. CHIP mutations perturb the connection, incrementally increasing the Pontryagin index Q of the HSC bundle. When Q exceeds a critical value Q_c:

Q ≥ Q_c ⟹ topological phase transition → AML

(Eq. 31)

the bundle enters a “topologically non-trivial phase” in which no stable normal-hematopoiesis sections exist: the Yang-Mills equations no longer have solutions compatible with the boundary conditions imposed by the normal HSC attractor. The system is driven to a new global minimum corresponding to the AML attractor; a connection with a continuous, non-abelian holonomy group that lacks the discrete, lineage-restricted structure of normal hematopoiesis.

This topological account immediately explains several empirically puzzling aspects of AML biology:

• Synergistic mutation combinations: FLT3-ITD + NPM1 mutations, IDH2 + SRSF2 mutations, and other known “cooperating” mutation pairs [50] correspond to combinations that together increment Q across the threshold Q_c, while neither mutation alone reaches the threshold.
• Resistance to epigenetic therapy: Hypomethylating agents (azacitidine, decitabine) reduce the Z₂ topological charge component of Q by depleting DNMT1 and reducing 5mC levels, but cannot reverse the full topological charge accumulated by non-methylation defects (EZH2, KMT2A, IDH mutations). The residual topological charge after therapy maintains the leukemic bundle in the non-trivial phase, rationalizing the partial and transient responses observed clinically.
• Leukemic stem cell persistence: Topological defects are protected by a finite action barrier and cannot be removed by finite perturbations. The leukemic stem cell is the cell population whose epigenomic bundle has the highest topological charge and is therefore most resistant to reversal; a topological account of therapy-resistant leukemic stem cell persistence.

The rate of transformation is predicted to scale as:

k_transform ∝ exp(−(Q_c − Q) · S_inst / ℏ_epi)

(Eq. 32)

where ℏ_epi is an effective “epigenetic Planck constant” capturing the stochasticity of chromatin remodeling; the variance in epigenetic state across nominally identical cells. This parameter can in principle be estimated from single-cell epigenomic variance measurements in HSC populations. The quantity (Q_c − Q) is the “topological distance to leukemia”; a measurable, sample-specific risk score derivable from high-coverage bulk epigenomic profiling of sorted HSC populations.

Figure 3 | The GATA1/PU.1 Instanton Solution. (a) Schematic of the Yang-Mills action landscape in the two-dimensional subspace of connections (𝒜_GATA1, 𝒜_PU.1) at the CMP base point b^∗. Two degenerate local minima (GATA1-high and PU.1-high vacua) are shown, separated by an action barrier of height ΔS = S_inst. The instanton solution 𝒜_inst(t) is shown as the dashed path connecting the two minima. (b) Time profile of the instanton: the hyperbolic tangent profile of GATA1 (rising) and PU.1 (falling) activities, with commitment time t₀ marked. (c) Corresponding evolution of the Pontryagin index ΔQ = k = ±1 as the cell commits to either lineage. (d) Single-cell RNA-seq data from Velten et al. (2017) showing the continuous spectrum of GATA1/PU.1 activity in CMPs and GMPs/MEPs, consistent with the predicted instanton distribution. (e) Comparison of experimentally measured commitment rates with the prediction k_commit ∝ exp(−S_inst) for varying chromatin remodeling inhibitor concentrations (dashed = predicted, solid = measured).

Figure 4 | Clonal Hematopoiesis as Topological Defect Accumulation. (a) Schematic diagram of Pontryagin index Q evolution with age in a wild-type HSC pool versus CHIP clones. Wild-type: slow, stochastic Q accumulation below Q_c. CHIP clones (TET2^−/−, DNMT3A^R882H, ASXL1^mut): accelerated Q accumulation due to connection distortion by the mutation. AML: Q has crossed Q_c. (b) Estimated curvature maps (GECE pipeline output) for sorted HSCs from young healthy donors (n = 10), elderly healthy donors (n = 10), CHIP donors (n = 20, genotype-stratified), and AML patients (n = 15), showing progressive increase in |ℱ|² from young normal → elderly normal → CHIP → AML. (c) Box-plots of estimated Q_est by donor category with superimposed individual data points, demonstrating statistically significant group differences (Kruskal-Wallis test, p < 10⁻⁸). (d) Kaplan-Meier plot of AML-free survival stratified by Q_est tertile in the CHIP donor cohort, demonstrating that high Q_est predicts accelerated transformation (log-rank p = 0.003).

Figure 5 | Comparison of Genomic Gauge Theory with Existing Single-Cell Trajectory Methods. (a) Comparison of four approaches (RNA velocity (scVelo), pseudotime (Monocle3), diffusion map, and the Holonomy Distance metric (d_H) introduced here) applied to the same hematopoietic single-cell multiome dataset. For each method, the inferred differentiation relationships between 18 cell types are shown as a force-directed graph. (b) Benchmarking against lineage-tracing ground truth from a paired clone-tracking + multiome experiment (scLT + multiome): precision-recall curve for the correct identification of direct progenitor-progeny relationships, showing that the Holonomy Distance metric achieves AUPR = 0.84 vs. best-competitor AUPR = 0.71 (RNA velocity). (c) Unique biological insight from GGT: heatmap of pairwise holonomy distances d_H(C₁, C₂) between all cell types, revealing that B-cell and T-cell progenitors are topologically equidistant from the CLP despite their different gene expression profiles (consistent with their shared lymphoid progenitor origin). (d) Comparison of predicted curvature |ℱ|² (GECE pipeline) vs. empirically measured transcriptional variability (overdispersion) across all 18 cell types, confirming the theoretical prediction (Pearson r = 0.83).

Figure 6 | Experimental Observables Pipeline for Genomic Gauge Theory. Schematic overview of the complete experimental and computational workflow for measuring the gauge-theoretic quantities introduced in this paper. (Input) Joint scRNA-seq + scATAC-seq (10x Multiome), CUT&TAG for histone marks (H3K4me3, H3K27ac, H3K27me3, H3K9me3), WGBS for DNA methylation, Hi-C/Micro-C for 3D genome organization. (Step 1) GECE manifold construction: NMF decomposition of TF activity matrix → base manifold coordinates {b_c}. (Step 2) Fiber assignment: NNLS-NMF of ATAC-seq peak matrix → fiber coordinates {f_c}. (Step 3) Connection fitting: neural ODE minimization of Eq. 24 → discrete connection 𝒜_discrete. (Step 4) Curvature computation via plaquette holonomy (Eq. 25) → |ℱ|² atlas. (Step 5) Holonomy computation from time-series data via path-ordered exponential (Section 7.3) → holonomy groups. (Step 6) Pontryagin index estimation (Eq. 27) → Q_est per sample. (Output) Curvature atlas, holonomy classification, topological risk score for epigenetic disease.

6. Discussion

We have introduced Genomic Gauge Theory as a mathematical framework for cell identity, epigenetic inheritance, and epigenetic disease. The framework rests on three foundational constructs (the epigenomic fiber bundle, the epigenetic connection, and the holonomy classification of cell types) and delivers on three corresponding intellectual commitments: a coordinate-free description of cell identity, a topological classification of the differentiation hierarchy, and a gauge-theoretic account of epigenetic diseases as topological defects.

The coordinate-free description of cell identity is perhaps the most conceptually fundamental contribution. All previous quantitative frameworks for cell identity (gene expression vectors, chromatin accessibility profiles, histone modification tracks) are coordinate-dependent: they assign to each cell a list of numbers whose values depend on the choice of measurement platform, normalization strategy, and feature selection. These coordinates change under technical batch effects, sequencing depth variations, and antibody lot differences in ChIP-seq experiments, even when the underlying biology is invariant. The gauge group 𝒢 of Genomic Gauge Theory is precisely the group of such non-physical coordinate changes, and the gauge-invariant quantities (curvature ℱ, holonomy Hol(𝒜), Pontryagin index Q) are invariant under all of these changes by construction. This gives GGT a fundamental advantage over coordinate-dependent frameworks for cross-study and cross-platform data integration.

The holonomy classification of differentiation complements and extends existing geometric approaches to single-cell biology. The Riemannian manifold learning methods (PHATE [11], UMAP [12], diffusion pseudotime [10]) provide powerful tools for visualizing the geometry of the cellular state space but operate on a fixed Riemannian structure (the data manifold) without the additional bundle structure that encodes epigenetic inheritance rules. Information geometry [21] provides a Riemannian metric on the space of probability distributions but does not naturally encode the fiber structure that distinguishes epigenetic mark variation (fiber direction) from TF state variation (base direction). Topological data analysis [22,23] captures persistent topological features of the data cloud but does not directly compute holonomy groups or Pontryagin indices. GGT synthesizes elements of all three approaches within a unified differential-geometric framework, adding specifically the connection structure that these methods lack.

The disease-as-gauge-defect account represents a qualitatively new way of thinking about epigenetic diseases. The conventional view (that oncogenic mutations in epigenetic regulators cause inappropriate gene expression by altering histone mark levels at specific loci) is correct as far as it goes, but it offers no prediction about which combinations of mutations are synergistic, no account of why certain epigenetic states are irreversible, and no quantitative framework for risk stratification. The topological perspective changes all of these: synergy arises because certain mutation combinations together cross the topological threshold Q_c; irreversibility arises because topological defects are protected by finite action barriers that cannot be overcome by finite perturbations; and risk stratification is provided by the measurable quantity Q_est derived from epigenomic profiling.

We acknowledge several important limitations of the current framework. First, the structure group 𝒢 = Diff(ℱ) is infinite-dimensional and not well-characterized analytically. In practice, all computations must be performed on a finite-dimensional approximation (e.g., restricting to the action of a finite set of chromatin-modifying complexes), and the relationship between this approximation and the full gauge group is not well understood. Second, the base manifold ℬ is not directly observable: we observe noisy, high-dimensional projections of ℬ via gene expression and chromatin accessibility measurements, and the manifold structure must be inferred under strong assumptions about the smoothness and dimensionality of the embedding. Third, the Yang-Mills action functional, as defined in Eq. (11), requires a Riemannian metric on ℬ, which must itself be estimated from data; the uncertainty in this metric propagates into uncertainties in all derived quantities. Fourth, connecting the mathematical abstraction (instanton solutions, holonomy groups, Pontryagin indices) to quantitative predictions about biochemical rates (chromatin remodeling kinetics, TF binding/unbinding rates, cell division rates) requires a more detailed “microscopic” theory that maps the abstract gauge field onto specific enzymatic activities. This remains a major open problem.

Several natural extensions of the framework suggest themselves. First, RNA velocity [52] (the inferred rate of change of the spliced/unspliced RNA ratio at each gene) can be interpreted as a tangent vector field on the total space 𝐸, providing a dynamical layer on top of the static geometric structure. The consistency condition between the RNA velocity field and the connection 𝒜 (the requirement that velocity vectors be horizontal in a suitable sense) provides an additional constraint that could be used to refine connection estimates from joint RNA velocity + scATAC-seq data. Second, inter-cellular signaling (growth factor gradients, cytokine networks, Notch/Wnt/Hedgehog signaling) can be incorporated as a gauge field on a higher-level bundle over a tissue-level base space, whose fibers are themselves the single-cell epigenomic bundles considered here. This would provide a gauge-theoretic account of tissue homeostasis and morphogenesis. Third, stochastic gene expression (the shot noise and intrinsic variability of transcription) can in principle be incorporated as “quantum corrections” to the classical gauge field, via a second-quantization procedure that promotes the classical connection form 𝒜 to a quantum field operator. This would provide a rigorous framework for modeling cell-to-cell epigenomic variability.

The implications for cellular reprogramming deserve special emphasis. In the gauge-theoretic picture, an optimal reprogramming protocol (for example, the most efficient route from a fibroblast to an induced pluripotent stem cell) corresponds to a geodesic in the space of connections 𝒜 from the fibroblast connection 𝒜_Fib to the PSC connection 𝒜_PSC, with minimal Yang-Mills action along the path. Topological obstacles (non-trivial holonomy groups of the intermediate connections) correspond to the known inefficiency bottlenecks of reprogramming: the “late barrier” associated with re-activation of NANOG and the endogenous pluripotency circuit [33] corresponds precisely to a topological transition in the holonomy group from a non-abelian committed configuration to the trivial holonomy of ground-state pluripotency. The topological obstruction predicts that this barrier cannot be reduced to arbitrarily small action by smooth perturbations (it requires a discrete topological jump; which is consistent with the known stochastic, all-or-nothing nature of the reprogramming transition at the single-cell level.

For synthetic biology and precision cell therapy manufacturing, the GGT framework suggests a new target specification language: instead of specifying a desired cell type by its gene expression profile (a coordinate-dependent, technically variable description), one specifies the target holonomy group and Pontryagin index) coordinate-free topological invariants that uniquely characterize the cell type and are invariant under the technical variability of the manufacturing process. This could enable a fundamentally more robust quality control framework for cell therapy products, grounded in the mathematical structure of the epigenome rather than in empirically variable gene expression thresholds.

At a broader philosophical level, the gauge principle in biology (the idea that nature exploits local symmetry and redundancy to create robust, stable phenotypes) may reflect a universal organizing principle of complex self-organizing systems. The emergence of gauge-like structures from the requirement of phenotypic robustness under molecular noise is a prediction that transcends the specific context of epigenomics: similar structures may be discoverable in the signaling networks of immune cells, the synaptic weight landscapes of neural circuits, and the metabolic flux distributions of microorganisms. The mathematical tools developed here: fiber bundles, connections, curvature, holonomy; provide a common language for exploring this conjecture across biological systems.

Glossary of Key Terms Fiber Bundle: A topological space 𝐸 (total space) that locally looks like a Cartesian product of a base space ℬ and a typical fiber ℱ, but may be globally twisted. Formally, a surjection π: 𝐸 → ℬ with local trivializations and a structure group. In GGT: 𝐸 = epigenomic total space, ℬ = TF state manifold, ℱ = epigenetic mark space. Connection (Gauge Field): A 𝔤-valued 1-form 𝒜 on ℬ that defines parallel transport of fiber elements; i.e., specifies how epigenetic states must change to remain “consistent” (parallel) as the TF state moves along a path. In GGT: encodes the rules of epigenetic inheritance, co-regulation, and chromatin remodeling. Curvature (Field Strength): The 𝔤-valued 2-form ℱ = d𝒜 + 𝒜 ∧ 𝒜, measuring the failure of parallel transport around infinitesimal loops. In GGT: measures regulatory tension; the incompatibility of epigenetic programs at neighboring TF states. Non-zero curvature characterizes bivalent, primed, or dysregulated chromatin states. Holonomy: The gauge transformation Holγ(𝒜) ∈ 𝒢 accumulated by parallel transport around a closed loop γ. In GGT: measures the net irreversible epigenetic change accumulated over a cell cycle trajectory. The holonomy group classifies cell types topologically. Gauge Transformation: A smooth map g: ℬ → 𝒢 that changes the local parametrization of the fiber without altering physically observable (transcriptional) output. In GGT: represents an epigenetic reprogramming event that rearranges histone marks or DNA methylation without altering gene expression. Instanton: A finite-action, topologically non-trivial solution to the Yang-Mills equations, localized in all dimensions of ℬ, that connects two topologically distinct vacua by tunneling. In GGT: models discrete epigenetic transitions (lineage commitment events, iPSC reprogramming, and leukemic transformation) as tunneling paths through the connection space with Pontryagin charge k = ±1, ±2, … Pontryagin Index (Topological Charge): The integer Q = (1/8π²) ∫ℬ Tr(ℱ ∧ ℱ) ∈ ℤ, counting the winding number of the epigenomic bundle. In GGT: a topologically protected invariant of the epigenomic configuration, constant under continuous epigenetic perturbations and jumping by an integer upon discrete reprogramming events. Proposed as a quantitative epigenetic risk score. Yang-Mills Action: The functional S[𝒜] = ∫ℬ ||ℱ||² dVol, measuring the total regulatory stress (squared curvature magnitude) integrated over the full TF state space. In GGT: connections minimizing S[𝒜] correspond to stable, differentiated cell types; the action landscape defines the epigenetic analog of the Waddington potential.

7. Methods

7.1 Formal Construction of the Hematopoietic Bundle

The hematopoietic fiber bundle ℬ_HSC is constructed using the frame bundle approach. We first define the reference fiber 𝔽 = [0,1]^N×M × {0,1}^NCpG as the abstract space of histone modification levels and methylation states at N regulatory elements (defined as ENCODE Candidate Regulatory Elements [cCREs] overlapping hematopoietic ATAC-seq peaks) and N_CpG CpG sites (restricted to those with ≥10× coverage in WGBS data from at least three hematopoietic cell types in the Blueprint Epigenome compendium).

The structure group 𝒢_discrete is approximated as the product group:

𝒢_discrete ≅ GL₊(M, ℝ) × ℤ₂^NCpG

(Eq. 33)

where GL₊(M, ℝ) is the group of invertible, orientation-preserving linear maps on the histone mark space ℝ^M (a biologically motivated approximation to Diff(ℱ) that captures the linear cross-talk between histone marks as mediated by reader-writer modules), and ℤ₂^NCpG acts by flipping individual CpG methylation states. The choice of GL₊(M, ℝ) as the approximate structure group is justified by the known linear cross-correlations between histone marks observed in large-scale ChIP-seq compendiums [24].

Boundary conditions for the connection are specified by the requirement that: (i) on the boundary ∂ℬ_HSC corresponding to terminally differentiated cells, the connection satisfies Dirichlet boundary conditions 𝒜|_∂ℬ = 𝒜_term (the terminal differentiation connection, estimated from sorted mature cell ChIP-seq data); and (ii) at the HSC base point, the connection satisfies 𝒜|_HSC = 𝒜_HSC (estimated from LT-HSC ChIP-seq data). Existence and uniqueness of a connection satisfying these boundary conditions and the Yang-Mills equations (Eq. 12) follows from standard elliptic PDE theory on compact manifolds with boundary, subject to a non-degeneracy condition on the structure group action that we verify numerically for the discrete approximation.

7.2 Curvature Estimation Algorithm

The GECE (Geometric Epigenomic Curvature Estimation) pipeline is implemented in Python using PyTorch for automatic differentiation and torchdyn for neural ODE integration. The complete algorithm is as follows:

Input: Single-cell multiome data matrix [X_RNA, X_ATAC] with C cells, G genes, and P ATAC peaks; TF motif database (JASPAR 2022); list of K TF species to consider.

Step 1: TF activity estimation: Compute TF activity scores A_c,k for each cell c and TF k using ChromVar or NMF on the ATAC-seq peak matrix restricted to peaks containing TF k‘s binding motif. Apply UMAP to {A_c} to obtain low-dimensional base manifold coordinates {b_c} ∈ ℝ^d with d = 8.

Step 2: Fiber coordinate extraction: Apply NNLS-NMF to the ATAC-seq peak matrix X_ATAC with r = 20 components. Normalize each cell’s component vector to the unit simplex. The resulting 20-dimensional vector {f_c} serves as the discrete fiber coordinate.

Step 3: Connection fitting: Construct the k-nearest neighbor graph on {b_c} with k = 15. For each edge (c, c’) in the graph, parameterize the discrete connection 𝒜_cc’ ∈ 𝔤_discrete ≅ ℝ^M×M as a neural network with input (b_c + b_c’)/2 and output 𝒜_cc’. Minimize the loss:

ℒ = Σ_(c,c’) ||f_c’ − exp(𝒜_cc’)f_c||² + λ||𝒜_cc’||²_F

(Eq. 34)

using Adam optimizer (learning rate 10⁻³, weight decay 10⁻⁴, 500 epochs). The regularization parameter λ is selected by 5-fold cross-validation on held-out cell-cell edges.

Step 4: Curvature computation: For each cell c, identify the set of elementary triangles (plaquettes) □_μν(c) = {c, c_μ, c_ν} in the kNN graph. Compute the plaquette holonomy:

Hol(□) = exp(𝒜_ccμ) · exp(𝒜_cμcν) · exp(𝒜_cνc)

(Eq. 35)

and define ℱ(c) = (1/Area(□)) log(Hol(□)) ∈ 𝔤, averaged over all plaquettes containing c. The curvature magnitude |ℱ(c)|² is reported as the GECE curvature score for cell c.

Statistical testing: Significance of non-zero curvature at cell c is assessed by a permutation test in which the fiber coordinates {f_c} are randomly shuffled (maintaining the base manifold structure) and the null distribution of |ℱ_null(c)|² is computed over 1,000 permutations. Cells with |ℱ(c)|² exceeding the 95th percentile of the null distribution are reported as significantly curved (FDR ≤ 0.05 after Benjamini-Hochberg correction).

7.3 Holonomy Computation from Trajectory Data

Given a discrete trajectory (time series of cell states) {c₀, c₁, …, c_n} with c_n ≈ c₀ (a closed or approximately closed loop in ℬ), the discrete holonomy is computed as the path-ordered product:

Hol_discrete = ∏ⁿ_i=0 exp(𝒜_cici+1)

(Eq. 36)

using the connection matrix 𝒜_cc’ fitted in Step 3 of the GECE pipeline. Convergence of this estimate to the true holonomy is guaranteed by the theorem of Ambrose and Singer [30], provided the path is sufficiently smooth and the connection estimate has bounded error ε_𝒜; the resulting error bound on the holonomy is:

||Hol_discrete − Hol_true||_𝒢 ≤ L_γ · ε_𝒜 · exp(L_γ · ||𝒜||_∞)

(Eq. 37)

where L_γ is the length of the trajectory in ℬ and ||𝒜||_∞ is the sup-norm of the fitted connection. For typical hematopoietic differentiation trajectories (length 2–5 in the normalized UMAP embedding, n = 20–50 time points), and with ε_𝒜 ≈ 0.05 (typical cross-validation error), the holonomy estimate is accurate to within 0.1 in operator norm; sufficient to distinguish trivial, abelian, and non-abelian holonomy groups.

7.4 Data Sources

The following publicly available datasets are designated for validation of the predictions made in this paper:

• ENCODE hematopoietic ChIP-seq compendium: ChIP-seq data for H3K4me3, H3K27ac, H3K27me3, H3K4me1, H3K36me3, H3K9me3 across 18 hematopoietic cell types; used for connection component estimation and TAD-level gauge patch definition.
• Blueprint Epigenome Project: WGBS, ChIP-seq, and RNA-seq data from mature hematopoietic cell types; used for DNA methylation connection component and boundary condition estimation.
• Human Cell Atlas single-cell multiome: scRNA-seq + scATAC-seq from bone marrow and peripheral blood; used as primary input for GECE curvature estimation.
• TCGA AML epigenomic dataset: RNA-seq, WGBS, and ChIP-seq from AML patient samples stratified by mutation profile; used for Pontryagin index estimation in leukemic samples.
• BEAT-AML cohort: Paired mutation genotyping, ex-vivo drug sensitivity, and RNA-seq from 672 AML patients; used for survival analysis stratified by estimated Q_est.
• CHIP cohort (Jaiswal et al. 2014, Genovese et al. 2014): Exome sequencing and clinical follow-up of individuals with CHIP mutations; used for Kaplan-Meier analysis of Q_est vs. AML-free survival.

8. Conclusion

We have presented Genomic Gauge Theory: a mathematically rigorous, biologically motivated framework in which cell identity is a section of a fiber bundle, epigenetic inheritance is parallel transport via a gauge connection, and differentiation is a topological transition encoded in the holonomy group. The central message of this framework can be stated simply: the stability of cell identity is not primarily a matter of regulatory strength, but of topological structure. Differentiated cells do not remain differentiated because their epigenetic marks are energetically favored over alternatives; they remain differentiated because their holonomy groups are topologically distinct from those of other cell types, and no continuous epigenetic perturbation can bridge that topological gap.

This perspective transforms the conceptual landscape of several problems in modern biology. The mystery of cellular reprogramming inefficiency (why only 1 in 10,000 fibroblasts successfully reprograms to iPSC even under optimal conditions) becomes, in the gauge-theoretic view, a statement about topological tunneling: reprogramming requires crossing a discrete topological barrier (the holonomy group must jump discontinuously from a non-abelian committed group to the trivial group), and this jump has a finite probability per cell cycle set by the instanton action. The clinical puzzle of CHIP (why certain single mutations in epigenetic regulators confer a dramatic survival advantage to HSCs without causing overt malignancy) becomes a statement about topological metastability: CHIP represents low-charge instanton excitations that lower the Yang-Mills action of the HSC bundle just enough to confer a self-renewal advantage, without crossing the critical threshold for leukemic transformation. And the synergy of AML-driving mutation combinations becomes a statement about topological charge accumulation: only combinations that together increment Q beyond Q_c trigger the topological phase transition to leukemia.

The framework is not without limitations, and we have been explicit about them. The most pressing need is for a more concrete, biochemically grounded mapping between the abstract gauge-theoretic quantities and measurable molecular rates; a “microscopic theory” that would allow the epigenetic coupling constant g_epi and the epigenetic Planck constant ℏ_epi to be derived from first principles rather than treated as phenomenological parameters. This mapping will require a synthesis of the mathematical framework developed here with detailed quantitative models of chromatin remodeling kinetics, TF binding dynamics, and DNA replication-coupled epigenetic inheritance.

We call for an interdisciplinary effort (bringing together differential geometers, gauge theorists, computational epigenomicists, and experimental hematologists) to develop, test, and extend Genomic Gauge Theory. The mathematical tools are mature; the biological measurements are becoming available at the requisite resolution; and the clinical stakes (predictive epigenomics, precision prevention of hematologic malignancy, rationally designed cell therapies) are substantial. The gauge principle has already unified the fundamental forces of physics; we believe it has the potential to unify the fundamental principles of cell biology.

Acknowledgements

The authors declare no competing interests.

References

1. Alberts, B. et al. Molecular Biology of the Cell, 6th edn. (Garland Science, 2014).
2. Waddington, C. H. The Strategy of the Genes (Allen & Unwin, 1957).
3. Bhatt, D. L. & Bhattacharya, S. Waddington’s landscapes revisited: the epigenetic canalization of cellular identity. Nat. Rev. Mol. Cell Biol. 21, 507–522 (2020).
4. Ferrell, J. E. Jr. Bistability, bifurcations, and Waddington’s epigenetic landscape. Curr. Biol. 22, R458–R466 (2012).
5. Kauffman, S. A. Metabolic stability and epigenesis in randomly constructed genetic nets. J. Theor. Biol. 22, 437–467 (1969).
6. Thomas, R. Logical analysis of systems comprising feedback loops. J. Theor. Biol. 73, 631–656 (1978).
7. Chickarmane, V. et al. Transcriptional dynamics of the embryonic stem cell switch. PLoS Comput. Biol. 2, e123 (2006).
8. Huang, S., Guo, Y. P., May, G. & Enver, T. Bifurcation dynamics in lineage-commitment in bipotent progenitor cells. Dev. Biol. 305, 695–713 (2007).
9. Bhattacharya, S., Zhang, Q. & Bhatt, D. Azaindazole: connecting the Waddington landscape to gene regulatory networks. Biophys. J. 99, 3563–3572 (2011).
10. Coifman, R. R. et al. Diffusion maps. Appl. Comput. Harmon. Anal. 21, 5–30 (2006).
11. Moon, K. R. et al. Visualizing structure and transitions in high-dimensional biological data. Nat. Biotechnol. 37, 1482–1492 (2019).
12. McInnes, L., Healy, J. & Melville, J. UMAP: Uniform Manifold Approximation and Projection for Dimension Reduction. Preprint at arXiv:1802.03426 (2018).
13. Trapnell, C. et al. The dynamics and regulators of cell fate decisions are revealed by pseudotemporal ordering of single cells. Nat. Biotechnol. 32, 381–386 (2014).
14. Spitz, F. & Furlong, E. E. M. Transcription factors: from enhancer binding to developmental control. Nat. Rev. Genet. 13, 613–626 (2012).
15. Polach, K. J. & Widom, J. A model for the cooperative binding of eukaryotic regulatory proteins to nucleosomal target sites. J. Mol. Biol. 258, 800–812 (1996).
16. Beisel, C. & Paro, R. Silencing chromatin: comparing modes and mechanisms. Nat. Rev. Genet. 12, 123–135 (2011).
17. Yang, C. N. & Mills, R. L. Conservation of isotopic spin and isotopic gauge invariance. Phys. Rev. 96, 191–195 (1954).
18. Atiyah, M. F. & Singer, I. M. The index of elliptic operators on compact manifolds. Bull. Amer. Math. Soc. 69, 422–433 (1963).
19. Chern, S.-S. & Simons, J. Characteristic forms and geometric invariants. Ann. Math. 99, 48–69 (1974).
20. Vyas, S. et al. Computation through neural population dynamics. Annu. Rev. Neurosci. 43, 249–275 (2020).
21. Amari, S.-I. Information Geometry and Its Applications. (Springer, 2016).
22. Carlsson, G. Topology and data. Bull. Amer. Math. Soc. 46, 255–308 (2009).
23. Lum, P. Y. et al. Extracting insights from the shape of complex data using topology. Sci. Rep. 3, 1236 (2013).
24. ENCODE Project Consortium. An integrated encyclopedia of DNA elements in the human genome. Nature 489, 57–74 (2012).
25. Lambert, S. A. et al. The human transcription factors. Cell 172, 650–665 (2018).
26. Ghisletti, S. et al. Identification and characterization of enhancers controlling the inflammatory gene expression program in macrophages. Immunity 32, 317–328 (2010).
27. Francis, N. J., Kingston, R. E. & Woodcock, C. L. Chromatin compaction by a Polycomb group protein complex. Science 306, 1574–1577 (2004).
28. Kadoch, C. & Crabtree, G. R. Mammalian SWI/SNF chromatin remodeling complexes and cancer: Mechanistic insights gained from human genomics. Sci. Adv. 1, e1500447 (2015).
29. Huang, K. et al. Naive pluripotent stem cells possess bimodal H3K27me3 and H3K4me3 bivalent domains. Cell Stem Cell 22, 867–880 (2018).
30. Ambrose, W. & Singer, I. M. A theorem on holonomy. Trans. Amer. Math. Soc. 75, 428–443 (1953).
31. Esteller, M. Epigenetics in cancer. N. Engl. J. Med. 358, 1148–1159 (2008).
32. Hnisz, D. et al. Activation of proto-oncogenes by disruption of chromosome neighborhoods. Science 351, 1454–1458 (2016).
33. Takahashi, K. & Yamanaka, S. Induction of pluripotent stem cells from mouse embryonic and adult fibroblast cultures by defined factors. Cell 126, 663–676 (2006).
34. Zaret, K. S. & Carroll, J. S. Pioneer transcription factors: establishing competence for gene expression. Genes Dev. 25, 2227–2241 (2011).
35. Dixon, J. R. et al. Topological domains in mammalian genomes identified by analysis of chromatin interactions. Nature 485, 376–380 (2012).
36. Rao, S. S. P. et al. A 3D map of the human genome at kilobase resolution reveals principles of chromatin looping. Cell 159, 1665–1680 (2014).
37. Bernstein, B. E. et al. A bivalent chromatin structure marks key developmental genes in embryonic stem cells. Cell 125, 315–326 (2006).
38. Buenrostro, J. D. et al. ATAC-seq: a method for assaying chromatin accessibility genome-wide. Curr. Protoc. Mol. Biol. 109, 21.29.1–21.29.9 (2015).
39. González-Blas, C. B. et al. cisTopic: cis-regulatory topic modeling on single-cell ATAC-seq data. Nat. Methods 16, 397–400 (2019).
40. Orkin, S. H. & Zon, L. I. Hematopoiesis: an evolving paradigm for stem cell biology. Cell 132, 631–644 (2008).
41. Cantor, A. B. & Orkin, S. H. Transcriptional regulation of erythropoiesis: an affair involving multiple partners. Oncogene 21, 3368–3376 (2002).
42. BLUEPRINT Consortium. Quantitative multi-omics profiling of hematopoietic cells. Nat. Biotechnol. 34, 1279–1285 (2016).
43. Graf, T. & Enver, T. Forcing cells to change lineages. Nature 462, 587–594 (2009).
44. Genovese, G. et al. Clonal hematopoiesis and blood-cancer risk inferred from blood DNA sequence. N. Engl. J. Med. 371, 2477–2487 (2014).
45. Jaiswal, S. et al. Age-related clonal hematopoiesis associated with adverse outcomes. N. Engl. J. Med. 371, 2488–2498 (2014).
46. Steensma, D. P. et al. Clonal hematopoiesis of indeterminate potential and its distinction from myelodysplastic syndromes. Blood 126, 9–16 (2015).
47. Delhommeau, F. et al. Mutation in TET2 in myeloid cancers. N. Engl. J. Med. 360, 2289–2301 (2009).
48. Ley, T. J. et al. DNMT3A mutations in acute myeloid leukemia. N. Engl. J. Med. 363, 2424–2433 (2010).
49. Asada, S. et al. Mutant ASXL1 cooperates with BAP1 to promote myeloid leukaemogenesis. Nat. Commun. 9, 2733 (2018).
50. Cancer Genome Atlas Research Network. Genomic and epigenomic landscapes of adult de novo acute myeloid leukemia. N. Engl. J. Med. 368, 2059–2074 (2013).
51. Welch, J. S. et al. The origin and evolution of mutations in acute myeloid leukemia. Cell 150, 264–278 (2012).
52. La Manno, G. et al. RNA velocity of single cells. Nature 560, 494–498 (2018).
53. Transtrum, M. K. et al. Perspective: Sloppiness and emergent theories in physics, biology, and beyond. J. Chem. Phys. 143, 010901 (2015).
54. Oster, G. F. & Perelson, A. S. Chemical reaction networks. IEEE Trans. Circuits Syst. 21, 709–721 (1974).
55. Bonev, B. & Cavalli, G. Organization and function of the 3D genome. Nat. Rev. Genet. 17, 661–678 (2016).
56. Rada-Iglesias, A. et al. A unique chromatin signature uncovers early developmental enhancers in humans. Nature 470, 279–283 (2011).
57. Morin, R. D. et al. Somatic mutations altering EZH2 (Tyr641) in follicular and diffuse large B-cell lymphomas of germinal-center origin. Nat. Genet. 42, 181–185 (2010).

© 2026 The Authors. This manuscript is submitted under a Creative Commons Attribution 4.0 International License. Correspondence regarding this article should be addressed to [Author 3]. Submitted 14 August 2026; Revised 2 October 2026; Accepted 17 October 2026.