paper
active
2024
paper:1-s2-0-s0006291x2400932x-main

Stress sharing as cognitive glue for collective intelligences: A computational model of stress as a coordinator for morphogenesis

ByLakshwin Shreesha·Michael Levin
DOI 10.1016/j.bbrc.2024.150396
Collective MorphogenesisCollective Stress CoordinationCognitive glue & collective intelligenceCollective intelligence & bioengineering morphogenesisAverage Euclidean cell movement distance"biological morphogenesis is not well-described by bottom-up, open-loop, mechanical models, but instead exhibits more complex policies as it navigates the space of possible anatomical shapes despite various perturbations"Distress SignalError minimizationExploratory activity level (temperature)FitnessGenotypeMorphogenetic efficiencyPositional information gradientQuote: stress as shared problemQuote: stress leak as signaling mechanismSelection (Darwinian)Stress SharingTarget morphology

TL;DR

Stress sharing between cells accelerates collective morphogenesis and expands individual cells' cognitive light cone, as demonstrated in a multiscale agent-based computational model of embryogenesis published in Biochemical and Biophysical Research Communications 731 (2024) 150396. In 30×30 grid populations evolved over 1000 generations using a genetic algorithm (GA) with three conditions — stress-sharing, without-stress-sharing, and hardwired — stress-sharing populations reached maximum phenotypic fitness by generation ~400, while hardwired populations achieved only 0.975 fitness and without-sharing populations plateaued at ~0.91 by generation 1000. The method introduced is a stress-propagation mechanism embedded in a 2D agent-based developmental layer within the GA, in which blocked stressed cells broadcast a binary stress signal across a 3×3 neighborhood, inducing fixed neighboring cells to form temporary tunnels that permit long-range cell migration; stress-sharing embryos showed a radius of cellular influence of 30 units at developmental step 1, declining to zero by step 85, versus only 5 units terminating at step 10 for non-sharing embryos. Critically, stress maps during development do not predict the target morphology to external observers — similarity between stress map and thumbs-up target rose from 0.53 to 0.63 while similarity to the smiling-face target fell from 0.48 to 0.25 — indicating that the goal state is functionally private to the system itself. The paper argues this implies that stress sharing is a candidate universal coordination mechanism for scaling single-cell homeostatic competency into robust collective intelligence, with direct implications for regenerative medicine, bioengineering of synthetic organisms, and adaptive robotics.

What to take away

  1. 1. In 30×30 grid populations evolved over 1000 generations, stress-sharing populations reached maximum phenotypic fitness by generation ~400, while without-sharing populations plateaued at a phenotypic fitness of ~0.91 and hardwired populations reached 0.975, both failing to achieve maximum fitness (p≪0.01 at generation 400 between stress-sharing and each control).
  2. 2. Stress-sharing embryos in 20×20 grids solved the morphogenetic target (a binary smiling-face pattern) by generation ~100, whereas hardwired populations required the full 1000 generations and without-sharing populations never reached maximum fitness at that scale.
  3. 3. At 50×50 grid size, none of the three population types reached maximum phenotypic fitness within 1000 generations, demonstrating that morphogenetic task difficulty scales with grid size and that the benefit of stress sharing manifests as a right-shift in time rather than a qualitative change in outcome.
  4. 4. The stress-sharing mechanism works by allowing blocked stressed cells to broadcast a binary stress signal across a 3×3 neighborhood, coercing fixed neighboring cells to form temporary movement tunnels, which permits unrestricted long-range cell migration toward target positions without dislodging already-correctly-placed cells.
  5. 5. Stress-sharing embryos exhibited a radius of cellular influence (cognitive light cone) of 30 units at developmental step 1, declining nonlinearly to zero by step 85, while without-sharing embryos had a radius of only 5 units that terminated entirely by step 10.
  6. 6. Cells in stress-sharing populations moved an average Euclidean distance of ~2500 units per generation until generation ~400, utilizing maximum competency of 4725 units, compared to ~200 Euclidean units and ~100 competency units for without-sharing populations throughout all 1000 generations.
  7. 7. The similarity between the stress map and the thumbs-up target pattern increased from ~0.53 to ~0.63 over 100 developmental stages, while similarity to the smiling-face target decreased from ~0.48 to ~0.25 over 80 stages, demonstrating that morphogenetic goal states are not deducible by external observation of stress dynamics alone.
  8. 8. Genomes of stress-sharing and hardwired populations evolved with statistically indistinguishable genotypic fitness until generation ~400 (p = 0.9 at 30×30), after which stress-sharing genomes plateaued while hardwired genomes continued improving, revealing that stress-sharing populations exploit developmental competency rather than purely genomic optimization.
  9. 9. The model raises the open question of whether the identified stress-sharing dynamics generalize to three-dimensional morphospaces and to multi-scale coupled homeostats operating simultaneously at molecular, cellular, tissue, and organismal levels, which the current 2D fixed-grid architecture cannot address.
  10. 10. To replicate the core evolutionary comparison, a researcher should initialize populations of M 2D matrices of size N by randomly scrambling the binary target pattern, encode a reorganization-marker gene in each genome specifying stress-sharing, without-sharing, or hardwired development, run a GA of development → selection of top 10% phenotypic fitness → random cell-pair mutation for repopulation, and track both phenotypic and genotypic fitness of the best individual across 1000 generations averaged over 10 independent runs with 95% confidence intervals.

Peer brief — for seminar discussion

Shreesha and Levin construct a multiscale agent-based computational model of morphogenesis to test whether stress — defined as a binary physiological signal proportional to a cell's distance from its homeostatic setpoint in anatomical morphospace — can function as a coordination mechanism for multicellular collectives when shared between cells. The model places cells on 2D grids of sizes 20×20, 30×30, and 50×50, tasks them with rearranging from a scrambled initial state into a target binary pattern (a smiling face), and embeds this developmental process within a genetic algorithm featuring selection on the top 10% of phenotypic fitness followed by stochastic cell-pair mutation. Three population types are compared across 1000 generations: stress-sharing embryos, in which stressed cells broadcast their stress signal across a 3×3 neighborhood to temporarily tunnel through fixed neighbors; without-sharing embryos, which possess the same homeostatic drive but cannot export stress; and hardwired embryos with no developmental reorganization. The introduced method is the stress-propagation tunnel mechanism, an alternative to which would be a purely reaction-diffusion morphogen gradient model without homeostatic error signaling. The load-bearing finding is that stress sharing produces dramatically faster morphogenetic convergence: at 30×30, stress-sharing populations reach maximum phenotypic fitness by generation ~400, while hardwired populations achieve only 0.975 and without-sharing populations plateau at ~0.91 by generation 1000 (p≪0.01 at generation 400 for both comparisons). The mechanism is quantified: stress-sharing cells travel ~2500 average Euclidean units per generation using ~4725 competency units, versus ~200 units and ~100 competency units for without-sharing cells. The spatial consequence is a cognitive light cone — the radius over which one cell alters another's future fate — of 30 units at developmental step 1 in sharing embryos, collapsing to zero by step 85, versus 5 units collapsing by step 10 without sharing. A secondary and conceptually distinct finding is that stress maps are not predictive of target morphology: similarity between stress gradients and the thumbs-up target rose from 0.53 to 0.63, while similarity to the smiling-face target fell from 0.48 to 0.25, with neither trend enabling goal-state inference. The paper interprets this as a primitive form of cognitive privacy — morphogenetic goal states are functionally inaccessible to external observers. The authors predict that HSP90, observed extracellularly, may serve as a candidate stress-leak molecule in vivo, a claim currently under experimental investigation in their lab. The most contestable aspect is the binary, 2D, fixed-grid abstraction: stress is treated as a simple on/off signal rather than a graded, spatiotemporally dynamic quantity, cells can only be in one of two states, and the grid cannot itself deform as cells move. This means the model cannot capture the recurrent mechanical feedbacks central to real morphogenesis, where cell movement reshapes the action landscape for other cells. A critical reader would also note that the genetic algorithm selects for phenotypic fitness without varying the stress-sharing parameter within a population — each population is homogeneous for its developmental mode — so the model does not directly demonstrate that stress-sharing would be evolutionarily selected over non-sharing in a mixed population, which would constitute a stronger test of the coordination hypothesis. The scalability ceiling (50×50 on a 256-core machine already shows incomplete convergence) further limits claims about whether the benefit quantitatively extends to biologically realistic cell numbers.

Findings (14)

Claims (12)

Hypotheses (6)

Questions (2)

Related work— refs + corpus + external arXiv

Cited / in-corpus / arXiv badges show which signals surfaced each row. Multi-source rows weighted higher.

+25 more

Similar preprints — Semantic Scholar

Cited by (1)

Cross-corpus bridges (6)

same_concept_as · Nomic cosine

External markdown files that talk about the same concept as this entity.

  • aboutblank_kb
    How does stress sharing enable cells to navigate morphogenetic task spaces that would be impossible for uncoordinated cells?questions/how-does-stress-sharing-enable-cells-to-navigate.md0.858
  • aboutblank_kb
    How do individual cells with their own homeostatic competencies coordinate to achieve larger-scale, reliable morphological endpoints?questions/how-do-individual-cells-with-their-own-homeostatic.md0.824
  • aboutblank_kb
    Virtual Embryo Modelframeworks/virtual-embryo-model.md0.809
  • aboutblank_kb
    Stress-Induced Evolutionary Innovationframeworks/stress-induced-evolutionary-innovation.md0.805
  • aboutblank_kb
    Agent-Based Model Of Morphogenesisframeworks/agent-based-model-of-morphogenesis.md0.802
  • aboutblank_kb
    Lakshwin Shreeshathinkers/lakshwin-shreesha.md0.797