Topological constraints in early multicellularity favor reproductive division of labor.

Reproductive division of labor (e.g. germ-soma specialization) is a hallmark of the evolution of multicellularity, signifying the emergence of a new type of individual and facilitating the evolution of increased organismal complexity. A large body of work from evolutionary biology, economics, and ecology has shown that specialization is beneficial when further division of labor produces an accelerating increase in absolute productivity (i.e. productivity is a convex function of specialization). Here we show that reproductive specialization is qualitatively different from classical models of resource sharing, and can evolve even when the benefits of specialization are saturating (i.e. productivity is a concave function of specialization). Through analytical theory and evolutionary individual-based simulations, we demonstrate that reproductive specialization is strongly favored in sparse networks of cellular interactions that reflect the morphology of early, simple multicellular organisms, highlighting the importance of restricted social interactions in the evolution of reproductive specialization.

Introduction

The evolution of multicellularity set the stage for unprecedented increases in organismal complexity (Szathmáry and Smith, 1995; Knoll, 2011). A key factor in the remarkable success of multicellular strategies is the ability to take advantage of within-organism specialization through cellular differentiation (Queller and Strassmann, 2009; Brunet and King, 2017; Cavalier-Smith, 2017). Reproductive specialization, which includes both the creation of a specialized germ line during ontogeny (as in animals and volvocine green algae) and functional differentiation into reproductive and non-reproductive tissues (as in plants, green and red macroalgae, and fungi), may be especially important (Cooper and West, 2018; Michod et al., 2006; Ispolatov et al., 2012; Solari et al., 2013; Michod, 2007; West et al., 2015). Reproductive specialization is an unambiguous indication that biological individuality rests firmly at the level of the multicellular organism (Michod, 1999; Folse and Roughgarden, 2010), and is thought to play an important role in spurring the evolution of further complexity by inhibiting within-organism (cell-level) evolution (Buss, 1988) and limiting reversion to unicellularity (Libby and Ratcliff, 2014). Despite the central importance of reproductive specialization, its origin and further evolution during the transition to multicellularity remain poorly understood (McShea, 2000).

The origin of specialization has long been of interest to evolutionary biologists, ecologists, and economists. A large body of theory from these fields shows that specialization pays off only when it increases total productivity, compared to the case where each individual simply produces what they need (Szathmáry and Smith, 1995; Smith and Szathmáry, 1997; Goldsby et al., 2012; Corning and Szathmáry, 2015; Hidalgo and Hausmann, 2009; Boza et al., 2014; Taborsky et al., 2016; Page et al., 2006; Rueffler et al., 2012; Szekely et al., 2013; Findlay, 2008; Amado et al., 2018). Certain types of trading arrangements maximize the benefits of specialization; highly reciprocal interactions, which facilitate exchange between complementary specialists, amplify cooperation (Allen et al., 2017; Pavlogiannis et al., 2018). Still, previous work finds that even when groups grow in an ideal spatial arrangement, increased specialization and trade is only favored by natural selection when productivity increases as an accelerating function of the degree of specialization (i.e., productivity is a convex, or super-linear, function of the degree of specialization). Conversely, saturating functional returns (i.e. productivity is a concave, or sub-linear, function of the degree of specialization) should inhibit the evolution of specialization (Cooper and West, 2018; Michod et al., 2006; Ispolatov et al., 2012; Solari et al., 2013; Michod, 2007; West et al., 2015).

Reproductive specialization differs from classical models of trade in several key respects. Trade between germ (reproductive) and somatic (non-reproductive) cells is intrinsically asymmetric, because the cooperative action, multicellular replication, is not a product that is shared evenly. Selection acts primarily on the fitness of the multicellular group as a whole (Folse and Roughgarden, 2010). As a result, optimal specialization can result in behaviors that reduce the short-term fitness of some cells within the multicellular group (Michod et al., 2006; Michod, 2007), often manifest as reproductive altruism.

Understanding the evolution of cell-cell trade, a classic form of social evolution (Kirk, 2005), requires understanding the extent of between-cell interactions. Network theory has proven to be an exceptionally powerful and versatile technique for analyzing social dynamics (Wey et al., 2008; Lieberman et al., 2005), and indeed, is uniquely well suited to understanding the evolution of early multicellular organisms. When cells adhere through permanent bonds, sparse network-like bodies (i.e. filaments and trees) often result (Amado et al., 2018). This mode of group formation is not only common today among simple multicellular organisms (Umen, 2014; Claessen et al., 2014), but is the dominant mode of group formation in the lineages evolving complex multicellularity (i.e. plants, red algae, brown algae, and fungi, but not animals).

In this paper, we develop and investigate a model for how the network topology of early multicellular organisms affects the evolution of reproductive specialization. We find that under a broad class of sparse networks, complete functional specialization can be adaptive even when returns from dividing labor are saturating (i.e. concave/sub linear). Sparse networks impose constraints on who can share with whom, which counterintuitively increases the benefit of specialization (McShea, 2000). By dividing labor, multicellular groups can capitalize on high between-cell variance in behavior, ultimately increasing group-level reproduction. Further, we consider group morphologies that naturally arise from simple biophysical mechanisms and show that these morphologies strongly promote reproductive specialization. Our results show that reproductive specialization can evolve under a far broader set of conditions than previously thought, lowering a key barrier to major evolutionary transitions.

Model

Reproductive specialization can be modeled as the separation of two key fitness parameters, those related to either viability or fecundity, into separate cells within the multicellular organism (Michod, 2006; Folse and Roughgarden, 2010). The dichotomy of viability versus fecundity was originally used by Michod, 2006 to partition components of cellular fitness into actions that contribute to keeping a cell alive (viability), and actions that directly contribute to reproduction (fecundity). Multicellular organisms often have evolved to divide labor along these two lines (i.e. reproduction by germ cells and survival provided by somatic cells), while their unicellular ancestors had to do both. We define viability as activities keeping the cell alive (e.g. investing in cellular homeostasis or behaviors that improve survival), and fecundity as activities involved in cellular reproduction.

At the cellular level, there appears to be a fundamental asymmetry in how viability effort and fecundity effort can be shared among cells: while multicellular organisms readily evolve differentiated cells that are completely reliant on helper cells (i.e. glial cells that support neurons in animals or companion cells that support sieve tube cells in plants), no cell can directly share its ability to reproduce. To better understand the intuition behind this, consider a cell that elongates prior to fission. This cell must grow to approximately twice its original length. Two cells cannot elongate by 50% and then combine their efforts; elongation is an intrinsically single cell effort. We thus use a model in which viability can be shared across connected cells, but fecundity cannot be shared (note, in order to test the sensitivity of our predictions to this assumption, in a later section we will consider the more general case in which viability and fecundity can both be shared, but by different amounts).

We consider a model of multicellular groups composed of clonal cells that each invest resources into viability and fecundity. Because there is no within-group genetic variation, within-group evolution is not possible, though selection can act on group-level fitness differences. Specifically, we consider the pattern of cellular investment in fecundity and viability, and their sharing of these resources with neighboring cells within the group, to be the result of a heritable developmental program. Thus, selection is able to act on the multicellular fitness consequences of different patterns of cellular behavior within the group. We let v denote each cell’s investment into viability, and b denote each cell’s investment into fecundity. Each cell's total investment is constrained so that . However, a cell's return on its investment is in general nonlinear. Here, we let α represent the ‘return on investment exponent': by tuning α above and below 1.0, we can simulate conditions with accelerating and saturating (i.e. convex and concave, or super- and sub-linear) returns on investment, respectively. We let and represent a cell’s return on viability and fecundity investments, respectively. Following Michod, 2006; Michod and Roze, 1997, we calculate a cell’s reproductive output as a multiplicative function of and (thus, both functions must be positive for a cell to grow). A single cell’s reproduction rate is . At the group level, fitness is the total contribution of all cells in the group toward the production of new groups (i.e. group level reproduction). The group level fitness is thus the sum of over all cells.

Finally, cells may share the products of their investment in viability with other cells to whom they are connected.For a given group, the details about who may share with whom, and how much, is encoded in a weighted adjacency matrix . The element defines what proportion of viability returns cell i shares with cell j. Cells cannot give away all of their viability returns, as they would no longer be viable; mathematically, we count a cell among its neighbors and thus ensure that they always ‘share’ a positive portion of viability returns with themselves, so that . Furthermore, since a cell cannot share more viability returns than the total it possesses, we have for a group of N cells. For the networks we consider, each cell takes a fraction β of its viability returns and shares that fraction equally among all of its neighbors (including itself), and keeps the rest of its returns for itself. Therefore cell i keeps a total fraction of of its returns for itself and gives to each of its non-self neighbors. In other words, , if cells i and j are connected, and if cells i and j are not connected. This means the total amount of returns kept by cell i depends on both the network topology and β. When there is no sharing, and when cells share everything equally among all connections and themselves. We refer to β as interaction strength. A given group topology (unweighted adjacency matrix) and β completely specify .

Within a group of N cells, the overall returns on viability that a given cell enjoys, then, comprises its own returns as well as whatever is shared with it by other members of the group. This can be written as , or equivalently, . Note that this is a column sum, since it describes the total incoming viability returns a cell receives as a result of its own effort and trade with neighboring cells. Therefore, we write the group level reproduction rate (i.e. the group fitness) for a group of N cells aswhere all three of the above equations are equivalent. We investigate evolutionary outcomes under this definition of group level fitness for groups with different topologies (who shares with whom), and in scenarios with various return on investment exponents α.

Results

Fixed resource sharing

We first consider cases wherein cells within a group share across fixed intercellular interactions. In each case we vary the return on investment exponent, α, between 0.5 and 1.5, and the interaction strength, β, between 0.0 and 1.0, both in increments of 0.1. For each combination of topology, α, and β, the group investment strategy ( for all i) was allowed to evolve for 1000 generations.

We begin with simple topologies: groups with no connections and groups that are maximally connected. They represent, respectively, the case in which all cells within the group are autonomous and the case in which every cell interacts with all others (i.e. a ‘well-mixed’ group). In the absence of interactions, cells cannot benefit from functions performed by others and therefore must perform both functions v and b; hence specialization is not favored, and does not evolve. In the fully connected case, a high degree of specialization is observed for many values of α and β (Figure 1a). Consistent with classic results (Cooper and West, 2018; Michod et al., 2006; Ispolatov et al., 2012; Solari et al., 2013; Michod, 2007; West et al., 2015), specialization is only achieved in the fully connected case for .

Figure 1.

Schematic of topology for a simplified ten cell group (first row), and mean specialization as a function of specialization power α and interaction strength β across the entire population.

(A) When each cell in the group is connected to all others, specialization is favored only when . (B) For the nearest neighbor topology, specialization is favorable for a wider range of parameters, including for some values of . Specifically, specialization is advantageous when . (C) Connecting alternating specialists creates a bipartite graph which maximizes the benefits of specialization and the range of parameters for which it is advantageous. In this case, specialization is favorable wherever . The red curves represent analytical predictions for , the lowest value of α for which complete generalization is disfavored, and the orange vertical lines are at to guide the eye. While analysis shows that some degree of specialization must occur in the regime upward and to the right of the red curves, simulations reveal that when complete generalization is disfavored complete specialization is favored in these networks.

Next, we consider a simple sparse network in which each cell within a group is connected to only two other cells, forming a complete ring (Figure 1b); we refer to this as the neighbor network. Surprisingly, preventing trade between most cells encourages division of labor. We find that specialization evolves even when , that is, when the returns on investment are saturating or concave. In our simulations, this topology leads to alternating specialists in viability and fecundity (Figure 1b). Analytically, we find that this topology always favors at least some degree of specialization whenever .

We next study a network with cells that can be separated into two disjoint sub-groups, where every edge of the network connects a cell in one sub-group to a cell in the other sub-group and no within sub-group connections exist, that is, a bipartite graph (Figure 1c). We refer to the specific network structure in Figure 1c as the ‘balanced bipartite’ network. We find that specialization evolves even when , similar to the neighbor network. However, we find that specialization evolves for a wider range of α and β values for the balanced bipartite network than for the neighbor network.

We can analytically determine under what conditions complete generalization is optimal. The complete generalist investment strategy is where every cell in the group invests equally into viability and fecundity, defined as: for all i. For these simple topologies, the complete generalist strategy is either a maximum or a saddle point, depending on the values of α and β. Complete generalization is only favored when the Hessian evaluated at the generalist investment strategy is negative definite, that is, all of its eigenvalues are negative. The largest eigenvalues of the Hessian for the complete, neighbor, and balanced bipartite networks are , , and , respectively. When α and β are chosen so that the largest eigenvalue becomes non-negative, complete generalization cannot maximize group fitness.

While we have not analytically shown where the fitness maximum occurs in cases where the generalist strategy becomes a saddle point, evolutionary simulations (Figure 1) suggest that when complete generalization is not a fitness maximum, a high degree of (or even complete) specialization typically does maximize fitness.

In all cases in which complete specialization is achieved in evolutionary simulations, terms for viability specialists go to zero, as they cannot reproduce on their own. Furthermore, the fecundity specialists are entirely reliant on the viability specialists for their survival; if viability sharing were suddenly prevented, their terms would also be zero. This amounts to complete reproductive specialization (Cooper and West, 2018; Kirk, 2005; Michod, 2006).

Evolving resource sharing

Until now, sharing has been included in every intercellular interaction within groups. Here, we consider the case in which there is initially no sharing, and sharing must evolve along with specialization. These simulations begin with no resource sharing (i.e. ); during every round, each group in the population has a 2% chance that a mutation will impact its developmental program, and the β value for one of its cells will change. The new β value is chosen from a truncated Gaussian with standard deviation of 10% of the mean, centered on the current value. Whatever is not retained is shared equally across all interactions, including the self term.

Evolutionary simulation results are similar to those from the fixed-sharing model (Appendix 1—figure 1). Saturating specialization (i.e. specialization despite a concave return function) still occurs for the neighbor and balanced bipartite topologies. Thus, for both fixed and evolved resource sharing, we observe specialization for the largest range of parameters (including ) not when the group is maximally connected, but rather when connections are fairly sparse. Therefore, a sparse group topology constitutes a cooperation-prone physical substrate that can favor the evolution of cellular.

Appendix 1—figure 1.

Evolution of resource sharing.

(A) Initially, individuals do not share resources; however, they may evolve to do so via random mutations. Here, the mean specialization of the fittest of 100 groups each with 10 cells after 100,000 steps is plotted as a function of specialization power. Error bars are standard deviations across 10 replicates. Blue is the fully connected network, red is the neighbor network, and green is the balanced bipartite topology. (B-D) The final distribution of specialization values for individual cells in fully connected (B), nearest-neighbor (C), and balanced bipartite topologies (D). The color of cells in B-D represents their degree of specialization, as indicated in the scale bar.

As an example of the benefit of evolving sharing, consider that the maximum fitness according to Equation 1 for a group of N disconnected cells scales as . On the other hand, for the balanced bipartite network with a complete specialization strategy (i.e. ), the fitness scales as . The ratio of these fitnesses is , where the approximation is for large N. So for larger groups and when , if a group can evolve resource sharing (i.e. letting and adopting the specialist investment strategy) its maximum fitness will increase.

Benefit of specialization

We now consider a simple example to highlight why specialization can be adaptive despite saturating (i.e., concave) returns from trade. Consider groups of four cells, connected via the nearest-neighbor topology (i.e. in a ring). We directly calculate the group-level fitness of generalists and specialists for two scenarios: and by summing the contributions of each cell within these groups (Figure 2). In this simple scenario, reproductive specialization strongly increases group fitness (33% for and 16% for ).

Figure 2.

To explore how specialization can be favored by the nearest-neighbor topology, we compare the fitness of a four member system when cells are (A) generalists and (B) specialists.

We first consider the case of linear functional returns (). For the case of generalists (A), each cell receives as much viability as it shares, and all nodes contribute equally to the fitness of the group. Therefore, the fitness of the group is . For the case of specialists, however, the viability specialist cells (blue) have , while the fecundity specialist cells have nonzero due to the fact that they receive of each viability specialist’s output. Thus the fitness of the group is . Thus, fitness is higher for the group of specialists, so specialization is favored. For , the fitness of generalists is 1.15, and the fitness of specialists is 1.33. Thus, even though the returns on investment are saturating (i.e. concave), specialization is favored.

The benefit of specialization in neighbor networks increases with group size. For a ring of size N, fitness under the specialist strategy is . For a ring of generalists the fitness is . Therefore, whenever , the ring of complete specialists enjoys a greater fitness than the ring of complete generalists. Again, note that complete generalization becomes disfavored when , so there is a narrow regime where during which neither complete generalization nor complete specialization is optimal. Numerical optimization and evolutionary simulations suggest that even in this region, however, the specialization score of the optimal strategy is large (Figure 1).

Effect of sparsity

Surprisingly, saturating specialization appears to be the rule, rather than the exception, for sparsely connected graphs. We investigated Erdős-Rényi random graphs with varying degrees of connectivity to systematically examine the relationship between sparsity and the value of α at which specialization is favored. We find that many randomly assembled graphs obtain maximum fitness through complete reproductive specialization even when α is below 1 (Figure 3b,c). It is only at the extremes of sparsity and connectivity (near the fully connected or fully unconnected points) that generalists maintain superior fitness for all values of . We further show that this general trend is independent of the size of a group; saturating specialization is favorable for groups of size , , and . When network connectivity is at its minimum, the group consists solely of isolated cells that cannot interact. Under these conditions generalists are favored. Similarly, at maximum connectivity every cell interacts with every other cell. Under these conditions generalists are favored unless . However, when connectivity is small but not zero, specialization arises most readily. We conjecture that the troughs in Figure 3b, where specialization occurs for the lowest values of α, occur when connectivity is just large enough so that the existence of a spanning tree is more likely than not.

Figure 3.

Sparsity encourages specialization.

Heat maps showing conditions that favor specialists (white) and generalists (black) for nearest neighbor topologies (A, left) and randomly generated graphs with the same connectivity as nearest neighbor topologies (A, right). Specialization is adaptive on a neighbor network for ; random networks with the same mean connectivity as the nearest neighbor topology behave similarly. (B) The sparsity of a random graph affects how likely it is to favor specialization. We numerically maximize fitness for random graphs of size (left), (middle), and (right) at different levels of sparsity, and subsequently measure the specialization of the fitness maximizing investment strategy. The horizontal axis is the fraction of possible connections present ranging from 0 (none) to 1 (all). The vertical axis is the specialization power α, and the colormap shows mean specialization.

Filaments and trees

Sparse topologies like the neighbor network configuration have significant biological relevance, and direct ties to early multicellularity. The first step in the evolution of multicellularity is the formation of groups of cells (Szathmáry and Smith, 1995; Kirk, 2005; Willensdorfer, 2008; Bonner, 1998; Fairclough et al., 2010). Simple groups readily arise through incomplete cell division, forming either simple filaments (Figure 4a) or tree-like morphologies (Figure 4b; Bengtson et al., 2017b; Droser and Gehling, 2008; Berman-Frank et al., 2007; Ratcliff et al., 2012). Filament topologies have been widely observed in independently-evolved simple multicellular organisms, from ancient fossils of early red algae (Butterfield, 2000; Figure 4a) to extant multicellular bacteria (Claessen et al., 2014) and algae (Umen, 2014). Branching multicellular phenotypes have also been observed to readily evolve from baker’s yeast (Ratcliff et al., 2015; Figure 4b), and are reminiscent of ancient fungus-like structures (Bengtson et al., 2017a) and early multicellular fossils of unknown phylogenetic position from the early Ediacaran (Droser and Gehling, 2008).

Figure 4.

Simple multicellular organisms with sparse topologies.

We show two examples of simple multicellular organisms with linear and branched topologies. The image in (A) is a fossilized rhodophyte specimen of Bangiomorpha pubescens, courtesy of Prof. Nicholas Butterfield (see e.g. Butterfield, 2000); the image in (B) is a confocal image of ‘snowflake yeast’ showing cell volumes in blue and cell-cell connections in green; the image in (C) is an epifluorescence image of individual yeast cells from a planktonic culture, with the same staining technique as in (B). Scale bars in pictures = 10 µm. Panels include cartoons depicting simplified topologies. Topologically similar to the two-neighbor configuration, these configurations yield similar simulation results. Specialization is plotted as a function of α. Solid (A) and blue (B) vertical lines (A and B) indicate analytical solutions for the transition point where the Hessian evaluated at stops being negative definite, that is, ; dotted lines indicate roughly where the simulation curves cross specialization of 0.5, that is, the 'true' transition value of α where specialization becomes favored. (C) In contrast, for a well-mixed group with fully connected topology, , indicating specialization only occurs when there are accelerating returns on investment. (D) To further explore trees and filaments we analytically solved for for various types of trees and filaments of different sizes. is plotted versus group size for several topologies. This is a proxy measure of how amenable a network structure is to specialization.

Simulations of populations of groups with filamentous and branched topologies reveal that specialization is indeed favored in the sub-linear regime (Figure 4a and b) ; conversely, sub-linear specialization is never observed for fully connected topologies (Figure 4c). While the generalist strategy is never a critical point for these networks (which have , see Materials and methods), we conjecture that there is a nearby critical point which maximizes fitness at small values of α and becomes unstable at larger values of α. We introduce a new metric, , defined as the value of α such that the largest (least negative) eigenvalue of the Hessian evaluated at the complete generalist strategy is zero when . For topologies in which each member has the same number of neighbors, is a critical value at which generalization is no longer an optimal strategy. However, even for groups where the number of neighbors for each cell varies, we can still use as a proxy for how amenable a topology is to saturating specialization. The smaller , the more specialization is likely to be favored. We plot vertical lines where (solid lines in Figure 4(a) Figure 4(b)), and dotted lines to indicate roughly where the simulation curves cross specialization of 0.5. These results show that, for these topologies, acts as an effective metric for how amenable a network is to saturating specialization. This metric only depends on topology and can in principle be calculated analytically given any network. We examined the value of as filaments and a variety of tree-like structures grow larger, and find that specialization becomes more strongly favored (Figure 4D ). While group size has no effect on specialization for some topologies, like the neighbor network, filaments and trees all see a decrease in as group size increases; eventually plateaus once groups are larger than a few tens of cells. Simple and easily accessible routes to multicellular group formation can readily evolve in response to selection for organismal size (Ratcliff et al., 2012), and this process may also strongly favor the evolution of cellular differentiation (McCarthy and Enquist, 2005; Heim et al., 2017; McClain and Boyer, 2009; Bonner, 1998).

Mean field model

Finally, to capture some general principles underlying this phenomenon, we consider a mean-field model with N cells (N >> 1), each of which is connected to z other cells. For simplicity we consider the case in which and . We pick as at this point, if the fitness of specialists is greater than that of generalists, specialization will be favored for at least some values of . If the fitness of generalists is greater than or equal to that of specialists, specialization will only be favored if .

For generalists, the fitness is simply , as each cell has and (before and after sharing). Viability specialists produce and , while fecundity specialists produce and . Viability specialists then share with each of their z neighbors. After sharing, fecundity specialists receive from each of their viability specialist neighbors. But how many of their neighbors are viability specialists? We label the fraction of cells connected to fecundity specialists that are viability specialists f, that is, f is the mean number of viability specialists connected to each fecundity specialist divided by z, averaged over all fecundity specialists. For a bipartite graph, ; for a randomly connected graph on which half of cells are viability specialists and half of cells are fecundity specialists, . Group fitness is thus:

Here, is the average viability returns each fecundity specialist has received after sharing, which is multiplied by the amount of fecundity each fecundity specialist has (1) and the number of fecundity specialists (). Writing in terms of :

Specialists will be favored if the ratio . This will be true if: which reduces to:

This inequality implies that specialization will only be favored if fecundity specialists are preferentially connected to viability specialists, that is, if . Further, for a fully connected network , so this inequality is never satisfied, that is, specialists cannot have larger fitness than generalists for and fully connected topologies, as classically predicted.

Further, f cannot be more than 1, so if the threshold from the inequality in Equation 5 is greater than or equal to 1, specialization cannot be favored for . Thus, specialization for is only possible if:which reduces to: . This again reproduces a classic result: specialization for is not possible for disconnected cells.

This analysis allows us to interrogate specific cases. For example, if , f must be greater than 2/3, while if , f must only be greater than 5/8. Can such networks be constructed? The answer will depend on both the number of cells and how they are connected. Ultimately, the question of if a graph can be made with particular values of f and z is a graph coloring problem, and beyond the scope of this manuscript. However, this inequality presents a useful heuristic which can be used to determine if specialization is favored by measuring just a few properties of the graph.

Effect of varying ratios of specialists

We now allow the fraction of fecundity specialists to be X (rather than forcing ). For generalists, the group fitness is unchanged, , while for specialists the group fitness is:

Writing in terms of gives:

Specialists will be favored if the ratio . This will be true if:

Compared to the threshold value of f when , if , that is, more than half of cells are fecundity specialists, the value of f necessary for specialization to be favored is lower. If , the threshold value of f is higher than if . In other words, 1:2 is different from 2:1, and they both are different from 1:1. Once again, the question of if a particular configuration can be created–and how–is a graph coloring problem beyond the scope of this manuscript. However, this mean field heuristic gives us some information about how to expect graphs with different ratios of specialists to generalists to behave.

We again ask what must be true for f to be less than 1 (if , specialization will not be favored). Thus, specialization is only possible if:which reduces to:

For a mean field model, specialization with is impossible if fewer than one fourth of cells are fecundity specialists. We stress here that this is a mean field model, and does not apply to scenarios in which cells have a wide range of values of z. If such networks do or do not favor specialization for will again be a graph coloring problem.

Discussion

During the evolution of multicellularity, formerly autonomous unicellular organisms evolve into functionally-integrated parts of a new higher level organism (West et al., 2015; Michod and Nedelcu, 2003). Evolutionary game theory (Corning and Szathmáry, 2015; Nash, 1950; Smith, 1988) argues that functional specialization should only evolve when increased investment in trade increases reproductive output. Conventionally, this requires returns from specialization to be accelerating, that is, convex or super-linear (Szathmáry and Smith, 1995; Smith and Szathmáry, 1997; Goldsby et al., 2012; Corning and Szathmáry, 2015; Boza et al., 2014; Taborsky et al., 2016; Page et al., 2006; Rueffler et al., 2012; Szekely et al., 2013). While this idea is intuitive, it is, in the case of fixed group topology, also overly restrictive. In this paper, we explore how social interactions within groups, measured by their network topology, affect the evolution of reproductive specialization. Indeed, when all cells within groups interact (with equal interaction strength), returns on investment must be an accelerating, that is, convex, function of investment for specialization to evolve (Figure 1a; Szathmáry and Smith, 1995; Smith and Szathmáry, 1997; Corning and Szathmáry, 2015; Cooper and West, 2018). Yet for a broad class of sparsely connected networks, complete specialization can evolve even when the viability and fecundity return on investment curves are saturating, that is, concave (Figure 3).

To understand how specialization can be favored despite concave return on investment (ROI) curves, consider Jensen's inequality. Jensen's inequality states that for a convex function , , that is, the average value of , , is larger than , where is the average value of x. A corollary of Jensen's inequality is that the opposite is true for concave functions, that is, for a concave function , . Jensen’s inequality guarantees that for concave ROI functions generalists produce more total viability and fecundity than specialists, and that for convex ROI functions specialists produce more total viability and fecundity than generalists.

Crucially, however, Jensen's inequality does not connect ROI convexity/concavity to group fitness. Jensen’s inequality relates the degree of specialization to the average viability and average fecundity produced, but does not itself say anything about group fitness, which is the product of viability and fecundity averaged across all cells. For fully connected topologies (i.e. Figure 4c), greater absolute productivity proportionally increases group fitness, and differentiation can only evolve with accelerating benefits of specialization. This is not the case for topologically structured organisms, where fitness also depends on how complementary specialist cells are connected. Natural selection acts on realized productivity, that is, average ; mutations that increase average v or average b without increasing average are not adaptive. The importance of connecting complementary specialists has long been appreciated in other contexts, such as metabolic cross-feeding, for which it has been shown that the spatial arrangement of unlike specialists plays a key role in determining their productivity (and thus fitness) (Co et al., 2020). Indeed, While Jensen's inequality ensures that generalists will produce more viability and fecundity than specialists given a concave ROI function, specialization can still increase the fitness of topologically structured groups by increasing realized productivity.

Rather than being unusual, networks favoring specialization readily arise as a consequence of physical processes structuring simple cellular groups (Allen et al., 2017). For example, septin defects during cell division create multicellular groups with simple graph structures (Figure 4a and b), where cells are connected only to parents and offspring (Bengtson et al., 2017b; Droser and Gehling, 2008; Ratcliff et al., 2012; Ratcliff et al., 2013). If cells share resources only with physically-attached neighbors, then the physical topology of the group describes its interaction topology, and these sparse networks strongly favor reproductive specialization. Finally, we note that the primary benefit of sparsity is that sparse networks are likely to be at least somewhat bipartite. The more bipartite-like a network is, the less effort is wasted, and the easier it is for specialization to be favored.

Disentangling the evolutionary underpinnings of ancient events is notoriously difficult. Still, it is worth examining the independent origins of complex multicellularity, which are independent runs of parallel natural experiments in extreme sociality. Complex multicellularity (large multicellular organisms with considerable cellular differentiation) has evolved in at least five eukaryotic lineages, once each in the animals (King, 2004), land plants (Kenrick and Crane, 1997), and brown algae (Silberfeld et al., 2010), two or three times in the red algae (Cock and Collén, 2015; Yoon et al., 2006), and 8–11 times in fungi (Nagy et al., 2018). In all cases other than animals, these organisms form multicellular bodies via permanent cell-cell bonds, creating long-lasting highly structured cellular networks. Both fossil and phylogenetic evidence suggests that early multicellular organisms in these lineages were considerably less complex, growing as relatively simple graph structures. For example, 1.2 billion year old red algae formed linear filaments of cells (Butterfield, 2000), basal multicellular charophyte algae formed circular sheets of cells radiating from a common center (Kenrick and Crane, 1997), the ancestor of the brown algae likely formed a branched haplostichous thallus that was either filamentous or pseudoparenchymatous (Silberfeld et al., 2010), and hyphal fungi are primarily composed of linear chains of cells. Much less is known about the topology of animals prior to the evolution of cellular specialization. One hypothesis is that early metazoans resembled extant colonial choanoflagellates (Fairclough et al., 2013), the closest-living protistan relatives of the animals (Fairclough et al., 2010). Extant colony-forming choanoflagellates have evolved a variety of multicellular structures with sparse cellular topologies and permanent cell-cell bonds. For example, many species form branched, tree-like structures (Leadbeater, 2015), Choanoeca flexa grows as a sheet of cells (Brunet et al., 2019), and Salpingoceca rosetta can form either linear chains or rosettes in which the cells are connected via cytoplasmic bridges formed through incomplete cytokinesis (Dayel et al., 2011). While these growth forms are quite diverse, they all share characteristics (i.e. permanent cellular bonds and sparse topologies) that promote the evolution of cellular differentiation.

The main differences between our work and previous investigations of the effect of group topology on specialization is that we consider the productivity of groups as a whole, not the cells within them, and we consider situations of highly asymmetric sharing. Our approach is general, and can be applied to other systems of trade and specialization, so long as (1) only the aggregate productivity of the group (and not the particles within it) is maximized, (2) the productivity of each particle within the group is a multiplicative function of returns on investment into two (or more) tasks, and (3) there is an asymmetry in how products of those investments are shared. While in this work we have focused on reproductive division of labor, a process in which fecundity returns are not shared at all, we show in the supplement that as long as sharing of two goods is sufficiently asymmetric, specialization with saturating returns on investment can still be adaptive (Appendix 1—figure 2).

Appendix 1—figure 2.

Effect of sharing two resources.

When two resources are shared to different degrees, specified by , specialization is sometimes favored under conditions of sublinear returns on investment . Interestingly, specialization is favored when one resource is shared liberally while the other resource is shared sparingly (though it is not necessary to have one resource remain totally unshared).

Finally, we note that alternative paths to specialization likely exist. For example, cells at different positions in a group may experience different local environments, which may produce cells with varied fecundity-viability trade-offs. A previous paper demonstrated that the evolution of specialization is favored if these ‘positional effects’ result in an initially heterogeneous population of cell types (Tverskoi et al., 2018). However, these positional effects were considered for the case of well-mixed groups (i.e. completely connected network topologies). We thus anticipate that future work examining the relationship between cellular interaction topology and cellular heterogeneity (as well as a wide range of complex and varied relationships between viability, fecundity, and multicellular fitness) will provide unique insight into the origin and diversity of multicellular forms.

Conclusion

We explored the evolution of reproductive specialization in multicellular groups with various cellular interaction topologies. Our results demonstrate that group topological structure can play a key role in the evolution of reproductive division of labor. Indeed, within a broad class of sparsely connected networks, specialization is favored even when the returns from cooperation are saturating (i.e. concave); this result is in direct contrast to the prevailing view that accelerating (i.e. convex), returns are required for natural selection to favor increased specialization (Cooper and West, 2018; Michod et al., 2006; Ispolatov et al., 2012; Solari et al., 2013; Michod, 2007; West et al., 2015). Our results underscore the central importance of life history trade-offs in the origin of reproductive specialization (Michod et al., 2006; Michod, 2007; Hammerschmidt et al., 2014; van Gestel and Tarnita, 2017; Noh et al., 2018), and support the emerging consensus that evolutionary transitions in individuality are not necessarily highly constrained (Ratcliff et al., 2012; Ratcliff et al., 2017; Fairclough et al., 2010; Brunet and King, 2017; Pennisi, 2018; Black et al., 2019; Rose et al., 2020; van Gestel and Tarnita, 2017; Black et al., 2019; Staps et al., 2019; Grosberg and Strathmann, 2007).

Materials and methods

Analysis

The gradient of the fitness with respect to the group investment strategy , iswhere is a unit vector in the direction. First notice that if , and where is a vector of ones, then the gradient is zero. This strategy, , corresponds to the ‘generalist’ strategy, where every cell invests equally into both tasks. Second, notice that if then the gradient is not zero under the generalist strategy, so at least some degree of specialization must be necessary to maximize fitness. To determine the stability of this solution we examine , the Hessian (see SI Equation 3) evaluated at the generalist critical point. If is negative definite, then the generalist strategy is a fitness maximum and is therefore an optimal strategy. If, on the other hand, has both positive and negative eigenvalues then the generalist strategy lies at a saddle point within the fitness landscape, and therefore the optimal strategy must be somewhere else in (or on the boundary of) the domain (i.e. for all ). Finally, note that is never positive definite since is always an eigenvector with negative eigenvalue (when ).

We also use the zero crossing of the largest eigenvalue of evaluated at and as an overall measure of how amenable a network is to specialization, even when .

Evolutionary simulations

Our evolutionary simulations maintain the same overall structure as the Wright-Fisher model: a discrete-time Markov chain framework with fitness-weighted multinomial sampling between generations and constant population size. Therefore we refer to them as Wright-Fisher evolutionary simulations. We initialize a population of groups, each of group size , with uniform random investment strategies. We then let them evolve for 1000 generations, selecting offspring according to the relative fitness of each group (see Equation 1). At each generation, there is a 2% chance for a mutation to a given group’s investment strategy . If a mutation occurs, a new investment strategy is selected from a truncated multivariate gaussian distribution centered at the current (pre-mutation) investment strategy and with standard deviation equal to . After mutations each group’s fitness is calculated according to Equation 1, and the population is ranked according to fitness. Finally, groups are selected (with replacement) to populate the next generation, according to a multinomial distribution weighted by the groups’ fitness ranks.

Measuring specialization

To quantify the degree of specialization associated with a given group’s optimal investment strategy—the one which maximizes the fitness—we introduce the following metric, which we refer to simply as ‘Specialization’:

Specialization ranges from 0 (for groups consisting of cells investing equally in functions v and b) to 1 for groups consisting of cells investing exclusively in either function.

Code availability

All evolutionary simulations and other computations associated with this work are available at github.com/dyanni3/topologicalConstraintsSpecialization (Yanni, 2020; copy archived at https://github.com/elifesciences-publications/topologicalConstraintsSpecialization).

In the interests of transparency, eLife publishes the most substantive revision requests and the accompanying author responses.

Acceptance summary:

The evolution of germ-soma differentiation is one of the most fundamental questions in evolutionary biology, and the present paper investigates the consequences of altering one of the most basic assumptions: the traditional (symmetric) division of labor that has been studied from biology to economics. The authors consider a diversity of network structures and fitness functions and they find that sparser networks lead to higher levels of specialization.

Decision letter after peer review:

Thank you for submitting your article "Topological constraints in early multicellularity favor reproductive division of labor" for consideration by eLife. Your article has been reviewed by two peer reviewers, and the evaluation has been overseen by a Reviewing Editor and Diethard Tautz as the Senior Editor. The following individual involved in review of your submission has agreed to reveal their identity: Pierrick Bourrat (Reviewer #2).

The reviewers have discussed the reviews with one another and the Reviewing Editor has drafted this decision to help you prepare a revised submission.

Summary:

This is a very interesting paper on the evolution of germ-soma differentiation in which the authors consider the topology of interactions between the cells that make up the whole. They find that classical considerations about the convexity or concavity of certain functions characterizing the advantages of specialization no longer hold when the network topology is nontrivial. And such topologies are indeed found in nature. The reviewers were generally very supportive of the work but raised a number of points that need to be addressed in a revised manuscript.

Essential revisions:

1) The authors use a notion of fitness in which clonal cells can have different fitnesses, or more accurately, clonal groups can have different fitness. We know that there are some precedents in the literature, but this notion of fitness does not correspond to the notion of fitness one can associate with natural selection. To illustrate why, consider the following plant example. Take a single genet with two ramets in two environmental patches, one rich and one poor. Each ramet might adopt a very different developmental strategy from the other, considering the ecological constraints it is subjected to. These two strategies would nevertheless not be heritable in the sense that two offspring ramets put in the same environmental patch would develop the same developmental strategies (excluding noise). Thus, the differential success of each ramet is not an evolutionary success that can be associated with natural selection. This is a case analogous to the one presented by the authors. The notion of fitness they refer to seems to be rather the notion of realized fitness. This has no implications for the author's results per se but instead leads to an interpretation in which natural selection is not at work for explaining the division of labor in situations of concavity.

2) Related to the previous point, there seems to be a tension between, on the one hand, the claim that a concave function can lead to an increase in reproductive specialization, and on the other hand, claims that it has something to do with fitness. Fitness is about expected values, and in a situation of concave function, two or more cells specializing would yield a lower collective fitness than when not specializing. From a purely analytical point of view (i.e., Fisher's fundamental theorem), this seems impossible. So my question to the authors is whether there is not hidden somewhere a convex function, which is the relevant one for the evolutionary dynamics observed. Otherwise, what is the ecological explanation of such a result? There must be some ecological constraints that give rise to this phenomenon, and it would be good to know what the authors think they are.

3) There are well-known cases presented in the population genetics literature in which Fisher's fundamental theorem seems violated, but this is because of the environment (including the social environment) changes over time, such as frequency-dependent effects on an individual's success. We wonder if the results of the authors could not related to this literature in some way.

4) The model description is a bit abstract and occasionally hard to follow. It would be great to have fecundity and viability defined, and even better to have some real biological example of what returns on viability might mean and how they might be shared (I don't find the filamentous fungi example informative, at least not in the way it is written). That would also help the reader understand why there are returns on viability but not on fecundity. That the vi vector is the "group investment strategy" also comes as a surprise and takes a bit for the reader to put it all together. Similarly, the existence of both a general adjancecy matrix and of a special case one that uses the β, is somewhat confusing the way it's described. If the authors anyway only work with the special case of equal sharing with the non-self neighbors then why not define the 1-β+β/ni quantities as cij when they appear in the text, and then write a fourth eqn for W in [1] that explicitly uses the β. That would certainly help the reader a lot.

5) Results subsection “Fixed resource sharing” first paragraph, we may be getting confused, but how can you vary β in the case when, as is now written, individual i "shares equally among interaction and self terms"? Doesn't this mean that β = 1?

6) “We conjecture that the troughs in Figure 3C, where specialization occurs for the lowest values of, occur when connectivity is just large enough so that a spanning tree is more likely to connect all individuals in the group than not”, we don't fully understand that conjecture: do the authors simply mean that the troughs occur when the random graph becomes connected with probability > 50%? (A spanning tree connects all individuals by definition.)

7) The authors suggest sparsity is the main determinant of whether a network supports reproductive specialization. But, their examples in 1B and 1C (where a ring is sparser than the bipartite network) to us suggest that it is not so much about "sparsity" as it is about "bipartiteness" – or how easy it is to subdivide the nodes into two classes such that most edges go between these two classes (that's what you'd want for specialization anyway, we guess), and that sparse graphs simply have a tendency to be close to bipartite. We suspect that a ring graph with an odd number of vertices will be less conducive to specialization (although you could still alternate germ/soma cells except at one point), and that a star graph where there is one node of degree n-1 and all the others have degree 1 may be an example of a sparse graph where evolving specialization is not so easy (because for this graph it's not clear how to divide the vertices into germ and soma).

8) Related to the previous point: we would be interested if the authors have considered what happens when the optimal strategy is not 1:1 but, say, 1:2. Does that make specialization more difficult? Here we think that, with a few additional simulations, the authors could add a lot to the paper in terms of the ability to connect properties of the graph (beyond comparing some explicit topologies and random graphs of varying sparsity) to its ability to support the evolution of reproductive specialization.

9) Finally, it would be nice to see how the different specialists are distributed on these networks (at least when the specialization is equal to 1). One can infer it, but we think it would visually help the reader to get the gist of how the model works very quickly.

Essential revisions:

1) The authors use a notion of fitness in which clonal cells can have different fitnesses, or more accurately, clonal groups can have different fitness. We know that there are some precedents in the literature, but this notion of fitness does not correspond to the notion of fitness one can associate with natural selection. To illustrate why, consider the following plant example. Take a single genet with two ramets in two environmental patches, one rich and one poor. Each ramet might adopt a very different developmental strategy from the other, considering the ecological constraints it is subjected to. These two strategies would nevertheless not be heritable in the sense that two offspring ramets put in the same environmental patch would develop the same developmental strategies (excluding noise). Thus, the differential success of each ramet is not an evolutionary success that can be associated with natural selection. This is a case analogous to the one presented by the authors. The notion of fitness they refer to seems to be rather the notion of realized fitness. This has no implications for the author's results per se but instead leads to an interpretation in which natural selection is not at work for explaining the division of labor in situations of concavity.

Thank you for the thought-provoking comment. We are not completely sure if we understand your point correctly (and if not, we are happy to continue the discussion), but we are using the concept of fitness in the standard sense. While cells within groups can have different numbers of surviving offspring, in our simplified modeling world, selection does not act on these differences. Instead, we assume that the cells within each group are clonal, and that there can be genetic differences between groups. These genetic differences are responsible for cellular behavior, namely the extent to which they specialize in viability or fecundity tasks. Selection acts at the group level, in a way that is simply proportional to cellular productivity within these groups. Thus, selection acts between groups on differences in group fitness, which is caused by heritable variation in cellular behaviors underlying differentiation that vary between the groups.

However, we appreciate the reviewer’s point. We previously described mutations as affecting individual cells, which would of course mean that groups are no longer clonal. It is more accurate to say that in our simulation model, mutations change the pattern of specialization and sharing at different positions in the group, which we think of as being driven by a heritable developmental program. As with our analytical results, selection only acts on group-level fitness. When model parameters favor specialization, we see initially uniform groups in which every cell is a generalist gradually evolve developmental programs featuring complete reproductive specialization, providing a simulation test of our analytical results. We have thus revised and clarified the description of these simulations. We also stopped referring to the “fitness” of individual cells in the paper, since it is potentially confusing, instead describing the direct consequences of cellular specialization on their productivity (each cell’s 𝑣 ∗ 𝑏).

In addition to the changes noted in the paragraph above, we modified the model description to read:

“We consider a model of multicellular groups composed of clonal cells that each invest resources into viability and fecundity. Because there is no within-group genetic variation, within-group evolution is not possible, though selection can act on group-level fitness differences. Specifically, we consider the pattern of cellular investment in fecundity and viability, and their sharing of these resources with neighboring cells within the group, to be the result of a heritable developmental program. Thus, selection is able to act on the multicellular fitness consequences of different patterns of cellular behavior within the group.”

We modified the simulation description to read:

“These simulations begin with no resource sharing (i.e., 𝛽 = 0); during every round, each group in the population has a 2% chance that a mutation will impact its developmental program, and the 𝛽 value for one of its cells will change.”

2) Related to the previous point, there seems to be a tension between, on the one hand, the claim that a concave function can lead to an increase in reproductive specialization, and on the other hand, claims that it has something to do with fitness. Fitness is about expected values, and in a situation of concave function, two or more cells specializing would yield a lower collective fitness than when not specializing. From a purely analytical point of view (i.e., Fisher's fundamental theorem), this seems impossible. So my question to the authors is whether there is not hidden somewhere a convex function, which is the relevant one for the evolutionary dynamics observed. Otherwise, what is the ecological explanation of such a result? There must be some ecological constraints that give rise to this phenomenon, and it would be good to know what the authors think they are.

Thank you again for this question. We believe the reviewers have found a very important point that lacked clarity in the previous version of our manuscript. We appreciate the deep question regarding the presence of a hidden convexity. To be frank, we were surprised by this result at first as well. However, this surprise stems from the fact that the mathematical rule connecting specialization and the second derivative of the return on investment (ROI) function does not apply to asymmetric trade on sparse networks.

The following response explains the logic underlying this result and shows why there are no hidden convexities in our model. We apologize for its length, but since it is central to the paper we wanted to make the math absolutely clear.

In our model we make the simple assumption that group fitness is directly proportional to the sum of cellular productivity (each cell’s 𝑣 ∗ 𝑏) within the group. We see this as having the simplest biological interpretation, groups that generate a larger number of cells can make more progeny (i.e., more groups), all else equal. As a result, while holding the number of cells per group constant, the group with the highest fitness will be composed of cells with the highest average fitness.

To understand why a hidden convexity need not be present here, we will first discuss the mathematical rule that typically connects specialization and the second derivative of the ROI function: Jensen’s inequality. Jensen’s inequality states that for a convex function F(x), the average value of 𝐹(𝑥), ⟨𝐹(𝑥)⟩, is larger than 𝐹(⟨𝑥⟩), where ⟨𝑥⟩ is the average value of 𝑥. In other words, ⟨𝐹(𝑥)⟩ > 𝐹(⟨𝑥⟩). A corollary of Jensen’s inequality is that the opposite is true for concave functions, i.e., for a concave function 𝐺(𝑥), ⟨𝐺(𝑥)⟩ < 𝐺(⟨𝑥⟩). We assume this is the rule the reviewers refer to when they say “in a situation of concave function, two or more cells specializing would yield a lower collective fitness than when not specializing.”

For the traditional case of fully connected topologies and symmetric sharing, Jensen’s inequality provides a mathematical connection between a group’s fitness and its degree of specialization. With fully connected topologies, each cell shares its returns on 𝑣 and 𝑏 equally with all cells. As a result, all cells end up with 𝑣 equal to the average 𝑣 and 𝑏 equal to the average 𝑏. As a result, each cell’s 𝑣 ∗ 𝑏 is thus the average of 𝑣 multiplied by the average of 𝑏. Thus, group fitness is directly proportional to average 𝑣 and average 𝑏. Jensen’s inequality guarantees that for convex ROI functions, the average v produced by specialists (⟨𝐹(𝑥)⟩) is higher than the average 𝑣 produced by generalists (𝐹(⟨𝑥⟩)). The same is true for 𝑏. Since specialists must have larger average 𝑣 and 𝑏 than generalists, group fitness must be higher as well. The same argument holds in reverse for concave functions, i.e., if the ROI function is concave, the average 𝑣 and average 𝑏 are lower for specialists than for generalists so group fitness is lower, too. Thus, for fully connected topologies and symmetric sharing, Jensen’s inequality allows one to connect group fitness to average 𝑣 and average 𝑏.

Crucially, however, the connection between ROI convexity/concavity and fitness is indirect. Jensen’s inequality directly relates the degree of specialization to the average 𝑣 and average 𝑏; Jensen’s inequality does not itself say anything about group fitness. Group fitness is proportional to the average of the product 𝑣 ∗ 𝑏, which does not have to be directly proportional to average 𝑣 or average 𝑏 (even though for fully connected topologies and symmetric sharing ⟨𝑣𝑏⟩ is directly proportional to ⟨𝑣⟩ and ⟨𝑏⟩).

That Jensen’s inequality connects the degree of specialization to average 𝑣 and average 𝑏, but not group fitness is a crucial distinction. Jensen’s inequality is a mathematical truism, and cannot be violated. Indeed, for any concave ROI function, generalists will produce more 𝑣 and 𝑏 than specialists. This fact is just as true for sparse topologies and asymmetric sharing as it is for fully connected topologies and symmetric sharing. However, for sparse topologies and asymmetric sharing, group fitness is not directly proportional to average 𝑣 and average 𝑏; instead, group fitness strongly depends on network structure as well.

To understand how group fitness decouples from average 𝑣 and average 𝑏 for sparse topologies and asymmetric sharing, consider a ring of four cells in three different configurations: one that alternates between viability and fecundity specialists, one in which like-specialists are connected to each other, and one in which all cells are generalists (pictured in Author response image 1; red cells are viability specialists, blue cells are fecundity specialists, and purple cells are generalists). For simplicity, we will set 𝛽 = 1, and we will initially consider the case when 𝛼 = 1. When 𝛼 = 1, Jensen’s inequality tells us that generalists and specialists will be equally productive. Classically, this would suggest that specialists and generalists should have the same fitness.

Author response image 1.

And, indeed, all three cases have the same average 𝑣 and the same average 𝑏 (½ for each). However, the group fitnesses are all different. Note, depending on the configuration, ⟨𝑣⟩⟨𝑏⟩ can be greater than, less than, or equal to ⟨𝑣𝑏⟩.Next, we consider the same three configurations, but with 𝛼 = 0.9. Jensen’s inequality tells us that for this value of 𝛼, generalists should have a higher average 𝑣 and average 𝑏. Indeed, the average 𝑣 and 𝑏 is higher for generalists than for specialists: 0.536 versus 0.5. However, the group fitness of generalists, 1.15, is still lower than the group fitness of alternating specialists (i.e., leftmost configuration in the schematic in Author response image 1), 1.333.

These examples show that Jensen’s inequality still holds, and still correctly tells us which configuration has the highest average 𝑣 and average 𝑏. However, average 𝑣 and average 𝑏 are no longer directly proportional to group fitness. Therefore, Jensen’s inequality does not directly inform group fitness, and we should not expect convex ROI functions to be required for specialists to be favored.

Ultimately, the difference between the classic fully connected topologies and the asymmetric sharing / sparse topologies we consider is the ability of viability specialists to preferentially share viability to fecundity specialists. In fully connected topologies, some of the potential benefit from specialization is wasted. Consider the case where viability specialists share viability with other viability specialists: this is entirely unhelpful, two viability specialists (with 0 fecundity) that help each other survive still both have zero fitness. When the return on investment function is linear (and the graph is fully connected), the effect of sharing with the same cell type (as opposed to a complementary cell type) exactly cancels out the benefits of specializing; trade is only beneficial once the return on investment function becomes convex. In this case, specialists make enough extra v and b that groups of specialists do better than groups of generalists.

Again, these results were surprising. To help establish intuition for the role of preferentially connecting unlike specialists, we developed a mean field model. Based on the average number of connections per cell, this model determines the fraction of connections that must be between unlike specialists for the network to support specialization with concave ROI. We find that if unlike specialists are preferentially connected, specialization despite concave ROI should be expected for a wide range of networks. These results are similar to our simulations of randomly generated graphs (see Figure 3 in the main text), in which we also observed specialization for a wide range of parameters. Combined, these observations suggest that specialization, despite concave ROI, does not require precisely designed topologies, but is a general principle applicable to many different network structures. We agree that this was not clear in the original manuscript, and have added this model to the main text as described below.

We added Discussion paragraphs on Jensen’s inequality:

“To understand how specialization can be favored despite concave return on investment (ROI) curves, consider Jensen’s inequality. […] With asymmetric sharing and sparse topologies, Jensen’s inequality still informs the average viability and fecundity produced, but does not directly inform the group fitness.”

We also added a mean field model to help the reader develop intuition for these results:

“Mean field model

Finally, to capture some general principles underlying this phenomenon, we consider a mean-field model with cells, each of which is connected to other cells. […]However, this inequality presents a useful heuristic which can be used to determine if specialization is favored by measuring just a few properties of the graph.”

We also added a Discussion paragraph on directed/wasted effort:

“Finally, we note that the primary benefit of sparsity is that sparse networks are likely to be at least somewhat bipartite. The more bipartite-like a network is, the less effort is wasted, and the easier it is for specialization to be favored.”

Finally we added a section on the four cell network:

“Jensen's inequality and sparse topologies and asymmetric sharing

To understand how average fitness decouples from average 𝑣 and average 𝑏 for sparse topologies and asymmetric sharing, consider a ring of four cells in three different configurations: one that alternates between viability and fecundity specialists, one in which like-specialists are connected to each other, and one in which all cells are generalists. […] Therefore, Jensen’s inequality does not directly inform average fitness, and we should not expect convex ROI functions to be required for specialists to be favored.”

3) There are well-known cases presented in the population genetics literature in which Fisher's fundamental theorem seems violated, but this is because of the environment (including the social environment) changes over time, such as frequency-dependent effects on an individual's success. We wonder if the results of the authors could not related to this literature in some way.

This question seems to be related to the central issue raised above – is there a hidden feature in the model that makes specialization beneficial despite a concave investment function (i.e., by adding convexity somewhere or introducing some frequency-dependent or environmental effect)? Above, we explained the logic of why specialization can be favored despite concave investment functions.

Also, to be clear, we don’t see our work as violating Fisher’s fundamental theorem, which states that selection will increase the mean fitness of a population at a rate proportional to the genetic variation in fitness in the population. Because our groups are clonal, all the genetic variation in our population occurs between groups, not within groups, and selection is acting only on group-level fitness. Thus, Fisher’s fundamental theorem is at a limit within groups (as there is no within-group genetic variation, there is no within-group evolution).

4) The model description is a bit abstract and occasionally hard to follow. It would be great to have fecundity and viability defined, and even better to have some real biological example of what returns on viability might mean and how they might be shared (I don't find the filamentous fungi example informative, at least not in the way it is written). That would also help the reader understand why there are returns on viability but not on fecundity. That the vi vector is the "group investment strategy" also comes as a surprise and takes a bit for the reader to put it all together. Similarly, the existence of both a general adjancecy matrix and of a special case one that uses the β, is somewhat confusing the way it's described. If the authors anyway only work with the special case of equal sharing with the non-self neighbors then why not define the 1-β+β/ni quantities as cij when they appear in the text, and then write a fourth eqn for W in [1] that explicitly uses the β. That would certainly help the reader a lot.

These are great points, and we agree that it is crucial that the model and its biological meaning are clear to the reader.

The dichotomy of viability vs. fecundity was originally used by Michod to partition components of cellular fitness into actions that contribute to keeping a cell alive (viability), and actions that directly contribute to reproduction (fecundity). The intuition underlying this is that multicellular organisms often have evolved to divide labor along these two lines (i.e., reproduction of the organism by germ cells and survival provided by somatic cells), while their unicellular ancestors had to do both. We define viability as an activity (Michod uses the term “effort”) that keeps the cell alive (e.g., investing in cellular homeostasis or behaviors that improve survival), and fecundity more narrowly as effort involved in cellular reproduction itself. At the cellular level, there appears to be a fundamental asymmetry in how viability effort and fecundity effort can be shared among cells: while multicellular organisms readily evolve differentiated cells that are completely reliant on helper cells for survival (i.e., glial cells that support neurons in animals or companion cells that support sieve tube cells in plants), no cell can directly share its ability to reproduce. To better understand the intuition behind this, consider a cell that elongates prior to fission. This cell must grow to approximately twice its original length. Two cells cannot elongate by 50% and then combine their efforts; elongation is an intrinsically single cell effort.

While we believe this was also Michod’s interpretation of “reproductive effort”, this concept, like many in biology, is subject to interpretation. While it is clear the cellular behaviors underlying replication cannot be shared, resources required for reproduction could have been provided by another cell. Fortunately, our main conclusions are general and can accommodate various definitions of reproductive effort, which include sharing. While we present the simplest case (no sharing of reproductive effort) to help explain our paper’s central idea as clearly as possible, we do not need to forbid sharing of reproductive effort in order for specialization to evolve with concave ROI functions. As long as there is a significant asymmetry in how much 𝑣 and 𝑏 are shared, specialization with concave ROI functions will evolve. We explore this generalization in Figure 2. Because much of the effort involved in reproduction is, by its nature, unshareable, we believe that asymmetry in sharing between 𝑣 and 𝑏 is a general feature of multicellular systems. Of course, our results are far more general than the evolution of multicellularity, and should apply to many systems in which entities trade along networks (biological, economic, etc.). We have updated the description in the text of how cells can share viability, illustrating the general point with additional examples.

Finally, we have updated our description of the model to clarify the roles of 𝛽 and 𝑐𝑖𝑗.

We modified the “Model” section to read:

“Reproductive specialization can be modeled as the separation of two key fitness parameters, those related to either viability or fecundity, into separate cells within the multicellular organism (13,35). […]Thus, selection is able to act on the multicellular fitness consequences of different patterns of cellular behavior within the group.”

And added clarifying statements like:

“In other words, 𝑐𝑖𝑖 = 1 − 𝛽, 𝑐𝑖𝑗 = 𝛽/𝑛𝑖 if cells 𝑖 and 𝑗 are connected, and 𝑐𝑖𝑗 = 0 if cells 𝑖 and 𝑗 are not connected.”

5) Results subsection “Fixed resource sharing” first paragraph, we may be getting confused, but how can you vary β in the case when, as is now written, individual i "shares equally among interaction and self terms"? Doesn't this mean that β = 1?

We do not allow cells to give all of their viability returns away, as this would result in cells that are not viable. Cells keep (1 − 𝛽) of their viability returns, and designate β of their viability returns for sharing. They then split their viability returns designated for sharing into 𝑁 + 1 equally sized portions, give one portion to each of their 𝑁 neighbors, and keep one portion for themselves. In other words, the β portion that is designated “for sharing” is not all given away, but is split equally among the connected cells and the cell itself. We have updated the explanation accordingly.

We removed the sentence “Individual 𝑖 shares 𝑣𝑖𝛼 equally among interaction and self terms.” that was on line 146. The model was explained in the previous section, so this statement was ultimately redundant and unclear. Instead, we modified the model explanation to read:

“Cells cannot give away all of their viability returns, as they would no longer be viable; mathematically, we count a cell among its neighbors and thus ensure that they always “share” a positive portion of viability returns with themselves, so that 𝑐𝑖𝑖 > 0.”

And

“…and when 𝛽 = 1 cells share everything equally among all connections and themselves.”

6) “We conjecture that the troughs in Figure 3C, where specialization occurs for the lowest values of, occur when connectivity is just large enough so that a spanning tree is more likely to connect all individuals in the group than not”, we don't fully understand that conjecture: do the authors simply mean that the troughs occur when the random graph becomes connected with probability > 50%? (A spanning tree connects all individuals by definition.)

The reviewers are correct that a spanning tree connects all individuals. We were attempting to state this fact using technical language that would be clear to experts (50% probability that a spanning tree was present) as well as with more general language accessible to non-experts (50% probability that all individuals connected). We have updated this statement to be clear to all.

We modified this statement to read:

“We conjecture that the troughs in Figure 3B, where specialization occurs for the lowest values of 𝛼, occur when connectivity is just large enough so that the existence of a spanning tree is more likely than not.”

7) The authors suggest sparsity is the main determinant of whether a network supports reproductive specialization. But, their examples in 1B and 1C (where a ring is sparser than the bipartite network) to us suggest that it is not so much about "sparsity" as it is about "bipartiteness" – or how easy it is to subdivide the nodes into two classes such that most edges go between these two classes (that's what you'd want for specialization anyway, we guess), and that sparse graphs simply have a tendency to be close to bipartite. We suspect that a ring graph with an odd number of vertices will be less conducive to specialization (although you could still alternate germ/soma cells except at one point), and that a star graph where there is one node of degree n-1 and all the others have degree 1 may be an example of a sparse graph where evolving specialization is not so easy (because for this graph it's not clear how to divide the vertices into germ and soma).

Thank you for this point, with which we largely agree; “bipartiteness” is more important than sparsity. Sparsity is important for two reasons. First, as the reviewers suggest, sparse graphs tend to be close to bipartite. Second, sparse topologies are common in nature. Thus, sparsity may be a common natural route to being somewhat close to bipartite. We have modified the text to make this point clear. However, as will be highlighted by our discussion below, additional features of the graph, such as connectivity, can play a role in determining if specialization is favored as well.

We also appreciate the star graph suggestion, which is a very interesting topology. This graph is also relevant to understanding the evolution of asymmetric investment in germ and somatic cells, as is common in among independently-evolved multicellular organisms, like animals, plants, and volvocine green algae. Hopefully after clarifying our model in response to previous questions it is now clear that the star graph can strongly favor specialization. We now work this example in detail in the supplement. Briefly, we will discuss a five cell star here.

Cartoon image generalists (a) and specialists (b) in a five cell star graph topology.

Wright-Fisher simulations of five cell star topologies for a range of 𝛼 and 𝛽 values.

If the central cell were a viability specialist, specialists would have a lower fitness than generalists. However, if the central cell is a fecundity specialist, then the four surrounding viability specialists provide the central cell with enough viability to make this configuration favored. In fact, for large groups (𝑁 > > 1), specialists are always favored – for any 𝛽 and any 𝛼. For the five cell examples shown above, specialists are favored for α values as low as 0.222; a heat map relating degree of specialization to 𝛼 and 𝛽 is shown in Author response image 3.

Author response image 3.

Finally, the reviewers are correct about rings with even and odd numbers of cells. The effect is small, but it is true that specialization is frustrated in rings with odd numbers of cells. In fact, we cut a discussion of this phenomenon from our manuscript as we worried that it would be a distraction. However, it is now clear that its absence in fact raises questions, so we have restored it.

Here, 𝛼∗, the lowest value of 𝛼 for which specialization is favored, is plotted versus the number of cells in the ring, for rings with even and odd numbers of cells.As discussed above, we added a Discussion paragraph that reads:

“Finally, we note that the primary benefit of sparsity is that sparse networks are likely to be at least somewhat bipartite. The more bipartite-like a network is, the less effort is wasted, and the easier it is for specialization to be favored.”

We added a section analytically working out when specialists and generalists are favored for star topologies with various N, 𝛼, and 𝛽.

We added a section discussing even and odd numbered rings.

8) Related to the previous point: we would be interested if the authors have considered what happens when the optimal strategy is not 1:1 but, say, 1:2. Does that make specialization more difficult? Here we think that, with a few additional simulations, the authors could add a lot to the paper in terms of the ability to connect properties of the graph (beyond comparing some explicit topologies and random graphs of varying sparsity) to its ability to support the evolution of reproductive specialization.

Thank you for this interesting suggestion. First, we return to the mean field model introduced in comment 2, but now allow the fraction of fecundity specialists to be 𝑋 (rather than forcing 𝑋 = ½). The model itself is presented in detail in the text changes below; here, we summarize the results.

Our mean field model suggests that a larger proportion of fecundity specialists makes concave specialization easier to achieve. Further, we find that concave specialization is only possible if more than one fourth of cells are fecundity specialists.

We stress here that this is a mean field model, and does not apply to scenarios like the star network, in which cells have very different individual values of 𝑧. If such networks do or do not favor specialization for 𝛼 < 1 will again be a graph coloring problem.

We added a section on varying ratios of specialists:

“Effect of varying ratios of specialists

We now allow the fraction of fecundity specialists to be 𝑋 (rather than forcing 𝑋 = 1/2). […] If such networks do or do not favor specialization for 𝛼 < 1 will again be a graph coloring problem.”

9) Finally, it would be nice to see how the different specialists are distributed on these networks (at least when the specialization is equal to 1). One can infer it, but we think it would visually help the reader to get the gist of how the model works very quickly.

We have visualized investment in different tasks on fully connected, ring, and bipartite graphs. We agree that these images are instructive for the reader.

In Author response image 4 is an image of specialists in a nearest neighbor topology:

Author response image 4.

In Author response image 5 for the bipartite network, complete specialization happens readily:

Author response image 5.

Author response image 6 for the complete network, generalists dominate with some fluctuations:

Author response image 6.

We modified Figure 1 to incorporate these images:

Appendix 1—table 1.

Largest eigenvalue of the Hessian evaluated at the generalist critical point as a function of , , and for three topologies.

When the group size , the balanced bipartite graph coincides with the neighbor graph, and indeed the eigenvalues agree. Similarly, when the balanced bipartite graph coincides with the complete graph and the eigenvalues agree. The interesting domain of is , so for the complete graph is always negative definite. However, the balanced bipartite and neighbor graphs show regions where the generalist strategy is not stable.

Topology	Largest eigenvalue
neighbor graph	α⁢(12)^2⁢α-3⁢(-1+43⁢α⁢β)
balanced bipartite graph	α⁢(12)^2⁢α-3⁢(-1+2⁢NN+2⁢α⁢β)
complete graph	α⁢(12)^2⁢α-3⁢(-1+α⁢β)