Embryos assist morphogenesis of others through calcium and ATP signaling mechanisms in collective teratogen resistance

Information for organismal patterning can come from a variety of sources. We investigate the possibility that instructive influences for normal embryonic development are provided not only at the level of cells within the embryo, but also via interactions between embryos. To explore this, we challenge groups of embryos with disruptors of normal development while varying group size. Here, we show that Xenopus laevis embryos are much more sensitive to a diverse set of chemical and molecular-biological perturbations when allowed to develop alone or in small groups, than in large groups. Keeping per-embryo exposure constant, we find that increasing the number of exposed embryos in a cohort increases the rate of survival while incidence of defects decreases. This inter-embryo assistance effect is mediated by short-range diffusible signals and involves the P2 ATP receptor. Our data and computational model emphasize that morphogenesis is a collective phenomenon not only at the level of cells, but also of whole bodies, and that cohort size is a crucial variable in studies of ecotoxicology, teratogenesis, and developmental plasticity.

During embryogenesis, a complex anatomical form is constructed via the interactions of large numbers of individual cells. A key question, with impacts ranging across evolution, regenerative medicine, and bioengineering, concerns the origin of the information enabling correct morphogenesis. Most of the emphasis to date has been placed upon the information provided vertically from parent to offspring—the genome—and the inanimate environment1–8. The typical perspective is of a single embryo and its local microenvironment as the arena within which cells compete and cooperate using their genomically specified cellular hardware to complete embryogenesis in a given situation. While work on developmental plasticity and extended phenotype do consider factors outside of individual bodies9–13, the embryo (and its outer perimeter) is most commonly taken to be the natural, self-contained unit of studies on control mechanisms and the origin of specific anatomies in evolutionary morphology, reproductive toxicology, and developmental genetics. However, many phenomena in biology exhibit scale-free or at least multi-scale dynamics14–16. Thus, we explored the possibility of relationships between development and external social environments, specifically whether instructive information could also propagate horizontally, enabling embryos to benefit from a kind of “wisdom of the crowd” in their cohort17–20.

While the focus of most studies has been the self-contained genome, a number of prior studies have also reported extra-genomic influences on developmental outcomes21, in the context of phenotypic plasticity11,22–24. Of particular interest were studies focusing on inter-conspecific interactions or interactions amongst members of the same cohort. For example, rat studies have shown that the uterine environment and position in the uterine horn can affect progeny. Females that were located contiguous to a caudal male or between two males exhibited more masculine characteristics, mounting behavior, and morphology than females contiguous to a cephalic male or near no males25,26. Adverse intrauterine conditions lead to females being bigger than their male littermates and increased uterine occupancy leads to smaller males27. In humans, studies have shown that skin-to-skin contact between caregivers and preterm infants can have positive cardio-respiratory impacts28,29. Inter-embryo communication in the context of predation has been observed in a number of species ranging from birds to frogs30–34.

These lateral interactions go beyond straightforward negative effects such as competition for nutrients, oxygen levels, and accumulating waste products, and suggest that there are also beneficial effects from being part of a collective. Postnatally, the Allee effect35 is a well-known effect of animal aggregation and its correlation with fitness. It is defined as a positive correlation between population density and individual fitness. Studies in fish, rodents, and planaria have shown that having conspecifics can positively impact future health outcomes. For example, large groups of planaria and goldfish survive colloidal silver exposure better than small groups. In line with other studies showing the beneficial effects of larger groups, dense populations of starfish were able to right themselves faster than sparse populations35–38.

These effects are seen at the cellular level as well, from social amoebas39 to metazoan cells in culture40. In mesenchymal cells in vitro, low population seeding impacts the differentiation potentials and alters cell fates and growth rates of tumors, all of which are affected by cell density41,42. It has long been realized that groups of cells resist transformation and cancer43–46, providing a collective dynamic that keeps cell activity orchestrated towards adaptive outcomes47. One such example of cellular resistance is the culturing of multiple ovarian follicles, which improves follicular survival and growth through paracrine signaling from one follicle to another48. In the case of tissue injury, paracrine factors have also been shown to play an important role. In Xenopus embryos, wounding induces calcium waves within a single organism that aid in cell-cell coordination and permit rapid wound closure in response to superficial injury49,50. ATP is one such ligand that can trigger cell contractility and plays a role in coordinating long-range contraction responses after injury51. However, the transfer of such information between individuals or distinct tissues and organs has not been visualized or quantified.

Thus, a number of cases of beneficial lateral interactions have been reported, and the study of cell cooperativity and instructive communication is an active field. However, a significant knowledge gap exists with respect to instructive lateral interactions for correct development at a level of organization above that of tissues, organs, and individual organisms. While epigenetic studies have explored how genes can be impacted on environment and behavior, the information transfer from this field is still vertical. Here, we sought to test the hypothesis that embryos can interact across distance in their medium to provide beneficial, specific information that assists morphogenesis. Our fundamental assay was resistance to teratogens: we exposed groups of Xenopus laevis embryos to several disruptors of normal development with diverse mechanisms of action and asked whether large groups can resist exposure better than small ones. We found that collective development is much more stable, with respect to morphological perturbations (birth defects), and analyzed an agent-based cellular automata computational model that explains the collective morphogenetic stability of larger embryonic groups through local interactions. Functional experiments reveal that this effect is mechanistically mediated by a short-range chemical signal, requiring calcium and P2 receptor signaling. It also does not involve detectable transcriptional changes at stage 25, but does implicate a small group of changes by stage 35. We also report the remarkable phenomenon of a mechanical injury-triggered calcium wave that propagates not only within embryos, as has been shown previously49–51, but also between embryos, identifying it as a potential candidate for the communication process within a network of developing individuals. Finally, our transcriptomic profiling of small vs.

large groups identifies a set of up- and down-regulated genes that provide a signature of the collective response to teratogen challenge.

To test the hypothesis of collective influences on development, we asked whether group size was a factor in embryos’ ability to undergo normal morphogenesis despite the presence of well-known teratogenic influences. Embryos for each experiment were from the same clutch but separated into differing group sizes. Importantly, we kept teratogen concentration constant between groups by scaling media volume to group size so that the per-embryo teratogen exposure was constant regardless of group size.

We began with thioridazine, a chemical reagent that targets dopamine pathways52,53. Xenopus laevis were raised in identical, standard conditions but at different group sizes within each Petri dish, and subjected to 90 μM of thioridazine across all densities (Fig. 1a). Media and dish size were scaled proportionally between large and small groups so that the only difference between experimental conditions was the number of animals in a dish. Large groups (n = 300) received 120 mL of media while small groups (n = 100) received 40 mL of media. To rule out greater drug degradation by larger cohorts as a potential confounder, we quantified thioridazine in the medium following the treatment of embryos in groups of various sizes. Liquid chromatography–mass spectrometry analysis indicated that there was no notable difference in thioridazine concentration between groups ranging in size from 25 to 400 (Fig. 1b) beyond that attributable to compound degradation during the preparation and transport of samples.Fig. 1Conspecifics help resist teratogenic effects.a Embryos were reared in large (n = 300) or small (n = 100) groups and exposed to the dopaminergic agent thioridazine during N&F stages (13-26). The concentration of the drug was kept equal at 0.4 μg/mL of media and the amount of media with the drug was scaled according to the group size to match the volume of the drug per embryo. b Embryos were treated with thioridazine in various group sizes. Post-treatment, media were harvested and thioridazine concentration was determined by LC-MS. The table shows the concentration of thioridazine in media from embryos treated in groups of 25, 50, 100, 200, 300, and 400. c Percentage of surviving animals in groups of increasing size from 1 to 300 following a 24-h treatment with 90 μM thioridazine.

Data was compared using a one-way ANOVA test and Tukey test; points are the average of 3 replicates. ****p < 0.0001, **p = 0.0024. d Craniofacial phenotype of a normal tadpole and a thioridazine-treated tadpole at stage 45. Control embryo with normal, round head shape and normal pigmentation (blue arrows). Thioridazine-treated embryo with square head shape and hyperpigmentation (purple arrows). e Total frequency of malformed square heads, and hyperpigmentation in large group thioridazine exposure (n = 300 in gray) and small group treatment (n = 100 in pink) across N = 6 trials. Comparisons were done using two-tailed Welch’s t-test with p = 0.0003 in square heads and p = 0.0489 in hyperpigmentation. f Quantification of hyperpigmentation between untreated (control) and thioridazine-treated hyperpigmented animals. A total of 30 control and 30 thioridazine-treated embryos were used in each trial. Two-tailed Welch’s t-test was used for comparison and p = 0.0029. g Ventral view of normal and square-headed animals with lines indicating where width measurements were taken. h Quantification of face width difference between untreated (left) and treated square head (right) animals. A total of 30 square head and 30 control embryos were examined for each trial. Comparison is done with a two-tailed Welch’s t-test with a p < 0.001. For all animal images, the anterior end is at the top of the image. All images unless noted are dorsal views. Data are plotted as mean ± SD and each point represents a trial. All treatments are normalized to the mean of non-treated controls’ spontaneous defects. Scale bars represent 2 mm. Source data are provided as a Source Data file.

Following treatment with 90 μM thioridazine for 18 h or stages 12.5–25, we observed both embryonic death and a range of defects in survivors and scored their incidence. Survival increased with increasing group size (Fig. 1c). For singlets and groups of 5 embryos, thioridazine exposure resulted in 0% survival. When exposed in groups of 25 and 75, averages of 13% and 44% of embryos survived, respectively. For group sizes of 100 and 300, survival rates increased to averages of 85% and 98%, respectively (N = 3, p = 0.0051 using ANOVA). Among survivors, thioridazine caused significant developmental abnormalities, as expected. Compared to age-matched controls, exposure to thioridazine resulted in a rectangular head shape, misshapen eyes, and hyperpigmentation (Fig. 1d and Supplementary Fig. 1A, B). After exposure, tadpoles in the largest group treatment (n = 300) and smaller group treatment (n = 100) were scored at NF stage 45 for defects (Fig. 1e). Large groups had significantly fewer average incidents of square head (22.7% vs. 40.7%) (see Methods section for quantitative procedure for assessing phenotype) and hyperpigmentation (37.8% vs. 54.2%) (N = 6, p = 0.00029 using Welch’s t-test). While other types of phenotypic defects were observed (Supplementary Fig. 1C), here we chose to focus on head shape and pigmentation for the in-depth analysis of cross-embryo effects.

As quantitative validation that individuals scored with the hyperpigmentation phenotype were truly different from controls, 10 tadpoles were pooled, dissolved, and measured for absorbance at 260 nm. Control animals had an average absorbance of 0.427 while hyperpigmented animals had an average absorbance of 0.615 (Fig. 1f) (N = 3, p = 0.0029 using Welch’s t-test). We used morphometric analysis to further quantify the square head shape phenotypes by determining the difference between the diameter of the head and the width at the base of the branchial arch (Fig. 1g). In treated animals with square heads, this number is significantly smaller than in controls (261.2 vs. 23.5, respectively. N = 3, p < 0.0001 using Welch’s t-test) (Fig. 1h). To prevent bias, blind scoring was used for three of the six experimental replicates of thioridazine.

We conclude that larger group size confers a significant protective effect against both thioridazine-induced death and craniofacial defects in a manner that cannot be explained by increased drug degradation in larger cohorts. We call this phenomenon the Cross-Embryo Morphogenetic Assistance (CEMA) effect.

We next asked whether the surprising CEMA effect is a feature of thioridazine teratogenesis specifically or whether it is a wider phenomenon, and, if so, what kinds of perturbations can embryos resist collectively better than singly? We explored additional teratogens orthologous to thioridazine: other drugs with a different target, and a non-pharmacological method using misexpression of mutant mRNA.

Two drugs with different targets from each other and from thioridazine were used to test the robustness of the CEMA effect. We first tested forskolin—a drug that targets adenylate cyclase and results in hyperpigmentation (Fig. 2a and Supplementary Fig. 2B)54. Tadpoles were subjected to 5 µM of forskolin at stages 10–45 in group sizes of small (n = 25), medium (n = 50), and large (n = 300) (Fig. 2b). In the small and medium groups, forskolin resulted in hyperpigmentation in an average of 93.3% and 86% of embryos, respectively, while the large group exhibited hyperpigmentation in an average of just 13.9% of embryos (N = 3, p < 0.001 using ANOVA). Blind scoring was done in two of three replicates to prevent any bias.Fig. 2Conspecific effect beyond thioridazine exposure.a Images of control (left) and forskolin-treated animals (right), demonstrating hyperpigmentation following forskolin exposure (purple arrows). b The graph shows the total frequency of hyperpigmentation in embryos exposed to forskolin in small (n = 25), medium (n = 50), and large (n = 300) groups. ****p < 0.001.. c Control animal raised in 0.1x MMR with normal brain structures (left). ****p <0.0001. Representative animals from cohorts subjected to nicotine treatment were then raised at a group size of 75 animals (right). Purple arrows highlight misshapen eyes, malformed forebrain, and midbrain. d Percent total of embryos with brain defects in groups treated with nicotine at densities of n = 10, 25, 50, 75, and 125. ****p < 0.0001, ***p = 0.0008, *p = 0.0197. e Effect of group size on incidence of defects in dominant-negative Kir6.1-injected embryos. Images of control (left) and injected (right) embryos at stage 45. Blue arrows indicate normal eyes and head shape while purple arrows point out misshapen heads and eyes. f Quantification of malformed and normal individuals at stage 45.

Group size of n = 10, 25, 50, and 100. Both cells at the 2-cell stage were injected and embryos were split into different size groups post injection. *p = 0.0111, ****p < 0.0001, ***p = 0.0004, and **p = 0.0028. All animal images are dorsal views and oriented so anterior is facing up and dorsal is down. For all data: values are plotted as mean ± SD, one-way ANOVA, and Tukey tests were conducted. Each dot on the graph is a separate replicate. All treatments are normalized to the mean of non-treated controls’ spontaneous defects. Scale bars represent 2 mm. Source data are provided as a Source Data file.

Next, we tested nicotine, which is known to cause developmental brain and other defects55–57. Embryos were separated into different group sizes at stage 11 and treated with nicotine until stage 35, and then allowed to develop in the regular medium until stage 45. Animals were then scored for abnormalities including missing or malformed eyes, fusion of eye to the brain, and brain defects including absence of forebrain, missing both forebrain and midbrain, malformed hindbrains, or incorrect separation of the forebrain and midbrain (Fig. 2c and Supplementary Fig. 2C). For tadpoles reared and exposed in the smaller groups (n = 10 and n = 25), an average of 96.7% and 94.7%, respectively, exhibited brain defects. Those reared at higher densities (50 and 75) had defects in an average of 46.9% and 33.3% of embryos, respectively. At the highest density, n = 125, only an average of 10.1% of embryos exhibited defects (N = 3, p < 0.001, ANOVA) (Fig. 2d).

We next examined the effect of group size on a mechanistically different kind of perturbation that has previously been shown to cause defects in eye and heart morphogenesis56: disrupting native bioelectric signaling among embryonic cells58–60 by microinjecting mRNA encoding a mutant ion channel. We found that large group size also protected embryos against the negative effects of mutant ion channel expression. At the 2-cell stage, both cells were microinjected with mRNA encoding dominant-negative Kir6.161–63. Following injection, animals were raised in dishes containing varying embryo group sizes. Defects observed at stage 45 included misshapen eyes, overall head malformations, hyperpigmentation, and death (Fig. 2e and Supplementary Fig. 2D). Unlike thioridazine, dominant-negative Kir6.1 microinjections did not induce significant embryo death and we saw no significant differences in survival between embryos raised in different group sizes and controls. There was, however, again a dramatic protective effect of group size on the incidence of craniofacial defects. At a density of 5 embryos per dish, 100% of the injected embryos had defects. Increasing the density to 25 and 50 animals lowered the average defect incidence to 88.2% and 79.7%, respectively. At the highest group size, n = 100 animals, the average defect incidence was 54.2% and the average of normal individuals was 43.0% (N = 3, p = 0.0141 using ANOVA) (Fig. 2f). We note that the mRNA microinjection experiments rule out explanations based on the pharmacokinetics of drugs in different sized groups. Because there is no significant embryo death, they also indicate that the protective effect of larger group size against craniofacial defects is not secondary to an effect on survival.

These results indicate that the protective effect of larger group size on craniofacial defects is not specific to one kind of pathway or teratogen. We conclude that CEMA is a more general developmental stability mechanism that can stabilize several diverse kinds of developmental processes against perturbations.

We next asked whether the assistive effect is only operational across identical conspecifics, or whether genetically diverse groups also reap the benefits. We mixed strains of wild-type and an albino strain of Xenopus laevis, then treated them to ask whether one genetic background can help another (Fig. 3a). A group of 150 wildtype embryos treated with 90 µM thioridazine (stages 12.5–25) had a square head average of 44.7% and 44.3% for 150 treated albino embryos. Increasing the group size to 300 leads to averages of 20.7% for wild types and 23.7% for albinos. Mixing a group of 150 wildtypes and 150 albino embryos had an average of 21% (Fig. 3b). While there were significant differences between the 150 and 300 group sizes, there were no significant differences between wildtype and albino embryos of the same group size. Given this, we conclude that CEMA does not require the large cohort to be genetically homogenous.Fig. 3Genetically diverse populations can also benefit from CEMA but unperturbed cohort members do not aid in stabilization.a Examples of wildtype and albino animals that were untreated (controls) or treated with thioridazine (treated). Blue arrows highlight normal head shape and pigmentation. Purple arrows indicate square heads and hyperpigmentation. b Percent total of embryos with square heads in two strains of Xenopus laevis, wildtype (either n = 150 or n = 300, respectively labeled on the graph) and albino (n = 150 or n = 300, respectively labeled on the graph). ****p < 0.0001. c Averages of the frequency of square head defects in a mixed group of n = 150 thioridazine-treated wildtype embryos + 150 untreated albinos, n = 300 treated wild-type embryos, and n = 150 treated wild-type embryos. **p = 0.0094 and *p = 0.0133. d Average of the frequency of brain defects and square heads in nicotine, thioridazine, and mixed treatments (n = 150 in each group).

**p = 0.0019 and *p = 0.0291 for brain defects and ****p < 0.0001 for square head. All animal images are dorsal views with the anterior end at the top of the image. For all data: values are plotted as mean ± SD, one-way ANOVA, and Tukey test were conducted. All treatments are normalized to the mean of non-treated controls’ spontaneous defects. Scale bars represent 2 mm. Source data are provided as a Source Data file.

In prior experiments, every animal in a given cohort was exposed to the same stressor. We next asked whether animals that had never experienced the teratogen could improve the resistance of conspecifics that were exposed. This would be expected if, for example, animals were providing developmental signals to each other—in that case, unperturbed control embryos would be even more efficient than exposed ones in stabilizing other embryos. Thus, we investigated CEMA in experiments in which wild-type animals exposed to thioridazine were reared with naïve, untreated albino embryos in mixed cohorts. After treatment with thioridazine, wild-type animals were rinsed three times before being mixed in with an equal number of untreated albinos. Consistent with our previous experiments, groups of 150 treated wildtype animals had an average square head incidence of 36.0%, which was significantly reduced to an average of 23.3% in groups of 300 treated wildtype animals (N = 3, p = 0.0068, ANOVA). However, in the mixed cohort (150 treated wildtype animals with 150 untreated albinos), an average of 37.0% had defects, not significantly different from that seen in the group of 150 treated embryos reared alone (Fig. 3c). The 300 mixed group had no significant difference from that of the 150 treated wildtype group, but there was a significantly increased incidence of square heads compared to that of the 300 all treated group (N = 3, p = 0.0068 using ANOVA). Thus, unperturbed individuals do not help stabilize teratogen-exposed individuals and we conclude that only perturbed individuals have a role in CEMA.

In order to distinguish whether CEMA contributes generic protective influence or encodes morphogenetic information for a specific defect, a cross-teratogen experiment was performed using nicotine and thioridazine. Nicotine primarily induces brain defects while thioridazine induces head shape, eyes, and pigmentation defects. Embryos were treated with either nicotine or thioridazine and then mixed together. If CEMA was a generic effect, then it could be expected that groups made up of cohorts of embryos exposed to different insults would function as a large group—exhibit CEMA-induced mitigation of the phenotypes. In contrast, if each type of insult involved phenotype-specific signals spread through the population, then such chimeric groups would not be additive in the way that CEMA requires, because each specific insult cohort would be too small to exhibit significant CEMA-induced repair. We observed that in a group of 100 animals, there was an average of 22% in normal nicotine treatment, 0% in normal thioridazine treatment, and 12.3% in nicotine treatment followed by mixing in with thioridazine conspecifics. With regards to square heads, nicotine treatment alone resulted in 2%, thioridazine alone resulted in 80.7%, and mix groups resulted in 76.7% (Fig. 3d). While there are decreases in the mixed group compared to either the normal nicotine or thioridazine treatment, the decreases are non-significant, suggesting that CEMA does not operate when the large group is made up of subgroups with a different history of teratogen exposure. These data are consistent with individual stressors eliciting their own specific repair information which is not additive and does not cross-protect.

To better understand inter-embryonic signaling within CEMA and guide future studies, we developed a model of the system-level collective dynamics and ran simulations to probe potential explanations and generate testable predictions. Specifically, we developed a computational approach to show how the development of embryos (or any morphogenic agent) may be more stable and harder to disrupt in larger groups than for singletons or small groups. We chose an agent-based approach to virtual embryogeny64–69, where each embryo was a cellular automaton70–77 that could interact with its neighbors through diffuse signaling and was parametrized to reflect the quantitative data in Fig. 1 with regard to survival as a function of cohort size.

For our model, we chose to use elementary cellular automata (ECA), which are 1D arrays consisting of cells that can hold a value of either 0 or 178. At each timestep in a simulation, these cells can change their values based on pre-programmed update rules, wherein they sample their own value and their close neighbor’s (within 1 to 3 cells to the left or right) and either maintain their existing value (0 or 1) or change to the opposite value. This inherent ‘local interactions only’ property makes ECAs a good choice for this model since many biological collective systems (especially during morphogenesis) appear to be governed by parallel local interactions not dependent on a central controller79–81.

We chose to simulate embryonic development using an ECA paradigm that recapitulates the core task: independent subunits (whether cells or larger structures) working toward a species-specific endpoint. The “majority problem” is a task in which given an initial configuration of 0 s and 1 s, the cells of the ECA update their states at each timestep eventually converging on a final state where all cells have the same value—either 0 or 1—matching the value in the majority at initiation (Fig. 4a and Supplementary Fig. 3)82–84. Conceptually, we equate this to normal embryogenesis, in which a healthy embryo starts in a given configuration and develops towards the correct target morphology. However, what update rule is suitable to follow to achieve this goal state? In previous ECA studies, one update rule that solves this task is the GKL rule85, which dictates that a cell’s value at each timestep is set to match the value shared by the majority of a set that includes itself and defined neighbors to the left or right (see Methods). Not only does the GKL rule solve this problem, but it also does so efficiently, defined as converging to the correct value before a number of steps equal to half of the total number of cells.Fig. 4Computational agent-based model of CEMA.a Cartoon example of healthy and teratogen-influenced development. Left, an ECA, a 1D array of 0's and 1's, has an initial configuration of 40% 0's (white) and 60% 1's (black). ECA follows the GKL update rule, solving the majority problem by converging to all 1's. Right, an ECA that has noise induced by teratogens, which does not successfully solve the majority problem. b A cartoon example of inter-embryonic signaling from time t to time t + 1. Embryos signal their current health (red) at time t and supportive signals (black) at the next timestep to their neighbors and nearest neighbors.

Embryos that have not been exposed to the teratogen cannot participate. c Data from the simulation shows that inter-embryonic signaling, CEMA, aids in development in the presence of noise (blue line). Without this communication, development fails in the presence of noise (red line). When half of a cohort is comprised of untreated embryos, they do not participate, and therefore the effect is lower (orange line). Each experiment included the number of embryos equal to the total embryos indicated on the x-axis and each experiment was repeated 50 times. The bars on the graph are 95% confidence intervals and the center represents the mean. Source data are provided as a Source Data file.

Teratogens cause defective development or death by disrupting the endogenous, exquisitely coordinated cellular processes (such as sensing and information processing) that are required for morphogenetic coordination. Thus, we chose to model teratogenic perturbations in development as noise introduced as an alternate update rule, causing cells to mimic a single neighbor’s state at random rather than following the GKL rule (see Supplementary Fig. 3). This noise is sufficient to cause dysfunction in how ECAs develop over time, resulting in incomplete or incorrect solutions at the end of the simulation (Supplementary Fig. 3).

To model conspecific interactions between embryos, we introduced a communication paradigm that relies on local communication in groups, similar to the local interaction between cells within ECAs (described above), but now at the level of interaction between ECAs. First, ECAs (representing embryos in a cohort) were given a spatial location on a 2D grid (i.e., growing embryos in a square configuration). Each ‘embryo’ was assigned a health value, which can vary between 1 (perfectly healthy) and 0 (dead) but was set at 1 for all ‘embryos’ at the start of the simulation (Supplementary Fig. 4). To model teratogen exposure, there is a chance (parameterized to 80%) that each embryo may be affected by noise (representing the deleterious effects of the reagent on the accurate functioning of the cellular machinery), in which case the embryo’s health value is decreased by a percentage (parameterized to 70%). At each timestep, embryos that have been ‘exposed’ in this timestep or in previous timesteps broadcast their health value to their immediate neighbors and to their neighbor’s neighbors (via a more diffuse and weaker signal; see Fig. 4 and “Methods” for details). This neighbor’s neighbors’ interaction proved to be a vital ingredient, as when the model was tested without it, survival plummeted for the same parameterization (Supplementary Fig. 5A). As could be expected for a biological stress signal, an ‘embryo’ in the model broadcasts if and only if it has been exposed. To simulate CEMA, at the next timestep (and until the end of the simulation), each embryo that received the health/stress signal shares a supportive signal in response. This supportive response scales with the cell’s health value and, modeling our experimental finding that only exposed embryos can provide the CEMA effect, only ‘exposed embryos’ send a supportive signal.

Incoming supportive signals are integrated as a weighted average. Additionally, we introduced a mechanism in which the healthier an embryo is (the closer its health status is to 1), the less it weights its neighbors’ input and increases its own signal’s weight; conversely, the unhealthier an embryo is, the more it weights its neighbor’s input and decreases its own weight. Importantly, the model is designed to ensure that an embryo has no ‘knowledge’ of its neighbors’ internal processes (we do not assume that embryos can sense the ground truth of the internal states of neighboring embryos). Instead, embryos can only send local signals that support (or destroy if the cell is noisy) update processes.