Robustness and biological validation of de Gennes' self-generated repellant-gradient mechanism for collective axon growth fronts

In 2007, P.-G. de Gennes proposed a mean-field mechanism by which a population of simultaneously growing axons organizes itself: neuron somata synthesize a chemorepellant, ship it by fast axonal transport to their growth cones, which secrete it into the extracellular matrix; the resulting self-generated repellant gradient biases the random motility of nearby cones toward low repellant, and the feedback between cone density and repellant concentration produces a collective growth band of finite width terminated by a relatively sharp front. The one-dimensional theory predicts a density-dependent ("hypodiffusive") effective diffusion for the growth-cone density, an exactly parabolic finite-support profile, and subdiffusive front widening scaling as $t^{1/3}$ in the strong-feedback regime. The paper itself flags the model as "highly tentative" and declares four idealizations: a single secreted repellant only, no degradation of the repellant during axonal transport, an extinction length assumed small compared with the profile width, and a crude jump-and-drift description of cone motility. No work in the eighteen years since publication resolves the mechanism: the 2022 review by Oliveri and Goriely reproduces the similarity solutions but states that the model's main biological assumptions "have not been directly justified and are greatly idealized". The open problem is to determine whether this self-generated-front mechanism is robust and biologically real — to analyze the declared idealizations (cue competition, transport losses, the long-extinction regime, realistic motility) plus the two-dimensional and multi-species extension, and to measure the model's parameters and test the $t^{1/3}$ front-widening law and finite-support parabolic profiles in a real neuronal system.

Background

During development of the nervous system, neurons extend long processes (axons) whose growing tips, called growth cones, crawl through tissue and wire neurons together. Growth cones read the chemical landscape around them: signaling molecules operating at very low concentrations either attract or repel them (chemoattractants and chemorepellants, collectively guidance cues). Experiments on single isolated growth cones show that their advance is intrinsically stochastic — a bistable alternation of protrusion and retraction, well described as a stochastic process (Betz, Lim and Käs 2006) — while the borders of developing brain regions (for example, visual-area borders) are remarkably sharp. Classically, such sharp boundaries were attributed to nonlinear chemical kinetics acting on morphogen gradients (spatial concentration patterns of diffusible developmental signaling molecules) imposed by other tissues (e.g. Kerszberg and Changeux 1994).

P.-G. de Gennes (2007) asked whether a sharp collective growth front could instead arise without imposed gradients, from a single self-generated repellant. His one-dimensional mean-field model — a continuum description in which only the density of growth cones, not individual axons, is tracked — considers neurons with somata (cell bodies) in a source plane (areal density $\nu$) sending out axons whose length grows linearly in time. Each soma synthesizes a chemorepellant (local concentration $\delta_N$) that is transported along the axon on microtubules (the intracellular filaments along which motor proteins shuttle cargo) at a speed much larger than the elongation speed, so that a fixed useful fraction $f$ of the synthesized molecules arrives at the growth cone. The cone secretes the repellant, which diffuses in the surrounding glial matrix, the supporting tissue between neurons (diffusion constant $D_i$), and is scavenged with lifetime $\tau$, producing an exponential spatial kernel with extinction length $\kappa^{-1}=(D_i\tau)^{1/2}$. In a local steady state the total repellant concentration becomes proportional to the local density of cones, $\delta(x)\propto G_t(x)$, where $G_t(x)$ is the growth-cone density profile at time $t$. Each cone is modeled as performing random jumps of size $a$ with correlation time $\theta$ (giving a bare diffusion constant $D_0\simeq a^2/\theta$), plus a drift velocity toward low repellant, $v=-k\,\partial\delta/\partial x$ with $k=D_c/C_0$, where $C_0$ is the threshold repellant concentration at which the jump statistics become significantly biased. The resulting nonlinear transport equation for $G_t$ has a density-dependent effective diffusion constant $D_{\mathrm{eff}}=D_0(1+(\phi_a/C_0)Ga)$, where $\phi_a$ collects the secretion parameters. In the "strong" regime the density-dependent term dominates and the equation becomes what de Gennes calls "hypodiffusive": effective diffusion vanishes where cones are sparse, so the density profile terminates abruptly at a finite front. The exact strong-regime solution is a finite-support parabolic profile

$$G_f(x)=\frac{3}{2\bar{x}}\left(1-\frac{x^2}{\bar{x}^2}\right),\qquad 0\le x\le\bar{x}(t),$$

whose front position (the width of the growth band) grows only slowly, subdiffusively, as $\bar{x}\simeq(aD_0\psi t)^{1/3}\propto t^{1/3}$, with the dimensionless feedback parameter $\psi\equiv4\pi\nu\delta_N f/(\kappa^2 C_0)$; the strong regime requires $\psi\gg1$ and holds for times $t/\theta\ll\psi^2$, and a thin linear-diffusion boundary layer of width $\lambda\ll\bar{x}$ smooths the front tip. Two collective predictions define the model: (i) subdiffusive $t^{1/3}$ widening of the axonal growth band, slower than the $t^{1/2}$ spreading of ordinary diffusion, and (ii) a sharp, finite-support front with a parabolic density profile.

The paper's Discussion declares the model "highly tentative" and lists its idealizations: only one secreted repellant is considered, whereas real growth cones also respond to many other signaling molecules from other territories; degradation of the repellant during its axonal transport is ignored; the extinction length $\kappa^{-1}$ is assumed small compared with the profile width, and in the opposite regime de Gennes notes that "most of the inhibitor would be spread in a useless region", potentially destroying the feedback; and the coarse jump-and-drift description of cone motility is "primitive" compared with the single-cone experiments. Fifteen years later, the review of Oliveri and Goriely (2022), which critically surveys mathematical models of neurite growth (neurites being the axon-and-dendrite processes that neurons extend), reproduces both similarity solutions (self-similar scaling solutions) of the model — a sharp front of slope $\sim t^{-2/3}$ advancing like $\sim t^{1/3}$ in the strong nonlinear regime, versus band widening $\sim t^{1/2}$ under regular diffusion — calls the paper "relatively unnoticed", states that "from a biological standpoint, the main assumptions of the model have not been directly justified and are greatly idealized", and observes that continuum mean-field approaches have "not been applied systematically to nerve cells".

Problem Statement

Determine whether de Gennes' self-generated repellant-gradient mechanism is a robust and biologically real mechanism for collective axon growth — by resolving the idealizations its one-dimensional theory deliberately made and by testing its two central predictions against a real neuronal system. Concretely, the following are open.

(1) Cue competition. How do additional external cues and competing autocrine/paracrine signaling molecules — signals secreted by the growing neurons themselves (autocrine) or by nearby cells (paracrine), the attractants and repellants that real growth cones also sense — alter or compete with the self-generated repellant front?

(2) Transport losses. What is the quantitative effect of degradation of the repellant during its axonal transport on the useful fraction $f$ (the fraction of synthesized repellant that arrives at the growth cone), on the feedback parameter $\psi\equiv4\pi\nu\delta_N f/(\kappa^2 C_0)$ (the dimensionless strength of the repellant-mediated feedback, with $\nu$ the areal density of somata, $\delta_N$ the synthesized repellant concentration per soma, and $C_0$ the threshold concentration for repellant action), on the existence of the strong regime, and on the resulting front?

(3) Long-extinction regime. Analyze the regime $\kappa\bar{x}<1$ in which the repellant's extinction length $\kappa^{-1}$ exceeds the profile width $\bar{x}$; there the local proportionality $\delta\propto G_t$ fails and the density-repellant coupling becomes nonlocal. Does any collective front survive, and with what scaling?

(4) Realistic motility. Do the $t^{1/3}$ widening law and the finite-support parabolic profile survive replacement of the coarse jump-and-drift cone description by realistic stochastic protrusion/retraction dynamics of the kind measured for single growth cones?

(5) Empirical status. Measure the model's parameters ($\nu$, $f$, $D_i$, $\tau$, $C_0$, $a$, $\theta$) in a developing or cultured neuronal system with an identified axonally transported, cone-secreted repellant, and directly test the predicted subdiffusive $t^{1/3}$ front widening (growth-band width growing like $t^{1/3}$, slower than the $t^{1/2}$ of ordinary diffusion) and the finite-support parabolic density profiles (nonzero only over a bounded interval) against alternatives such as $t^{1/2}$ diffusive spreading.

(6) Geometry and multi-species extension. Extend the mechanism from its one-dimensional reduction to two-dimensional tissue geometries and to interacting multiple cue species, including the stability of the front against transverse perturbations (perturbations varying along the front rather than across it).

The common objective uniting these requests: decide the scientific status of the mechanism — whether a single self-generated, axonally transported chemorepellant can self-organize collective axonal growth fronts in real neural tissue, or whether the sharp-front prediction is an artifact of the idealizations. The requested theoretical results (items 1-4, 6) jointly establish the mechanism's regime of validity, and the empirical test (item 5) jointly decides whether that regime is realized in a real system.

References

1. de Gennes, P.-G. (2007). Collective neuronal growth and self organization of axons. Proceedings of the National Academy of Sciences 104(12), 4904-4906. DOI: 10.1073/pnas.0609871104. https://doi.org/10.1073/pnas.0609871104
2. Betz, T., Lim, D., Käs, J. A. (2006). Neuronal Growth: A Bistable Stochastic Process. Physical Review Letters 96(9), 098103. DOI: 10.1103/PhysRevLett.96.098103. https://doi.org/10.1103/PhysRevLett.96.098103
3. Betz, T., Koch, D., Lim, D., Käs, J. A. (2009). Stochastic actin polymerization and steady retrograde flow determine growth cone advancement. Biophysical Journal 96(12), 5130-5138. DOI: 10.1016/j.bpj.2009.03.045. https://doi.org/10.1016/j.bpj.2009.03.045
4. Zubler, F., Douglas, R. (2009). A framework for modeling the growth and development of neurons and networks. Frontiers in Computational Neuroscience 3, 25. DOI: 10.3389/neuro.10.025.2009. https://doi.org/10.3389/neuro.10.025.2009
5. Kerszberg, M., Changeux, J.-P. (1994). A model for reading morphogenetic gradients: autocatalysis and competition at the gene level. Proceedings of the National Academy of Sciences 91(13), 5823-5827. DOI: 10.1073/pnas.91.13.5823. https://doi.org/10.1073/pnas.91.13.5823
6. Khan, G. M., Miller, J. F. (2014). In search of intelligence: evolving a developmental neuron capable of learning. Connection Science 26(4), 297-333. DOI: 10.1080/09540091.2014.922052. https://doi.org/10.1080/09540091.2014.922052
7. Buldyrev, S. V. (2012). Fractals in Biology. In: Mathematics of Complexity and Dynamical Systems, 488-511. DOI: 10.1007/978-1-4614-1806-1_31. https://doi.org/10.1007/978-1-4614-1806-1_31
8. Bastien, R., Porat, A., Meroz, Y. (2019). Towards a framework for collective behavior in growth-driven systems, based on plant-inspired allotropic pairwise interactions. Bioinspiration & Biomimetics 14(5), 055004. DOI: 10.1088/1748-3190/ab30d3. https://doi.org/10.1088/1748-3190/ab30d3
9. Tweedy, L., Insall, R. H. (2020). Self-Generated Gradients Yield Exceptionally Robust Steering Cues. Frontiers in Cell and Developmental Biology 8, 133. DOI: 10.3389/fcell.2020.00133. https://doi.org/10.3389/fcell.2020.00133
10. Tweedy, L., Thomason, P. A., Paschke, P. I., Martin, K., Machesky, L. M., Zagnoni, M., Insall, R. H. (2020). Seeing around corners: Cells solve mazes and respond at a distance using attractant breakdown. Science 369, eaay9792. DOI: 10.1126/science.aay9792. https://doi.org/10.1126/science.aay9792
11. Oliveri, H., Goriely, A. (2022). Mathematical models of neuronal growth. Biomechanics and Modeling in Mechanobiology 21(1), 89-118. DOI: 10.1007/s10237-021-01539-0. https://doi.org/10.1007/s10237-021-01539-0
12. Cheng, K., Kumarasinghe, U., Staii, C. (2025). Feedback-Driven Dynamical Model for Axonal Extension on Parallel Micropatterns. Biomimetics 10(7), 456. DOI: 10.3390/biomimetics10070456. https://doi.org/10.3390/biomimetics10070456
13. Kassianides, C., Goriely, A., Oliveri, H. (2025). The multiscale mechanics of axon durotaxis. Journal of the Mechanics and Physics of Solids 200, 106134. DOI: 10.1016/j.jmps.2025.106134. https://doi.org/10.1016/j.jmps.2025.106134 (preprint: arXiv:2501.01868)
14. Staii, C. (2025). Stochastic Models of Neuronal Growth. AppliedMath 5(4), 170. DOI: 10.3390/appliedmath5040170. https://doi.org/10.3390/appliedmath5040170