Stem Systems: Control, Dynamics, and Models

• Stem systems are multidisciplinary frameworks combining network theory, embryonic stem-cell models, and stochastic mechanics to describe tissue self-renewal and differentiation.
• They integrate experimental reconstructions and computational models to quantify feedback loops, geometric constraints, and morphogenetic signals in developmental and regenerative contexts.
• Their dynamic nature emphasizes emergent stemness driven by reversible interactions, spatial organization, and non-genetic cues rather than static marker sets.

Searching arXiv for the cited stem-systems literature and closely related papers to ground the article in current research. arXiv search query: "stem cell systems developmental control networks embryonic stem cells morphogenesis niche stochastic competition volume exclusion stem cell differentiation" Stem systems are formal and experimental frameworks for analyzing how stem-cell populations self-renew, differentiate, organize tissues, and maintain homeostasis across developmental, regenerative, and pathological contexts. In the literature surveyed here, the term spans several partially overlapping usages: embryonic stem cell model systems that reconstruct vertebrate pattern formation in vitro; genomic or developmental control networks that specify stem-cell lineage dynamics; stochastic and mechanochemical population models for niche competition, crowding, and lineage survival; and computational morphogenetic systems in which stem-cell-like division of labour underpins reproducible shape generation. Across these usages, a common theme is that stemness is not reducible to a static marker set. Rather, it is specified by dynamical control structure, spatial context, interaction topology, and feedbacks that couple proliferation, differentiation, signaling, mechanics, and geometry (Siggia, 2018, Werner, 2016, Corominas-Murtra et al., 2020, Devlin et al., 25 Mar 2025).

1. Conceptual scope and definitions

A first, network-theoretic definition treats stem systems as genomic-encoded developmental control networks governing normal tissue renewal and, when dysregulated, cancer stem cell behavior (Werner, 2016). In this formulation, a stem cell network is a directed control graph whose nodes represent discrete cell states and whose edges represent developmental transitions or divisions. Self-renewal corresponds to loop structure, while lineage production corresponds to non-looping outgoing transitions. This framework distinguishes first-order, second-order, and higher-order stem-cell networks by loop order, with growth laws linked to combinatorial structure (Werner, 2016).

A second, developmental-systems definition treats stem systems as experimentally tractable multicellular assemblies built from embryonic stem cells (ESCs). ESC colonies in defined geometries—2D micropatterns, 3D cysts, and organoids—recapitulate germ-layer specification, axis formation, and symmetry breaking while remaining genetically accessible (Siggia, 2018). In this sense, a stem system is a reduced but informative reconstruction of vertebrate self-organization.

A third definition emphasizes stochastic population dynamics in niches and tissues. Here, stem systems are modeled as branching, birth–death, diffusion, or competition processes that capture clone survival, crowding feedback, dedifferentiation, or lineage extinction under demographic noise (García-Tejera et al., 2022, Corominas-Murtra et al., 2020, Yu et al., 12 Jan 2026, Nguyen et al., 2022). In these models, stemness may be an emergent positional or dynamical property rather than a fixed intrinsic identity.

A fourth usage appears in computational morphogenesis, where evolved organisms develop stem-cell systems characterized by upstream strongly connected components (SCCs) that are motile and proliferative, and downstream SCCs that are static and non-dividing (Devlin et al., 25 Mar 2025). This suggests that a stem system can also be understood as an irreversible division-of-labour architecture that stabilizes shape generation under noise.

These definitions are not identical, but they converge on a shared systems-level picture: stem-cell behavior depends on network loops, signaling logic, physical constraints, and spatially localized feedbacks rather than on single-gene descriptors alone.

2. Embryonic stem-cell model systems and vertebrate self-organization

ESC-based stem systems are central to quantitative developmental biology because they reconstruct core embryological events in experimentally simplified settings. ESC colonies grown in defined geometries reproduce germ-layer specification, axis formation, and symmetry breaking, making them suitable for dissecting the multiple layers of regulation that render embryonic development robust (Siggia, 2018).

In 2D circular micropatterns of diameter $0.5$–$1$ mm, a uniform BMP4 pulse elicits four concentric territories—extra-embryonic, endoderm, mesoderm, and ectoderm—whose relative widths depend only on colony size, not total cell number (Siggia, 2018). This is a precise demonstration that global cues can be transformed into reproducible spatial fate maps by intercellular circuits and feedbacks intrinsic to the colony. In 3D ESC cysts formed from single ESCs embedded in extracellular matrix, very low BMP triggers a true symmetry-breaking event, with one side acquiring primitive streak markers and the opposite side anterior-like identity without externally imposed asymmetry (Siggia, 2018).

The signaling architecture organizing these systems includes BMP, Nodal/Activin, Wnt, and FGF/MAPK, treated as morphogens. In Xenopus animal-cap assays, a $10$–$20\times$ ligand concentration range produces discrete, dose-dependent mesodermal fates (Siggia, 2018). Siggia’s review further highlights an adaptive interpretation model, by analogy to bacterial chemotaxis, in which cells respond to the time derivative of morphogen level rather than only to steady concentration, allowing positional inference as tissues move and secrete inhibitors (Siggia, 2018). Secreted antagonists such as Chordin sharpen borders, and juxtacrine or adhesion-based interactions enforce local coherence.

Several additional physical motifs recur. In the blastocyst, the TE versus ICM decision uses apical–basal polarity, oriented divisions, and mutual transcriptional inhibition, after which ICM splits into epiblast and primitive endoderm through FGF4-mediated signaling, cell sorting analogous to liquid-driven phase separation via differential surface tensions, and cell-competition-driven elimination of minority states (Siggia, 2018). Gastrulation movements generate forces that fold epithelia and provide positional cues, coupling morphogenesis and patterning.

These ESC systems are not presented as replacements for the embryo. Rather, they are benchmarked against embryo data, including micropattern fate maps matched to time-and-space co-staining in mouse gastrula, so that in vitro models remain tied to in vivo axis organization (Siggia, 2018). A plausible implication is that ESC systems are most informative when used as controlled reconstructions of conserved signaling, mechanical, and lineage circuits rather than as isolated cell-culture assays.

3. Control architectures: networks, bifurcations, and lineage semantics

Network-theoretic stem-system accounts center on the proposition that developmental dynamics are specified by control topology. In Werner’s formulation, developmental control networks, or “cenes,” are genome-encoded graphs whose active nodes and edges determine cell division, differentiation, and signaling behavior, while the organism-scale developmental network is the “cenome” (Werner, 2016). A stem cell network is a special class of cene containing one or more self-looping control states that encode self-renewal.

This framework classifies deterministic geometric networks by loop order. A first-order network produces one self-renewing daughter and one differentiating daughter; a second-order network produces a first-order stem cell lineage while also renewing the higher-order cell; and a $k$-th-order network contains $k$ nested loops (Werner, 2016). The total number of cells after $n$ ideal synchronous divisions is given by

$\mathrm{Cells}(n,k)=\sum_{i=0}^{k}\binom{n}{i},$

which connects stem-cell proliferation to Pascal’s Triangle and discrete figurate-number growth laws (Werner, 2016). This is not merely a descriptive analogy: the stated claim is that loop topology determines proliferative semantics.

The same literature argues that cancer stem cell networks can be topologically indistinguishable from normal stem-cell networks, with the crucial difference lying in spatial–temporal locality of activation, downstream signaling context, and non-network mutations that enable invasion and metastasis (Werner, 2016). A stem network activated ectopically in neural tissue may yield a glioblastoma-like phenotype, whereas the same topology in hematopoietic marrow yields leukemia (Werner, 2016). Metastatic potential is attributed not primarily to network architecture but to additional changes in adhesion, proteolysis, chemokine response, or anchorage independence (Werner, 2016).

Minimal dynamical-systems models reach related conclusions by a different route. Goto and Kaneko studied all two-gene expression networks with diffusive coupling of one protein and identified both symmetric differentiation via a Turing-type mechanism and asymmetric differentiation via a saddle-node bifurcation on an invariant circle (SNIC) (Goto et al., 2013). In the asymmetric case, cells with oscillatory intracellular dynamics function as stem-cell-like states that can both proliferate and differentiate, whereas differentiated cells correspond to fixed-point-type behavior (Goto et al., 2013). Importantly, the ratio of cell types is robust because the effective bifurcation parameter is determined self-consistently through cell–cell interaction.

Andrecut’s multi-dimensional switch-like model likewise links stem-cell fate decisions to bifurcation structure. In the perfectly symmetric case, the model exhibits a degenerate bifurcation with a critical hyperplane of infinitely many critical steady states, interpreted as multi-lineage priming states of the progenitor (Andrecut, 2013). Symmetry breaking in parameter space converts this continuum into discrete lineage-committed attractors. This provides a specific mathematical account of how metastable priming and all-or-none fate choice can arise within a single switch-like regulatory architecture.

Taken together, these works establish a central stem-systems principle: stemness is encoded in dynamical semantics—looping, reversibility, attractor structure, and bifurcation accessibility—rather than solely in snapshot expression states.

4. Stochastic population dynamics, niche access, and homeostasis

A major branch of stem-systems research models stem cells as stochastic populations embedded in spatial niches. Corominas-Murtra and colleagues proposed a stochastic conveyor-belt model in which symmetric divisions advect cells away from a niche while random rearrangements return some cells toward it (Corominas-Murtra et al., 2020). In a one-dimensional setting, lineage density $\rho_n(z,t)$ follows a reaction–advection–diffusion equation,

$$\frac{\partial \rho_n}{\partial t} = -\frac{\partial}{\partial z}\bigl(v(z)\rho_n\bigr) +\frac{D}{2}\frac{\partial^2 \rho_n}{\partial z^2} +\lambda \rho_n,$$

with $1$0 in the simplest geometry (Corominas-Murtra et al., 2020). The long-term clonal survival probability decays as a Gaussian with distance from the niche,

$1$1

which the authors describe as “universal” under broad conditions (Corominas-Murtra et al., 2020). Functional stem-cell number is thus set by the ratio $1$2, not by preassigned intrinsic stem-cell labels.

Crowding-based feedback yields a different but complementary homeostatic mechanism. The birth–death process with volume exclusion (vBD) introduces a carrying capacity $1$3 through the effective birth propensity

$1$4

leading in mean field to

$1$5

and nontrivial fixed point

$1$6

for $1$7 (García-Tejera et al., 2022). At the master-equation level, however, extinction remains possible, and the system exhibits subcritical, weakly supercritical, strongly supercritical, and transient bimodal regimes (García-Tejera et al., 2022). In the strongly supercritical case, the quasi-steady-state clone-size distribution is approximately Gaussian around $1$8, while mean extinction time scales exponentially as

$1$9

in a WKB approximation (García-Tejera et al., 2022). This makes explicit that homeostasis in small niches is metastable rather than absolute.

Related branching-process models study two-type systems of stem cells and irreversibly altered descendants under different criticality assumptions (Nguyen et al., 2022). These yield closed-form probability generating functions and explicit extinction asymptotics in bi-critical, subcritical, and supercritical regimes. The resulting picture is that lineage persistence depends sensitively on whether upstream stem-like cells, downstream progenitors, or both operate near criticality (Nguyen et al., 2022).

A more recent density-dependent CTMC framework adds dedifferentiation explicitly. The five-channel model includes symmetric self-renewal, symmetric differentiation, asymmetric division, dedifferentiation, and terminally differentiated-cell death (Yu et al., 12 Jan 2026). In the diffusion approximation, removal of the dedifferentiation flux produces a sharp dichotomy: under uniformly negative net stem drift the stem coordinate becomes extinct asymptotically almost surely, whereas under uniformly positive drift polynomial moments diverge exponentially (Yu et al., 12 Jan 2026). The derived steady-state “ratio law”

$10$0

and “equalization law”

$10$1

formalize the idea that dedifferentiation can balance fate bias and stabilize homeostasis (Yu et al., 12 Jan 2026).

These models collectively oppose a purely deterministic view of stem-cell maintenance. They instead imply that survival, functional identity, and steady state emerge from the interaction of noise, drift, crowding, and lineage reversibility.

5. Mechanics, geometry, and effective interactions

Stem systems are also physical systems. Mechanical constraints, tissue geometry, and active material properties can determine stem-cell distribution even when molecular identities are held fixed.

Krämer and colleagues modeled epithelial tissues containing stem cells (SC), transient-amplifying cells (TA), and terminally differentiated cells (TD), with SCs dividing asymmetrically, TA cells dividing symmetrically for a fixed number of generations, and TA/TD cells removed stochastically (Krämer et al., 2023). Each biological cell is represented by a pair of particles subject to growth forces, short-range repulsion, adhesion, drag, and noise. In overdamped form,

$10$2

or equivalently

$10$3

with Gaussian white noise (Krämer et al., 2023).

A key result is that stem cells become effectively separated by a soft repulsion generated by the outflow of proliferating daughter cells. The pair correlation $10$4 in the full tissue can be fit by an equilibrium-like soft exponential potential,

$10$5

with interaction length $10$6, where

$10$7

is set by the spatial extent of the progeny halo around one stem cell (Krämer et al., 2023). Structurally, SC organization in tissue is reproduced by Brownian colloids with this effective repulsion, although the actual SC trajectories remain active due to asymmetric pushes from progeny. This offers a mechanistic explanation for observed stem-cell spacing in epithelia and suggests a similar spatial principle for cancer stem cells.

At the developmental scale, Siggia’s ESC review emphasizes mechanochemical coupling and scaling laws. A tissue deformation field $10$8 is modeled by

$10$9

where $20\times$0 is active stress generated by cells in state $20\times$1 (Siggia, 2018). Reaction–diffusion patterning with a self-activator $20\times$2 and longer-range inhibitor $20\times$3 can generate wavelength

$20\times$4

when $20\times$5 (Siggia, 2018). Dimensionless length scales such as

$20\times$6

predict whether a secreted inhibitor leaks at the edge of a colony and therefore whether signaling is confined to the periphery (Siggia, 2018). Xenopus embryos of different size establish BMP gradients that rescale so that pattern boundaries track fractional rather than absolute distance, providing an instance of robust morphogen scaling (Siggia, 2018).

These results support a general physical interpretation of stem systems. Positional information, fate segregation, and stem-cell spacing may arise from phase-separation-like sorting, crowding-limited kinetics, active-stress generation, or effective interactions induced by progeny flow. This suggests that “niche” is often a geometric and mechanical construct in addition to a molecular one.

6. Morphogenetic reproducibility and stem-cell division of labour

A distinct but increasingly important line of work studies how stem-cell systems contribute to reproducible morphogenesis. Devlin and colleagues used an in silico evolution–development platform on a $20\times$7 pixel 2D grid with Cellular Potts Model mechanics, GRNs containing 9 transcription factors including three diffusible morphogens, 15 adhesion proteins, and two contractile proteins (Devlin et al., 25 Mar 2025). Cells grow and divide by size-threshold rules, and state changes are inferred from Booleanized protein-expression vectors.

The central result is that reproducible complex morphologies evolve together with stem-cell systems, despite no direct selection for reproducibility (Devlin et al., 25 Mar 2025). By constructing cell-state graphs from observed state transitions and identifying SCCs, the study distinguishes poorly reproducible organisms with exactly one SCC from highly reproducible organisms with at least two SCCs and a unidirectional link from an upstream SCC to a downstream SCC (Devlin et al., 25 Mar 2025). Upstream SCCs divide $20\times$8 more frequently and have $20\times$9 higher mean momentum than downstream SCCs (Devlin et al., 25 Mar 2025).

Morphogenetic reproducibility is quantified by a score $k$0 derived from pairwise Jaccard similarity between final shapes under rotation, reflection, and translation invariance. Across 90 complex shapes, the distribution of $k$1 is bimodal with coefficient $k$2; poorly reproducible organisms have mean $k$3, whereas highly reproducible organisms have mean $k$4 (Devlin et al., 25 Mar 2025). Under selection for shape complexity alone, $k$5 complex-shape organisms evolved stem-cell systems with irreversible transitions and high reproducibility; under combined selection for shape complexity and directional center-of-mass shift, $k$6 surpassed the threshold, all 29 evolved stem-cell systems, and all but one were highly reproducible (Devlin et al., 25 Mar 2025).

The proposed mechanism is a division of labour: moving, dividing stem-cell types form a leading cap that shapes morphology, while differentiated types form a static stalk that anchors and preserves shape (Devlin et al., 25 Mar 2025). Irreversible differentiation at the stem–differentiated boundary, guided by local morphogen gradients and differential adhesion energies satisfying

$k$7

continuously replenishes new adhesive contacts and drives coherent advance (Devlin et al., 25 Mar 2025).

This work does not claim direct equivalence to any one biological organ. Its stronger claim is more abstract: stem-cell systems are not only devices for producing specialized cell types, but can also be fundamental morphogenetic strategies for achieving reproducibility in noisy developmental processes (Devlin et al., 25 Mar 2025). A plausible implication is that irreversible progenitor-to-differentiated architecture should be viewed as a structural regulator of form, not merely of lineage.

7. Pathology, inference, and adjacent meanings of “stem systems”

Cancer-oriented stem-system theories extend the network perspective into pathology. Werner’s “Stem Cells: The Good, the Bad and the Ugly” argues that the distinction between normal and malignant stem systems lies primarily in spatial–temporal locality of network activation and in non-network mutations that confer invasive phenotypes, not in gross differences of network topology (Werner, 2016). In a stochastic linear stem-cell network with progenitor dedifferentiation probability $k$8, the mean-field equations

$k$9

yield

$k$0

so that $k$1 gives ideal linear output while $k$2 drives exponential expansion (Werner, 2016). The corresponding therapeutic thesis is that highest-order stem cells must be targeted first, since lower-order progenitors can be replenished from upstream reservoirs (Werner, 2016).

Methodologically, statistical and mathematical modeling of hematopoietic stem and progenitor cells provides a broader inferential workflow for stem systems (Back et al., 2018). Imaging-derived positions, trajectories, shapes, and lineages are summarized by point-pattern intensity, Ripley’s $k$3-function, pair correlation $k$4, MSD, autocorrelation, shape descriptors, and genealogy statistics, then embedded into center-based or Cellular Potts models for mechanistic testing (Back et al., 2018). This workflow formalizes a closed loop from image to statistics to model to validation, and is explicitly intended for normal and aberrant hematopoiesis.

The phrase “stem systems” also has adjacent meanings outside stem-cell biology. In plant bio-hybrid control, a technical report on shaping natural bean plants models plant stem stiffening and motion dynamics with an LSTM forward model and evolves controllers to direct growth around obstacles (Wahby et al., 2018). Here the “stem” is botanical rather than stem-cell related, and the system consists of a plant–robot hybrid governed by a learned discrete-time map $k$5 (Wahby et al., 2018). Likewise, in materials characterization, STEM denotes scanning transmission electron microscopy; Pelz and coworkers describe partitioned PRISM algorithms for STEM imaging and 4D-STEM diffraction simulation (Pelz et al., 2021). These usages are terminologically adjacent but conceptually separate from biological stem systems.

The coexistence of these meanings creates a recurrent ambiguity. In contemporary arXiv usage, “stem systems” may denote stem-cell developmental systems, institutional STEM ecosystems, plant stem control systems, or scanning transmission electron microscopy systems. Within biological research, however, the dominant technical sense concerns dynamical control of self-renewal, differentiation, and tissue organization.

8. Synthesis and open interpretive themes

Across experimental embryology, network theory, stochastic processes, mechanics, and computational morphogenesis, stem systems are consistently presented as multi-level control systems. Genetic programs such as sequential HOX activation, biochemical cascades such as BMP $k$6 Wnt $k$7 Nodal, mechanical events such as convergent extension and ingression, and physical principles such as symmetry breaking, scaling, and phase separation jointly determine lineage outcomes and tissue-level form (Siggia, 2018).

Several recurring conclusions emerge. First, stemness is often emergent rather than intrinsic: in the stochastic conveyor-belt model, functional stem-cell number depends on positional fluctuations and niche geometry (Corominas-Murtra et al., 2020); in crowding-based models, homeostasis depends on negative feedback mediated by limited space (García-Tejera et al., 2022); and in dedifferentiation models, cyclic return flux can rescue homeostasis that strictly hierarchical lineages cannot maintain under noise (Yu et al., 12 Jan 2026). Second, irreversible differentiation is not merely a terminal commitment process; in morphogenetic simulations it is the basis of a robust division of labour between shape-generating and shape-stabilizing cell types (Devlin et al., 25 Mar 2025). Third, network topology matters, but topology alone is insufficient: locality of activation, interaction context, and non-network mutations can separate normal from malignant outcomes even when the underlying control graphs are isomorphic (Werner, 2016).

A persistent controversy concerns whether stem-cell potential is fundamentally cell-intrinsic or imposed by tissue-level dynamics. The available models do not resolve this uniformly. Rather, they partition the problem: some frameworks encode stemness in internal loop structure and bifurcation architecture (Werner, 2016, Goto et al., 2013), while others derive it from niche access, geometry, or noise-mediated competition among otherwise equivalent cells (Corominas-Murtra et al., 2020). This suggests that intrinsic and extrinsic accounts are not mutually exclusive but operate at different levels of description.

Another important interpretive theme is the relation between in vitro and in vivo systems. ESC colonies, cysts, and organoids are valuable because they reconstitute core signaling and mechanical circuits while remaining tractable (Siggia, 2018). Yet their legitimacy depends on careful embryo benchmarking rather than on phenomenological resemblance alone. A similar caution applies to abstract network and morphogenetic simulations: they are strongest as generators of mechanistic hypotheses and scaling predictions, not as literal copies of any specific tissue.

Stem systems, in the biological sense, are therefore best understood as an integrated research domain centered on how multicellular lineages achieve controlled self-renewal, discrete fate choice, robust patterning, and stable tissue architecture under noise. The field’s distinctive contribution lies in linking these phenomena across scales—from intracellular attractors and developmental control loops to colony geometry, tissue mechanics, and organismal morphogenesis (Siggia, 2018, Werner, 2016, Devlin et al., 25 Mar 2025).