Module 5

Colony Organization & Division of Labor

How individual response thresholds, age-based roles, and size polymorphism create a self-organized workforce

5.1 Response Threshold Model

One of the central questions in ant biology is: how does a colony allocate its workforce to different tasks (foraging, brood care, nest maintenance, defence) without any central coordinator? The response threshold model, developed by Bonabeau, Theraulaz, and Deneubourg (1996), provides an elegant answer based on individual variation.

The Model

Each ant $i$ has an internal response threshold $\theta_i$ for each task. When the stimulus intensity $s$ for a particular task (e.g., the amount of brood needing care, the number of hungry nestmates) exceeds the ant's threshold, the ant is likely to perform that task. The probability follows a sigmoidal (Hill-type) function:

\[\boxed{P_i(\text{task}) = \frac{s^n}{s^n + \theta_i^n}}\]

where $n$ is the Hill coefficient (steepness of the response). For$n = 2$ (commonly used), the response is:

\[P_i = \frac{s^2}{s^2 + \theta_i^2}\]

Key properties of this function:

When $s \ll \theta_i$: $P_i \approx 0$ (ant ignores the task)
When $s = \theta_i$: $P_i = 0.5$ (half-maximum response)
When $s \gg \theta_i$: $P_i \approx 1$ (ant always performs the task)

The threshold $\theta_i$ can be thought of as the ant's “sensitivity” to a particular task demand. Low-threshold ants respond to weak stimuli (they are the first to act), while high-threshold ants only respond when demand is extreme.

Population Heterogeneity and Task Allocation

The key insight is that thresholds vary across individuals. If$\theta_i$ is drawn from a distribution (e.g., Gamma distribution):

\[\theta_i \sim \text{Gamma}(k, \lambda), \qquad f(\theta) = \frac{\lambda^k \theta^{k-1} e^{-\lambda\theta}}{\Gamma(k)}\]

then the fraction of the colony performing a task at stimulus level $s$ is:

\[\bar{P}(s) = \int_0^\infty \frac{s^n}{s^n + \theta^n} f(\theta) \, d\theta\]

This creates a graded colony-level response: at low stimulus, only a few low-threshold specialists respond; as demand increases, more and more ants are recruited. This is far more efficient than a fixed allocation, because:

Resources are not wasted on tasks with low demand
Emergency demands can recruit the entire workforce
No central controller is needed — each ant independently decides based on local information

Multi-Task Extension

When multiple tasks compete for the same ant's attention, the probability of performing task $j$ is normalized across all tasks:

\[P_i(\text{task}_j) = \frac{s_j^n / (s_j^n + \theta_{ij}^n)}{\sum_{k=1}^{M} s_k^n / (s_k^n + \theta_{ik}^n)}\]

where $M$ is the number of tasks and $\theta_{ij}$ is ant $i$'s threshold for task $j$. An ant with a very low threshold for foraging but a high threshold for brood care will preferentially forage unless brood care demand becomes extremely high.

Reinforcement Learning

Thresholds are not fixed — they change with experience. An ant that performs a task repeatedly tends to lower its threshold for that task (practice effect):$\theta_i(t+1) = \theta_i(t) - \epsilon$. An idle ant's threshold drifts upward: $\theta_i(t+1) = \theta_i(t) + \delta$. This creates specialists that preferentially perform a narrow set of tasks, even if they started with identical thresholds — a spontaneous symmetry breaking.

5.2 Age Polyethism (Temporal Division of Labor)

In virtually all ant species studied, workers progress through a predictable sequence of tasks as they age. This pattern is called age polyethism or temporal division of labor. The typical progression is:

Age-Task Sequence

Days 1-10

Brood care (nursing)

Youngest workers stay deep in the nest, tending eggs, larvae, and pupae. They feed larvae with trophallactic secretions from their crop and labial glands.

Days 10-30

Nest maintenance

Middle-aged workers excavate new chambers, repair damaged tunnels, manage waste disposal, and process incoming food. They operate in the middle zones of the nest.

Days 30+

Foraging & defense

The oldest workers take on the most dangerous tasks: foraging outside the nest (high predation risk) and colony defense. This progression makes evolutionary sense: older workers have less remaining reproductive value.

Hormonal Control: Juvenile Hormone (JH)

The physiological driver of age polyethism is juvenile hormone (JH), produced by the corpora allata. JH titres increase with age:

\[\text{JH}(t) = \text{JH}_{\max} \cdot \left(1 - e^{-t/\tau_{\text{JH}}}\right)\]

where $\tau_{\text{JH}} \approx 15$–20 days is the characteristic time constant. The behavioural transition thresholds are:

Low JH (young): brood care behaviour dominant
Medium JH: transition to nest work
High JH (old): foraging behaviour activated

This mechanism is strikingly similar to the JH-mediated age polyethism in honeybees, suggesting deep evolutionary conservation of this hormonal control system across social Hymenoptera. Importantly, the JH clock can be accelerated or delayed by social signals: if foragers are experimentally removed, young workers mature faster to fill the gap.

Optimal Age Allocation Theory

From an evolutionary perspective, age polyethism can be understood as an optimization of colony fitness. Define the colony fitness function:

\[W = \sum_{j=1}^{M} B_j\!\left(\sum_{i} n_{ij}\right) - \sum_{j=1}^{M} \sum_{i} c_{ij} \, n_{ij}\]

where $B_j$ is the benefit function of task $j$,$n_{ij}$ is the number of age-class $i$ workers allocated to task $j$, and $c_{ij}$ is the cost (mortality risk) of age-class $i$ performing task $j$.

Maximizing $W$ subject to the constraint that each age class has a fixed number of workers yields the optimal allocation. When foraging mortality $c_{\text{forage}}$is high, the optimum assigns oldest workers to foraging because their remaining reproductive contribution (via inclusive fitness) is lowest:

\[\frac{\partial W}{\partial n_{ij}} = B_j' - c_{ij} = 0 \implies \text{oldest to riskiest tasks}\]

This is sometimes called the expendable worker hypothesis: colonies should “spend” their oldest, most dispensable workers on dangerous tasks.

5.3 Worker Size Polymorphism

Many ant species (especially in the subfamilies Myrmicinae and Formicinae) produce workers of dramatically different sizes. This physical caste polymorphism is distinct from age polyethism — it is determined during larval development and is permanent.

Size Castes

The most common system involves two or three discrete size classes:

Minor workers: the smallest and most numerous caste. Perform foraging, brood care, and general maintenance. Body mass typically 1–3 mg.
Major workers (soldiers): 5–10 times heavier than minors. Disproportionately large heads with powerful mandible muscles. Primary role: colony defence, seed milling (in harvester ants), and food storage (in repletes).
Supermajors: present in some species (e.g.,Pheidologeton diversus), these are the largest workers, sometimes 500 times the mass of the smallest minors. They specialize in defence and heavy-duty tasks.

Allometric Scaling

The relationship between head width and body mass follows an allometric power law:

\[\boxed{W_{\text{head}} = a \cdot W_{\text{body}}^\alpha}\]

where $\alpha$ is the allometric exponent and $a$ is a normalization constant. Three regimes exist:

$\alpha = 1$: Isometry — head grows proportionally with body. No size-based specialization.
$\alpha > 1$: Positive allometry— larger ants have disproportionately bigger heads. This is the soldier morphology. Measured values: $\alpha \approx 1.2$–1.5 for major workers.
$\alpha < 1$: Negative allometry— rare in ants but seen in some body parts (legs, antennae).

Taking the logarithm of both sides:

\[\log W_{\text{head}} = \log a + \alpha \log W_{\text{body}}\]

On a log-log plot, the allometric relationship appears as a straight line with slope$\alpha$. A break or change in slope indicates the transition between castes. In species with discrete castes (dimorphic or trimorphic workers), the log-log plot shows two or three distinct linear segments.

Optimal Caste Ratios

The ratio of minors to majors in a colony is not random — it is optimized by natural selection. Consider a simplified model with two castes and two tasks (foraging and defence):

\[W = F(n_{\text{min}}^F + \epsilon \cdot n_{\text{maj}}^F) + D(\epsilon \cdot n_{\text{min}}^D + n_{\text{maj}}^D) - c_{\text{min}} N_{\text{min}} - c_{\text{maj}} N_{\text{maj}}\]

where $F$ and $D$ are diminishing-returns benefit functions,$\epsilon < 1$ represents the reduced efficiency of the “wrong” caste at a task, and $c$ are maintenance costs. Because majors are larger,$c_{\text{maj}} \approx (m_{\text{maj}}/m_{\text{min}})^{0.75} \cdot c_{\text{min}}$(metabolic scaling).

Optimization shows that the fraction of majors should increase with predation pressure and decrease with food scarcity — matching empirical observations across species. Typical minor:major ratios range from 3:1 to 20:1 depending on ecological context.

5.4 Self-Organization of Labor

A fundamental misconception about ant colonies is that the queen directs worker activities. In reality, the queen's role is purely reproductive — she lays eggs and emits pheromones that signal her presence and fertility, but she does not coordinate or direct worker behaviour. Task allocation is entirely self-organized.

Gordon's Interaction Rate Model

Deborah Gordon's pioneering work on harvester ants (Pogonomyrmex barbatus) revealed that task switching is governed by interaction rate. When a forager returns to the nest, she briefly contacts (antennates) other ants near the entrance. The rate at which a waiting ant encounters returning foragers determines whether she will go out:

\[\boxed{P(\text{switch to foraging}) = \frac{r^n}{r^n + r_0^n}}\]

where $r$ is the rate of encounters with returning foragers (encounters per minute), $r_0$ is a threshold encounter rate, and $n$ is the Hill coefficient.

The logic is simple: if many foragers are returning (high $r$), it means foraging is going well, so more foragers should go out. If few are returning (low$r$), conditions may be bad (predators, no food), so stay inside.

This creates a negative feedback loop that stabilizes the foraging workforce: if too many foragers are out, the encounter rate at the entrance drops (fewer returning per unit time per forager), suppressing further activation. If too few are out, the encounter rate rises, activating more foragers.

Task Demand Feedback

More generally, the stimulus level $s_j$ for task $j$ is coupled to the number of ants currently performing that task:

\[\frac{ds_j}{dt} = \sigma_j - \beta_j \cdot n_j(t)\]

where $\sigma_j$ is the rate of stimulus accumulation (e.g., rate at which brood become hungry, rate of tunnel damage) and $\beta_j \cdot n_j$ is the rate at which workers reduce the stimulus by performing the task.

At steady state:

\[n_j^* = \frac{\sigma_j}{\beta_j}\]

The equilibrium number of workers on each task is proportional to the demand rate and inversely proportional to worker efficiency. This means the colony automatically reallocates labour when conditions change — a perturbation (e.g., tunnel damage increases $\sigma_{\text{maint}}$) causes more workers to switch to maintenance until the new equilibrium is reached.

Perturbation Experiments

The self-organized nature of task allocation is demonstrated by perturbation experiments:

Remove foragers: within hours, nest workers begin foraging (younger workers accelerate their behavioural maturation)
Remove nurses: foragers can revert to brood care, though less efficiently (behavioural regression is possible but imperfect)
Increase food availability: the foraging workforce expands, drawing workers from maintenance and other tasks
Simulate predator attack: soldiers are recruited from interior tasks; if soldiers are depleted, large minors take over defence

These experiments confirm that the colony functions as a homeostatic system with distributed control: no single ant knows the colony's state, but collective behaviour maintains optimal allocation.

Analogy to Immune System

The ant colony's division of labour is strikingly analogous to the vertebrate immune system: both use threshold-based activation of specialized agents (T cells / workers), both feature positive feedback (clonal expansion / pheromone recruitment), and both achieve robust system-level responses without central coordination. This convergence suggests deep principles of distributed biological computation.

5.5 Colony Organization Chart

Schematic of colony organization showing the queen at the centre, age-based task zones radiating outward, and size castes with allometric head scaling.

5.6 Simulation: Response Thresholds, Task Allocation & Allometry

This simulation demonstrates four aspects of colony organization: (1) the sigmoidal response threshold function for different individual thresholds, (2) dynamic task allocation of 1000 ants responding to changing demands, (3) age polyethism showing the transition from brood care to foraging with age, and (4) allometric head-width scaling across minor, major, and supermajor castes.

Python

script.py163 lines

import numpy as np
import matplotlib
matplotlib.use('Agg')
import matplotlib.pyplot as plt

np.random.seed(42)

fig, axes = plt.subplots(2, 2, figsize=(14, 12))
fig.patch.set_facecolor('#0a0a0a')
for ax in axes.flat:
    ax.set_facecolor('#0a0a0a')
    for spine in ax.spines.values():
        spine.set_color('#444')
    ax.tick_params(colors='#aaa', labelsize=9)

# --- Panel 1: Response Threshold Model ---
ax1 = axes[0, 0]
s = np.linspace(0, 10, 200)
thresholds = [1.0, 2.0, 4.0, 6.0, 8.0]
colors_t = ['#f87171', '#facc15', '#4ade80', '#38bdf8', '#c084fc']
hill_n = 2

for theta, col in zip(thresholds, colors_t):
    P = s**hill_n / (s**hill_n + theta**hill_n)
    ax1.plot(s, P, '-', color=col, linewidth=2, label=f'theta = {theta:.0f}')

ax1.set_title('Response Threshold Model', color='#f87171', fontsize=12, fontweight='bold')
ax1.set_xlabel('Stimulus intensity (s)', color='#ccc', fontsize=10)
ax1.set_ylabel('P(perform task)', color='#ccc', fontsize=10)
ax1.legend(fontsize=9, facecolor='#111', edgecolor='#444', labelcolor='#ccc')
ax1.grid(True, alpha=0.15)
ax1.set_ylim(-0.05, 1.05)

ax1.text(0.5, 0.15, r'$P(s) = s^n / (s^n + \theta^n)$, n=2',
    transform=ax1.transAxes, color='#888', fontsize=10, ha='center',
    bbox=dict(boxstyle='round', facecolor='#1a1a1a', edgecolor='#444'))

# --- Panel 2: Task Allocation Dynamics ---
ax2 = axes[0, 1]
n_ants = 1000
n_tasks = 3
task_names = ['Foraging', 'Brood care', 'Nest maintenance']
task_colors = ['#f87171', '#4ade80', '#38bdf8']

thresholds_pop = np.random.gamma(shape=3, scale=1.5, size=(n_ants, n_tasks))

n_steps = 100
stimuli = np.zeros((n_steps, n_tasks))
stimuli[:, 0] = np.concatenate([np.linspace(1, 3, 30), np.linspace(3, 8, 20),
    np.linspace(8, 4, 25), np.linspace(4, 2, 25)])
stimuli[:, 1] = np.concatenate([np.linspace(5, 5, 30), np.linspace(5, 2, 20),
    np.linspace(2, 6, 25), np.linspace(6, 4, 25)])
stimuli[:, 2] = np.concatenate([np.linspace(3, 3, 30), np.linspace(3, 3, 20),
    np.linspace(3, 7, 25), np.linspace(7, 3, 25)])

allocation = np.zeros((n_steps, n_tasks))
for t in range(n_steps):
    probs = np.zeros((n_ants, n_tasks))
    for task in range(n_tasks):
        s_val = stimuli[t, task]
        probs[:, task] = s_val**hill_n / (s_val**hill_n + thresholds_pop[:, task]**hill_n)
    probs_sum = probs.sum(axis=1, keepdims=True)
    probs_sum[probs_sum == 0] = 1
    probs_norm = probs / probs_sum
    for ant in range(n_ants):
        chosen = np.random.choice(n_tasks, p=probs_norm[ant])
        allocation[t, chosen] += 1

time_steps = np.arange(n_steps)
ax2.stackplot(time_steps, allocation[:, 0], allocation[:, 1], allocation[:, 2],
    colors=task_colors, alpha=0.7, labels=task_names)
ax2.set_title('Task Allocation Dynamics (1000 ants)', color='#f87171', fontsize=12, fontweight='bold')
ax2.set_xlabel('Time step', color='#ccc', fontsize=10)
ax2.set_ylabel('Number of ants', color='#ccc', fontsize=10)
ax2.legend(fontsize=9, facecolor='#111', edgecolor='#444', labelcolor='#ccc', loc='upper right')
ax2.grid(True, alpha=0.15)

ax2_twin = ax2.twinx()
for task in range(n_tasks):
    ax2_twin.plot(time_steps, stimuli[:, task], '--', color=task_colors[task], linewidth=1.5, alpha=0.5)
ax2_twin.set_ylabel('Stimulus level (dashed)', color='#888', fontsize=9)
ax2_twin.tick_params(colors='#666', labelsize=8)

# --- Panel 3: Age Polyethism ---
ax3 = axes[1, 0]
ages = np.arange(0, 90)

brood_care = np.exp(-((ages - 10)**2) / (2*8**2))
maintenance = np.exp(-((ages - 35)**2) / (2*10**2))
foraging = 1 / (1 + np.exp(-(ages - 50) / 5))

brood_care = brood_care / (brood_care + maintenance + foraging + 0.01)
maintenance = maintenance / (brood_care + maintenance + foraging + 0.01)
foraging_frac = foraging / (brood_care + maintenance + foraging + 0.01)

total = brood_care + maintenance + foraging_frac
brood_care /= total
maintenance /= total
foraging_frac /= total

ax3.fill_between(ages, 0, brood_care, color='#4ade80', alpha=0.6, label='Brood care')
ax3.fill_between(ages, brood_care, brood_care + maintenance, color='#38bdf8', alpha=0.6, label='Nest maintenance')
ax3.fill_between(ages, brood_care + maintenance, 1, color='#f87171', alpha=0.6, label='Foraging')

ax3.axvline(x=20, color='white', linewidth=0.8, linestyle=':', alpha=0.3)
ax3.axvline(x=45, color='white', linewidth=0.8, linestyle=':', alpha=0.3)
ax3.text(10, 1.05, 'Young', color='#4ade80', fontsize=10, ha='center')
ax3.text(32, 1.05, 'Middle', color='#38bdf8', fontsize=10, ha='center')
ax3.text(65, 1.05, 'Old', color='#f87171', fontsize=10, ha='center')

ax3.set_title('Age Polyethism', color='#f87171', fontsize=12, fontweight='bold')
ax3.set_xlabel('Worker age (days)', color='#ccc', fontsize=10)
ax3.set_ylabel('Fraction of time', color='#ccc', fontsize=10)
ax3.legend(fontsize=9, facecolor='#111', edgecolor='#444', labelcolor='#ccc', loc='center right')
ax3.set_xlim(0, 89)
ax3.set_ylim(0, 1.12)
ax3.grid(True, alpha=0.15)

# --- Panel 4: Allometric Head Width ---
ax4 = axes[1, 1]
n_workers = 300
body_mass = np.random.lognormal(mean=0.5, sigma=0.8, size=n_workers)
body_mass = np.clip(body_mass, 0.3, 30)

minor_mask = body_mass < 3
major_mask = (body_mass >= 3) & (body_mass < 12)
super_mask = body_mass >= 12

alpha_minor = 0.8
alpha_major = 1.3
alpha_super = 1.5

head_width = np.zeros_like(body_mass)
head_width[minor_mask] = 0.8 * body_mass[minor_mask]**alpha_minor + np.random.normal(0, 0.1, minor_mask.sum())
head_width[major_mask] = 0.6 * body_mass[major_mask]**alpha_major + np.random.normal(0, 0.15, major_mask.sum())
head_width[super_mask] = 0.45 * body_mass[super_mask]**alpha_super + np.random.normal(0, 0.3, super_mask.sum())
head_width = np.clip(head_width, 0.3, 30)

ax4.scatter(body_mass[minor_mask], head_width[minor_mask], c='#4ade80', s=20, alpha=0.6, label=f'Minor (alpha={alpha_minor})')
ax4.scatter(body_mass[major_mask], head_width[major_mask], c='#facc15', s=40, alpha=0.6, label=f'Major (alpha={alpha_major})')
ax4.scatter(body_mass[super_mask], head_width[super_mask], c='#f87171', s=80, alpha=0.6, label=f'Super (alpha={alpha_super})')

bm_fit = np.linspace(0.3, 30, 200)
ax4.plot(bm_fit, 0.7*bm_fit**1.0, '--', color='white', linewidth=1, alpha=0.3, label='Isometry (alpha=1)')
ax4.plot(bm_fit, 0.6*bm_fit**1.3, '-', color='#facc15', linewidth=1.5, alpha=0.4)

ax4.set_xscale('log')
ax4.set_yscale('log')
ax4.set_title('Worker Size Polymorphism: Head Allometry', color='#f87171', fontsize=12, fontweight='bold')
ax4.set_xlabel('Body mass (mg)', color='#ccc', fontsize=10)
ax4.set_ylabel('Head width (mm)', color='#ccc', fontsize=10)
ax4.legend(fontsize=8, facecolor='#111', edgecolor='#444', labelcolor='#ccc')
ax4.grid(True, alpha=0.15, which='both')

ax4.text(0.05, 0.05, r'$W_{head} \propto W_{body}^\alpha$' + '\nalpha > 1: positive allometry',
    transform=ax4.transAxes, color='#888', fontsize=9, va='bottom',
    bbox=dict(boxstyle='round', facecolor='#1a1a1a', edgecolor='#444'))

plt.tight_layout()
plt.savefig('output.png', dpi=150, bbox_inches='tight', facecolor='#0a0a0a')
plt.close()

Click Run to execute the Python code

Code will be executed with Python 3 on the server

Key Observations

Response thresholds: low-threshold ants (red curve) respond at weak stimulus levels, while high-threshold ants (purple) require strong stimuli. This heterogeneity is the key to flexible task allocation.
Task dynamics: when foraging demand spikes (step ~30), ants are rapidly recruited from other tasks. The dashed lines show stimulus levels; the filled areas show workforce distribution. Notice the response lag.
Age polyethism: the smooth transition from brood care (green, young) through maintenance (blue, middle) to foraging (red, old) matches empirical observations across many ant species.
Head allometry: on the log-log plot, majors (yellow, $\alpha = 1.3$) and supermajors (red, $\alpha = 1.5$) have steeper slopes than isometry (white dashed line), confirming positive allometry of head size.

References

Bonabeau, E., Theraulaz, G., & Deneubourg, J.-L. (1996). Quantitative study of the fixed threshold model for the regulation of division of labour in insect societies. Proceedings of the Royal Society B, 263(1376), 1565–1569.
Gordon, D. M. (2010). Ant Encounters: Interaction Networks and Colony Behavior. Princeton University Press.
Wilson, E. O. (1978). Division of labor in fire ants (Solenopsis invicta) based on physical castes: a review. Journal of the Kansas Entomological Society, 51(4), 615–636.
Robinson, G. E. (1992). Regulation of division of labor in insect societies. Annual Review of Entomology, 37, 637–665.
Beshers, S. N., & Fewell, J. H. (2001). Models of division of labor in social insects. Annual Review of Entomology, 46, 413–440.
Hölldobler, B., & Wilson, E. O. (1990). The Ants. Harvard University Press.
Gordon, D. M. (1996). The organization of work in social insect colonies. Nature, 380, 121–124.
Tschinkel, W. R. (1988). Colony growth and the ontogeny of worker polymorphism in the fire ant, Solenopsis invicta. Behavioral Ecology and Sociobiology, 22(2), 103–115.
Mersch, D. P., Crespi, A., & Keller, L. (2013). Tracking individuals shows spatial fidelity is a key regulator of ant social organization. Science, 340(6136), 1090–1093.

←M4: Nest Architecture M6: Collective Intelligence→