Contribution
Earlier models assumed clonal groups or a single form of division (often sterile helpers). This one lets the helping level of each phenotype and their ratio co-evolve under kin selection with relatedness as a free parameter, and finds which of the four forms of division (two generalists, generalist with sterile helper, generalist with pure reproductive, sterile helper with pure reproductive) can be stable.
Key insight
Specialization feeds on itself. The more a reproductive phenotype stops helping, the more the helper is selected to help, and the more the helper helps, the faster the reproductive is driven to pure reproduction. So whenever division of labour is favoured, the reproductive end goes to its extreme, and the helper end often follows, which makes some individuals wholly dependent on others’ help.
Assumptions
- Infinite population in fixed-size groups, island model, global competition; asexual reproduction, non-overlapping generations; rare mutations of small effect (weak selection).
- Relatedness R a fixed parameter (“open” model), except in SI 6.
- Fitness (Methods): W = p(1 − q1)[(1 − e) + e((1 − λ)q1^α + λ(PQ1^α + (1 − P)Q2^α))] + (1 − p)(1 − q2)[same with q2], where λ is the share of benefit going to others, α the returns to investment, e the trait’s essentiality, and capitals the neighbours’ mean traits. The neighbours’ term replaces a mean of powers by a power of means, which the authors justify by weak selection.
- Linear costs (relaxed in SI 4, not read).
Key results
- Forms. Only two forms of division are ever stable: pure reproductive with a generalist helper (0 < q1 < 1, q2 = 0) and pure reproductive with a sterile helper (q1 = 1, q2 = 0). Combinations with a generalist reproductive “never” were.
- Efficiency. α > 1 is necessary but not sufficient for division (SI 7.1); whenever division arises, α > 1 also stabilizes q2 = 0.
- Relatedness. For an essential trait (e = 1), R has no influence on whether division is favoured, “because the fitness benefit of being a pure reproductive is then exactly cancelled by the indirect fitness cost of not helping relatives”. For e < 1, higher R favours it. Higher R favours the sterile-helper form whether or not the trait is essential. R = 1 is not required for any form.
- Feedback. Individual-based simulations: with q2 fixed above 0, q1 evolves to an intermediate value; released, both go to the extremes within 10,000 generations (Fig. 4d legend).
- Evidence (Table 1). Of seven microbial examples, five are sterile helper with pure reproductive, one generalist with pure reproductive, one sterile helper with generalist (volvocine algae), the last a form the model does not allow, explained by a resource discontinuity outside the model.
Claims
| id | claim | strength | support |
|---|---|---|---|
| C1 | When division of labour is stable, one phenotype is a pure reproductive | strong, within the model | equilibrium analysis (SI 1–3, not read); gradient run agrees |
| C2 | α > 1 is necessary, not sufficient | moderate | SI 7.1 (not read) |
| C3 | For essential traits relatedness does not affect whether division evolves | strong, within the model | cancellation argument; gradient run agrees at α = 3 |
| C4 | Higher relatedness favours sterile helpers | moderate | model; consistent with comparative and experimental data cited |
| C5 | Division of labour drives major transitions, not the reverse | weak | interpretive; the model has no transition |
Concepts
- Essentiality e: the fraction of fitness that comes from cooperation rather than the asocial environment; e = 1 means no cooperation, no fitness.
- Trait sociality λ: the share of a helper’s benefit that goes to group neighbours rather than itself.
- Efficiency α: returns to investment in helping; α > 1 accelerating.
- Extreme specialization: a phenotype that does only one task, so others depend on it wholly.
Connections
It works in the Taylor–Frank neighbour-modulated fitness framework and Brown and Taylor’s multi-trait method. The authors’ review of division of labour in microbes (West & Cooper 2016) and Michod’s volvocine models are its context; none is registered. The record’s major-transitions readings (LIT-749) are the frame for C5.
Bearing on the record
- CLAIM-279 quotes A46: specialization “may improve local coordination while simultaneously increasing the organization’s vulnerability to the loss or capture of a specialist”. The paper says neither. It has no notion of coordination, and no loss or capture of a specialist is modelled or discussed. What it shows is dependence: “some individuals are completely dependent upon the helping behaviour of others”. A46’s vulnerability point is a fair inference from dependence, not the paper’s finding, and the claim’s note should say so.
- CLAIM-279’s thesis, that roles are regimes rather than kinds, gets partial support of a specific shape: phenotypes here are not given roles, they are two draws of one developmental lottery whose investments evolve. But the regimes do not stay intermediate; selection drives them to the extremes. A model of niche specialization among operators built on this logic would predict sharp, not graded, roles.
- QUESTION-083: the model’s division emerges with no group-level selection and no coded roles, from individual-level (kin) selection on one shared rule. It is a positive instance for the question’s first half, in a setting with relatedness and a single task.
- QUESTION-277 lists it as a sixth route to individuality. The route is asserted in the conclusion (“a driver, not a consequence”), not shown: the model ends at dependence and has no higher-level unit.
- No instruction for machine-learning practice; nothing for the anthology.
Limitations
- Table 1 contradicts Figure 5. Prediction 11 reads: “(a) If the helpers are sterile (q1 = 1), a higher efficiency benefits to specialisation (higher α) favours a lower proportion of helpers (lower p*). (b) Otherwise, there is no effect.” The Figure 5 legend says the reverse: with sterile helpers “higher α has no effect upon the optimal proportion”, with generalist helpers “a higher α decreases the optimal proportion of helpers”. One of the two is wrong in this manuscript; the published version was not checked.
- Table 1 numbers two predictions “5”.
- One trait, two phenotypes, open-model relatedness, no group-size effect except through assumed α(N), λ(N).
- The proofs and simulations are in the supplement, which was not read.
Open questions
- Does the feedback to extremes survive several cooperative traits, where a pure reproductive for one trait may help with another (the authors’ cyanobacteria example)?
- What would a model need for the dependence to become a new unit of selection, which C5 asserts?