Acknowledging selection at sub-organismal levels resolves controversy on pro-cooperation mechanisms.

Cooperators who pay a cost to produce publically-available benefits can be exploited by cheaters who do not contribute fairly. How might cooperation persist against cheaters? Two classes of mechanisms are known to promote cooperation: 'partner choice', where a cooperator preferentially interacts with cooperative over cheating partners; and 'partner fidelity feedback', where repeated interactions between individuals ensure that cheaters suffer as their cooperative partners languish (see, for example, Momeni et al., 2013). However when both mechanisms can act, differentiating them has generated controversy. Here, I resolve this controversy by noting that selection can operate on organismal and sub-organismal 'entities' such that partner fidelity feedback at sub-organismal level can appear as partner choice at organismal level. I also show that cooperation between multicellular eukaryotes and mitochondria is promoted by partner fidelity feedback and partner choice between sub-organismal entities, in addition to being promoted by partner fidelity feedback between hosts and symbionts, as was previously known.

Introduction

Cooperation: why is it important and how can it persist?

Cooperation is wide-spread (Axelrod and Hamilton, 1981; Sachs et al., 2004; West et al., 2006; Frederickson, 2013). Cooperation is thought to drive major evolutionary transitions such as the emergence of eukaryotes and multicellularity (Maynard Smith and Szathmary, 1998). Cooperation between pathogenic microbes can worsen microbial infection (Sandoz et al., 2007; Falsetta et al., 2014). Cooperation between cells in tumor microenvironment can hasten cancer progression (Rattigan et al., 2012; Cleary et al., 2014). In industrial fermentation, we strive to keep microbes in a cooperative state (producing products useful to us despite the metabolic burden to microbes), and exploit inter-species cooperation to increase product yield (Zhou et al., 2011). Thus, understanding cooperation has broad implications in basic and applied biology.

Cooperation poses an evolutionary puzzle. A cooperator pays a cost to help other individuals. If other individuals cooperate by reciprocating, then the original cooperator may enjoy a net gain. However, other individuals are better off not reciprocating ('cheating'), in which case, the original cooperator will suffer a net loss. How might cooperation evolve and persist despite the incentive to cheat?

To state more formally, let us consider a community of interacting (Appendix note 1) individuals. A focal cooperator pays a net fitness cost (Appendix note 2) to generate benefits that can be shared or exchanged (Appendix note 3) with partners. Partners may belong to the same species as the focal cooperator (e.g. kin cooperation, Figure 1A) or a different species (mutualistic cooperation, Figure 1B). Partners may cooperate by reciprocating to the focal cooperator a similar benefit (homotypic cooperation, Figure 1A) or a different costly benefit (heterotypic cooperation, Figure 1B) (Appendix note 4), or cheat by not reciprocating (Figure 1A and Figure 1B, unfilled blue). When cooperative and cheating partners share equal access to the focal cooperator, under certain circumstances cheating partners will always be more fit than cooperative partners (Appendix note 5). In this case, cooperation may persist only if the community is partitioned into interaction groups (groups) (Appendix note 6) that are sufficiently variable in cooperator frequency (Price, 1970; Trivers, 1971; Bull and Rice, 1991; Frank, 1994; 1997; Foster and Wenseleers, 2006; Fletcher and Doebeli, 2009) (Figure 2). This way, since cooperator-dominated groups will produce more cooperative benefits than cheater-dominated groups, individuals in cooperator-dominated groups (likely cooperators) will reproduce more than individuals in cheater-dominated groups (likely cheaters). Consequently, cooperator frequency can increase in a community despite decreasing in each group (Figure 2).

Figure 1.

Examples of cooperation, cheating, partner choice (PC), and partner fidelity feedback (PFF).

(A) An example of homotypic (sharing an identical benefit) and kin (between genetic relatives) cooperation, and cheating. During iron limitation, a focal cooperative Pseudomonas aeruginosa cell (grey filled oval) pays a fitness cost to synthesize siderophores (grey triangles, color indicating origin) which are released into the environment to scavenge iron. Siderophore-Fe3+ complex can be taken up by the focal bacterium and partner cells. A cooperative partner (blue filled oval) also contributes siderophores which can be used by the focal cooperator. In contrast, a cheating partner (blue unfilled oval) uses siderophores without contributing any, and is competitively superior to cooperators (Griffin et al., 2004). (B) An example of heterotypic (exchanging different benefits) and mutualistic (between different species) cooperation, and cheating. In the obligatory cooperation between a yucca (grey) and yucca moths, moths can cooperate (left) or cheat (right). (C) In PC, a focal individual (grey) displays different responses ('conditional response', i and ii) or a fixed response ('screening', iii) that favors cooperative over cheating partners. (i and ii) A focal client (grey, bottom) will preferentially visit a cleaner (smaller blue fish) if the cleaner previously cooperated with (filled blue) instead of cheated (unfilled blue) another client (yellow, top) (Bshary and Grutter, 2006) or the focal client itself (grey, top) (Bshary and Schäffer, 2002). Cooperative cleaners only eat parasites (black squares), while cheating cleaners bite off nutritious client tissue. Thus, PC can operate before (i) or after (ii) interactions with partners. (iii) A focal cooperative yeast cell (grey) pays a fitness cost to express cell surface adhesive proteins. The focal adhesive cell will bind better to other adhesive cooperators (filled blue) than to non-adhesive cheaters (unfilled blue). This differential binding (solid versus dashed grey box) favors cooperation between adhesive cells, and allows the formation of cell clumps with enhanced stress resistance (Smukalla et al., 2008). (D) Cooperation between hosts (grey) and endosymbionts such as mitochondria (blue) has been traditionally used to illustrate PFF (Sachs et al., 2004; 2011). Cooperative mitochondria (filled blue) serve the host cell at a cost to their own reproduction, and are therefore displaced by cheating mitochondria (unfilled blue) in the same cell (center panel). However, vertical transmission of mitochondria means that host and mitochondria repeatedly interact with each other. This ensures that mitochondria will harm their own fitness if they do not serve their host (compare left with right). Consequently, cooperative mitochondria can persist (compare bottom with top) if their frequency varies greatly among different hosts. For this and all following figures, filled and unfilled symbols differentiate cooperative versus cheating partners.

DOI: http://dx.doi.org/10.7554/eLife.10106.003

Figure 2.

Variability in group cooperator frequency promotes cooperation.

(A) Rules of interactions: A cooperator (filled circle) gives birth to a cooperator and helps each of the rest of group members to sire one additional offspring before dying. A cheater (unfilled circle) gives birth to two cheaters and offers no help to other group members before dying. (B) Community-wide cooperator frequency can increase over time when groups vary significantly in their cooperator frequencies (left), but not when they share similar cooperator frequencies (right). Suppose that after group formation and interactions and reproduction of individuals in a group, a group with initial cooperator frequency P will have final size π and final cooperator frequency of . The Price equation states that , where is the average final group size, is the difference between and , the final and initial community-wide cooperator frequency, respectively. , the covariance between final group size and initial cooperator frequency of group, should be positive. It may be rewritten as βVar(P), and should increase as Var(P), the variance in initial cooperator frequency P across groups, increases. reflects , the change in cooperator frequency in each group, and should thus be negative. Consequently, for community-wide cooperator frequency to increase, the absolute value of must exceed that of . A numerical demonstration of the Price Equation is provided below. In the case on the left, two groups form. The group starting with 1 cooperator (G1) has an initial cooperator frequency of P=1/6, and ends up with final group size π= 16 and final cooperator frequency P=1/16. The group starting with 5 cooperators (G5) has an initial cooperator frequency of P=5/6, and ends up with final group size π= 32 and final cooperator frequency P=25/32. The average group size is (16+32)/2=24 while the change in global cooperator frequency is =26/48−6/12=0.042. Thus, is 1. , where E[X] represents the expected value of X. Here, 1 out of 2 groups started with 1 cooperator and 1 out of 2 groups started with 5 cooperators. Thus, ; . Hence, is also 1, equal to . This figure is simplified from a lecture given by Prof. Benjamin Kerr (University of Washington, Seattle).

DOI: http://dx.doi.org/10.7554/eLife.10106.004

Current definitions of partner choice and partner fidelity feedback

Identifying orthogonal mechanisms that independently promote cooperation will help us understand contributions from different sources. This will in turn move us closer toward predicting the persistence of cooperation in a natural or an industrial system.

Several conceptual frameworks have attempted to classify pro-cooperation mechanisms (Sachs et al., 2004; Nowak, 2006; Lehmann and Keller, 2006; West et al., 2007). Unlike most frameworks which focus on kin cooperation, the framework by Sachs et al. considers both kin and mutualistic cooperation. It classifies pro-cooperation mechanisms known under a plethora of names ('kin selection', 'host sanctioning/policing', 'green beard', 'reciprocity' - to name a few) to either 'partner choice (PC)' or 'partner fidelity feedback (PFF)' (Sachs et al., 2004) (Appendix note 7).

In PC, a focal cooperator 'recognizes' and 'chooses' cooperative instead of cheating partners to interact with. Choice can occur before (Figure 1C,i) or after (Figure 1C,ii) interacting with partners ('conditional response') (Sachs et al., 2004). Archetti et al. has broadened PC to include 'screening' where a focal individual displays a single fixed response, but this response favors cooperative over cheating partners (Archetti et al., 2011a; 2011b) (Figure 1C,iii; Figure 3D). Thus, PC causes cooperator frequency to vary among groups as a focal cooperator favors cooperative over cheating partners during (Figure 1C) or after (Figure 3D) the formation of interaction groups. This leaves cheating partners isolated in groups devoid of cooperators.

Figure 3.

Revised definitions of PC and PFF.

(A) Adapted from Figure 3 in (Momeni et al., 2013) by B. Momeni. [L− A+], [A− L+], and [L−] are reproductively isolated yeast cells expressing different fluorescent proteins. On an agarose pad lacking adenine and lysine supplements, [A− L+] (green) and its cooperative [L− A+] (magenta) and cheating [L−] (blue) partners undergo self-organization from an initially random spatial distribution. [A− L+] and [L− A+] physically associate with each other and grow tall. [L−], isolated from [A− L+], fail to grow tall. ‘xy’: top-views; ‘z’: vertical sections. In simulated top-views, higher color intensity indicates a greater number of cells. In simulated vertical cross-sections, low and high color intensities represent dead and live cells, respectively. Scale bar: 100 μm. (B) In PC (i), a focal cooperative entity (grey filled circle) preferentially directs cooperative benefits to cooperative over cheating partner entities (filled and unfilled blue squares, respectively) despite their spatially equivalence. In PFF (ii), such discrimination does not exist. Instead, cooperative benefits are more available toward partner entities closer-by regardless of their cooperativeness. Beneficial interactions are marked by black arrows, with directions of benefit flow marked by arrowheads and interaction strength marked by line thickness. (C) A comparison of PC and PFF. (D) Top: Anti-competition cooperation in bacteria. Cooperators (filled ovals) but not cheaters (unfilled ovals) harbor toxin-antitoxin gene pair. Cooperators grow slower than cheaters. When encountering stress such as high cell density, a fraction of cooperators will 'commit suicide' (yellow) by lysing themselves (dashed outline) and release stable toxin (black). The remaining cooperators (filled blue) are immune to toxin because they express antitoxins (grey), while cheaters (unfilled blue) are sensitive to toxin-killing (dashed outline). This cell killing reduces competition, a benefit shared among surviving cooperators (filled blue) and cheaters (unfilled blue, solid outline). Bottom left: When PC acts alone in a well-mixed environment, cooperators can increase in frequency only if their initial abundance is sufficiently high. To see how this works, suppose that whenever total population size has reached 250, 10% of cooperators will commit suicide, and each suicidal cooperator can kill 2 cheaters. If starting at 210 cooperators and 40 cheaters, all cheaters will be killed off. However, if starting at 50 cooperators and 200 cheaters, the cheater population will diminish by 10 individuals (5%) only, compared to the 10% reduction in cooperators. Bottom right: In a spatially-structured environment, the joint action of PC and PFF favors cooperators even when cooperators are initially very rare. Data figures are reproduced from Chao and Levin’s work (Chao and Levin, 1981) with full permission from the authors. (E) Yucca flowers (grey ovals) dominated by cheating (unfilled blue crescent) but not cooperative (filled blue crescent) pollinator offspring are aborted. This mechanism can be interpreted as PC between a focal yucca plant and its internal, spatially-equivalent pollinator offspring (left) or PFF between yucca flowers and pollinator offspring (right). For simplicity, pollens delivered by cooperative moths are omitted from this figure. Grey cross-bar: death of an entity.

DOI: http://dx.doi.org/10.7554/eLife.10106.005

In PFF, individuals are "associated for an extended series of exchanges that last long enough that a feedback operates" (Sachs et al., 2004). That is, during repeated interactions (e.g. in a spatially-structured environment), an individual who does not help its cooperative partner will eventually suffer because its cooperative partner suffers (Figure 1D) or even dies. In PFF, high inter-group variability in cooperator frequency, which is required for promoting cooperation, can be achieved via population bottlenecks (drastic reductions in population sizes) (Brockhurst and Svensson, 2007; Chuang et al., 2009; Nadell et al., 2010; Harcombe, 2010; Van Dyken et al., 2013; Datta et al., 2013; van Gestel et al., 2014; Waite et al., 2015) (Appendix note 8). For example in kin cooperation, interaction groups originating from single individuals will comprise the founder and its progenies, thus exhibiting a cooperator frequency of either 1 or 0 depending on whether the founder is a cooperator or a cheater. Population bottlenecks can be realized through range expansion of a few individuals to new locations (Hallatschek et al., 2007; Mitri et al., 2015, population dilutions (Chuang et al., 2009), population crashes caused by cheaters (Waite et al., 2015), and selection for rare adaptive mutants (Morgan et al., 2012; Waite and Shou, 2012; Asfahl et al., 2015).

Sachs et al framework is compatible with or simplifies other conceptual frameworks (Appendix note 9). For example in kin cooperation, Hamilton’s “'kin discrimination' and 'viscous population' (Hamilton, 1964a; Hamilton, 1964b) would correspond to PC and PFF, respectively. To summarize current knowledge, PC promotes cooperation through 'active' recognition/screening by a focal individual. In contrast, PFF promotes cooperation through repeated interactions among group members in groups where cooperator frequencies vary significantly due to stochasticity.

Ambiguities associated with current interpretations of PFF and PC

What differentiates PC and PFF? Sachs et al. pointed out that unlike PFF, PC does not require repeated interactions between individuals (Sachs et al., 2004). This is certainly true. However, PC can lead to PFF (e.g. a client fish returning again and again to the same cooperative cleaner fish [(Bshary and Schäffer, 2002]), thus blurring the distinction between PC and PFF.

Sachs et al. and other further suggested that unlike PC, PFF is 'automatic', 'passive', and 'does not require recognition or conditional response' (Sachs et al., 2004; Foster and Wenseleers, 2006). However, definitions of these terms are unclear. For example, a biochemist might interpret 'passive' as not requiring ATP (e.g. passive diffusion), while a geneticist might interpret 'passive' as not requiring genes (e.g. death from some environmental assaults does not require gene activities). To further illustrate this problem, let’s examine PFF acting alone in the absence of PC in a community of engineered yeast strains (Momeni et al., 2013). In this community, adenine-requiring cooperators release lysine ([A− L+]) while lysine-requiring partners cooperate by reciprocating adenine ([L− A+]) or cheat by not reciprocating adenine ([L−]) (Figure 3A). Supplying metabolites is costly (e.g. [L−] is more fit than [L− A+]) (Waite and Shou, 2012; Momeni et al., 2013). PC does not operate here: lysine released from a focal [A− L+] cell is available to both [L−] and [L− A+]. In a spatially-structured environment (e.g. on an agarose pad) without adenine or lysine supplements, [A− L+] and [L− A+] physically mix and grow to form tall 'hills' while cheating [L−] segregate to 'foothills' and fail to grow tall (Figure 3A). Consistent with this being PFF, disruption of repeated interactions via frequent mixing favors cheater [L−] over cooperator [L− A+]. Furthermore, population bottleneck during spatial range expansion (Hallatschek et al., 2007; Mitri et al., 2015), which presumably increases variability in cooperator frequency across different locations, favors cooperator (L− A+] over cheater (L−] (Momeni et al., 2013). However one could argue, albeit somewhat absurdly, that this PFF relies on the fact that cells can 'recognize' metabolites from partners via permeases and mount 'conditional responses' (growth versus no growth to cooperative versus cheating partners). And this process is not 'passive' either, because cell growth requires energy and numerous genes.

Difficulties in distinguishing PFF and PC can lead to controversy, especially when interactions occur in a spatially-structured environment where both PFF and PC can operate. In fact, mutualisms such as those between fig trees and fig wasps (Jander and Herre, 2010), between yuccas and yucca moths (Pellmyr and Huth, 1994), and between legumes and rhizobia (Kiers et al., 2003) have been thought to be stabilized by PC by some authors (Sachs et al., 2004; Kiers et al., 2011) and PFF by other (Archetti et al., 2011a; Weyl et al., 2010). For example, in yucca-yucca moth mutualisms, a yucca selectively aborts flowers dominated by cheating pollinator offspring (Pellmyr and Huth, 1994). The PC camp argues that a plant partitions its resource into separate flowers, and 'decides' whether a flower has enough cheating pollinators to be aborted (Sachs et al., 2004). The PFF camp counters that flower abortion is not PC, but rather the fitness consequence of being damaged by cheating pollinators during excessive oviposition, since experimentally-inflicted damages also trigger flower abortion (Weyl et al., 2010; Archetti et al., 2011a; Weyl et al., 2011).

This controversy has led to the proposal of an alternative criterion for distinguishing PFF and PC: If an individual’s response is specific to the cheating behavior of a partner, then PC operates; if the response is toward general stress (which may or may not be inflicted by a cheating partner), then PFF operates (Archetti et al., 2011a). However, this alternative criterion can also encounter difficulties. Suppose that PC initially did not exist. It would seem reasonable that an evolved response to cheater-specific signals should be channeled to preexisting stress-response pathways which are already in place to facilitate survival. That is, flower abortion in response to physical damages does not preclude the possibility that the same flower abortion pathway can also be activated by cheater-specific signals. Thus, a clear criterion to differentiate PFF and PC is, to date, lacking (Appendix note 10).

Defining PC and PFF in terms of 'entities'

In hierarchically-organized biological systems, interactions and selections can occur at multiple (including sub-organismal) levels. The controversy above arises from attempting to compress a multi-level process into a single-level process. If instead we consider 'entities' at multiple levels and allow PC and PFF to act on these entities, then this controversy is resolved.

Definition of entity

A biological entity (entity) is a biological structure with a boundary such that the birth, or growth, or survival, or death of an entity is separable from that of other similar entities due to chemical or physical coupling within an entity and the lack of equivalent coupling between entities (Appendix note 11). An organism such as a yucca is an entity, because different parts of the yucca are coupled not only physically but also chemically (e.g. metabolism), and because equivalent intra-yucca coupling does not exist between yuccas. Consequently, the birth, or growth, or survival, or death of a yucca can be separated from other similar entities (Appendix note 12). Ascending above organism, an entity can be a collection of organisms. For example, a yucca and its internal moth offspring together can be considered an entity (Appendix note 13). This is because moth offspring depend on host yucca for survival, and this dependency does not extend to another yucca. The collection of moth offspring inside a yucca is also an entity, because the survival of these moth offspring are coupled through the survival of their shared host yucca, and such coupling does not exist with moth offspring in another yucca. Similarly, an ant colony where the task of maintaining the colony is divided among different castes is also an entity. In contrast, a collection of trees in a forest of similar trees does not constitute an entity, because coupling within a collection (e.g. competition between trees within the collection) extends to between collections. Descending below organism, an entity can also be an organismal part. For example, a yucca flower is an entity because it can be aborted independently of other flowers. Similarly, a yucca flower and its internal moth offspring together can be considered an entity, and the collection of moth offspring inside a yucca flower can also be considered an entity. Cells or gametes are entities in a multicellular organism, because they can divide and/or die independently of other similar entities. Similarly, mitochondria are entities in a cell, and mitochondrial genomes (mtDNAs) are independently replicating entities in a mitochondrion (see section below). Thus, an entity (e.g. a yucca-and its internal moth larvae) can contain smaller entities (e.g. the moth larvae; the yucca) which contain even smaller entities (e.g. flowers) which contain even smaller entities (e.g. cells, mitochondria, mtDNAs), reflecting the hierarchical organization of multicellular organisms (Appendix note 14).

Revised definition of PC

PC requires that a focal cooperative entity can direct more cooperative benefits to cooperative over cheating partner entities despite their spatial equivalence (Appendix note 15) (Figure 3B left, Figure 3C). To a focal entity, partner entities are spatially-equivalent if the focal entity would have interacted with partner entities equally had they been identical. For example, partner entities of equal distance to a focal entity can be spatially-equivalent, so are partner entities in a well-mixed environment. As discussed earlier, PC may be achieved through conditional response of the focal entity or screening by the focal entity. I emphasize that PC must occur between a focal cooperative entity (instead of a focal entity population) and partner entities. Otherwise, PFF could be mis-interpreted as PC (Appendix note 16). In essence, my definition of PC can serve as a definition for terms such as 'active' or 'recognition'.

Revised definition of PFF

Let us now consider potential mechanisms that promote cooperation when PC does not exist or does not operate fully (Appendix note 17) to exclude all cheaters. One mechanism orthogonal to PC is PFF. PFF requires the formation of interaction groups that vary significantly in cooperator frequency due to stochasticity and that last long enough so that cooperators benefit from their cooperative acts (Figure 3B right, Figure 3C). PFF can operate within a population of entities (e.g. homotypic cooperation in Figure 3D) or between populations of entities (e.g. heterotypic cooperation in Figure 1D and Figure 3A). As an example of PFF, let’s consider 'proportional tit for tat (pTFT)', a game theory strategy. In pTFT, the focal individual starts cooperating and subsequently cooperates with a probability equal to the fraction of cooperative partners from the previous round of interactions (Hilbe et al., 2014). PC does not operate in pTFT, since the focal individual does not differentiate cooperative versus cheating partners in the same interaction group. Indeed, for pTFT to survive cheaters, both requirements of PFF must be satisfied: repeated interactions in groups and large variation in the frequency of pTFT across groups (Axelrod and Hamilton, 1981).

PC and PFF can act synergistically, an idea compatible with existing mathematical framework (e.g. Eq. 1a in [Foster and Wenseleers, 2006]) and can be found in the anti-competition cooperation among bacteria (Figure 3D) (Chao and Levin, 1981).

Experimental tests of PC and PFF

The revised definition of PC leads to a conceptually simple experimental test. One can place partner entities of varying cooperative qualities at the same distance to the focal entity or together in a well-mixed environment. One can then test whether the focal entity preferentially directs benefit to the cooperative instead of cheating partner entity. If so, then PC operates.

To experimentally demonstrate the contribution of PFF in promoting cooperation, one will need to compare cooperator frequency when both PFF and PC are operative (e.g. in a spatially-structured environment) with when only PC is operative (e.g. in a well-mixed environment). Such an example is provided in Figure 3D. However, this experiment is sometimes not possible to do. For example, if PC also relies on a spatially-structured environment (e.g. a focal entity senses spatial gradients of benefits emanating from partner entities), then a well-mixed environment will destroy both PFF and PC. In this case, one can compare cooperator frequency when PFF acts alone (in a spatially-structured environment where PC has been mutationally inactivated) with when neither PFF nor PC is operative (in a well-mixed environment with PC mutationally inactivated).

Ideally, one would quantify processes involved in PC and PFF, and mathematically model how they contribute to cooperator survival. I illustrate how to do so using the bacterial anti-competition cooperation as an example (Figure 3D, Appendix note 18). If predictions on cooperator frequency match experimental observations without 'tweaking' experimentally-measured model parameters, then there is no need to invoke additional pro-cooperation mechanisms. Otherwise, we will need to look deeper into potential causes of model-experiment mismatch.

Revisiting yucca-yucca moth cooperation

We now return to the cooperation between yuccas and yucca moths. Selection against cheating pollinators can occur via PC between a focal yucca and its internal pollinator offspring (Figure 3E, left). This is because despite the spatial equivalence of pollinator offspring (i.e. all in the yucca), the focal yucca can selectively abort flowers dominated by cheating pollinator offspring. However, we can also view this as PFF between yucca flowers and pollinator offspring (Figure 3E, right). Note that since a focal flower can not discriminate its internal, spatially-equivalent pollinator offspring, PC does not operate here. Instead, PFF between flowers and pollinator offspring requires the spatially-structured environment defined by flowers (Appendix note 19). In summary, a single mechanism (aborting cheater-dominated flowers) can be viewed as PC between a focal yucca and its internal pollinator offspring and as PFF between yucca flowers and pollinator offspring. Even though Sachs et al. and Weyl et al. respectively classify this cooperation as being stabilized by PC and PFF, my revised definition suggests that both are correct depending on entities under consideration.

Thus, to understand and potentially predict how persistent cooperation is, we will need to consider entities at all relevant levels. Between entities at given levels, we should be mindful that PC and PFF may independently promote cooperation (Figure 3D). At the same time, since PFF at lower levels (e.g. between yucca flowers and moth offspring) can appear as PC at higher levels (e.g. between a focal yucca and moth offspring) (Figure 3E), we should avoid double-counting.

PFF and PC in mitochondrion-eukaryote mutualistic cooperation

I will use the mutualistic cooperation between multicellular eukaryotes and mitochondria (Appendix note 20) as an example to illustrate PFF and PC at various levels. A cooperative mitochondrial genome (mtDNA) harbors genes that are necessary for the host eukaryote, but the transcription of these genes can slow down the replication of mtDNA itself because these two processes compete for the same protein factor(s) (Larsson et al., 1998). Thus, mtDNAs not performing host-serving functions can (Hofhaus and Gattermann, 1999; Taylor et al., 2002; Harrison et al., 2014; Jasmin and Zeyl, 2014) (Appendix note 21), though not always (Appendix note 22), gain an advantage in self-replication compared to normal mtDNAs. Here, I use 'dysfunctional mtDNAs' to describe mtDNAs that, compared to functional mtDNAs in the same 'heteroplasmic' mixture, can replicate themselves at least as well but generate less or no benefit for the host eukaryote. Dysfunctional mtDNAs can be devastating to the host eukaryote (Chinnery et al., 2000; Chan, 2006). In a heteroplasmic mixture of functional and dysfunctional mtDNAs (Fig 4A), what mechanisms oppose dysfunctional mtDNAs and their dysfunctional mitochondria? (Appendix note 23)

The importance of identifying PFF and PC at various levels becomes evident if we are interested in mitochondrial disease prognosis. Consider a human mother who is heteroplasmic for functional and dysfunctional mtDNAs/mitochondria (Chinnery et al., 2000). Since her children would only inherit mitochondria from her (Pyle et al., 2015), what would their disease prognosis be? Suppose that dysfunctional mtDNAs/mitochondria are purged from the human population solely because they harm the fitness of their host organisms (Appendix note 24) (i.e. via PFF between host organisms and mtDNAs/mitochondria, Figure 4B), as traditionally thought (Sachs et al., 2004; 2011). Then, offspring should on average suffer a similar or a severer fitness defect than their mother does, since dysfunctional mtDNAs/mitochondria accumulate and are not selected against in the mother. Fortunately, as I will discuss below, PFF and PC at sub-organismal levels create somatic and germline 'filters' against dysfunctional mtDNAs/mitochondria (Figure 4C–E), thus increasing the fitness of the mother and her offspring.

Figure 4.

PFF and PC act at multiple levels in the mutualistic cooperation between mitochondria and multicellular eukaryotes.

(A) An eukaryotic cell contains multiple mitochondria (dashed versus solid boundaries representing dysfunctional versus functional mitochondria). Each mitochondrion contains multiple mtDNAs (unfilled versus filled circles representing dysfunctional versus functional mtDNAs). (B) The mutualistic cooperation between mitochondria and eukaryotes is traditionally thought to be stabilized by PFF between eukaryotic host organisms and their mtDNAs/mitochondria. A host organism dominated by dysfunctional mtDNAs/mitochondria leaves fewer offspring (left) than a host organism dominated by functional mtDNAs/mitochondria (right). Due to mtDNA bottleneck, oocytes and offspring exhibit large variability in the level of dysfunctional mtDNAs/mitochondria they inherit from their mother. This facilitates PFF (see text). (C) Mitochondria dominated by functional mtDNAs replicate faster than those dominated by dysfunctional mtDNAs, even though dysfunctional mtDNAs replicate as fast as or faster than functional mtDNAs in the same mitochondrion. (D) PC between a focal cell and mitochondria/mtDNAs can occur through selective import, asymmetric segregation, or mitophagy. (E) Cells dominated by functional mtDNAs/mitochondria have a higher fitness than those dominated by dysfunctional mtDNAs/mitochondria. Grey crossbars: elimination; grey arrows: time.

DOI: http://dx.doi.org/10.7554/eLife.10106.006

Starting from the lowest levels, selection against dysfunctional mtDNAs can occur via PFF between mitochondria and mtDNAs (Figure 4C) (see Appendix note 25). Dysfunctional mtDNAs can be selected against if they harm the reproduction of their host mitochondria. Indeed, during Drosophila melanogaster oogenesis (Hill et al., 2014; Ma et al., 2014), mitochondria dominated by functional mtDNAs proliferate faster than mitochondria dominated by dysfunctional mtDNAs (Figure 4C). This could occur if, for example, dysfunctional mtDNAs render their host mitochondria incapable of importing proteins required for mtDNA replication.

Moving one level up, PC can operate between a focal cell and mitochondria/mtDNAs when the cell preferentially imports, inherits, or retains functional mitochondria (Figure 4D). An example of PC can be found between an oocyte and spatially-equivalent nurse cell mitochondria during D. melanogaster oogenesis: To be transported into the developing oocyte, mitochondria from the surrounding nurse cells need to be localized to the cytoplasmic bridge connecting nurse cells to the oocyte, which in turn depends on mitochondrial function (Cox and Spradling, 2003; Hill et al., 2014) (Figure 4D, left). A second example can be found during cell division when the stem daughter cell or the young daughter cell preferentially inherits young (functional) mitochondria (McFaline-Figueroa et al., 2011; Katajisto et al., 2015), despite the spatial equivalence of mitochondria in the mother cell (Figure 4D, center). A third example of PC is a host cell using mitophagy, cellular digestion of mitochondria, to selectively eliminate dysfunctional mitochondria while retaining functional mitochondria (Kim and Lemasters, 2011; Ashrafi and Schwarz, 2013) (Figure 4D, right) (Appendix note 26). Mitophagy-mediated elimination of dysfunctional mtDNAs/mitochondria can conceivably operate in germ cells, given that mitophagy is active in fertilized oocytes to eliminate paternal mitochondria (Ashrafi and Schwarz, 2013). As discussed earlier (Figure 3E), PFF mechanisms at a lower level (i.e. between mitochondria and mtDNAs) may appear as PC mechanisms at a higher level (i.e. between a cell and mitochondria/mtDNAs). For example, slower net proliferation of dysfunctional mitochondria (Figure 4C) may or may not be linked to mitophagy (Figure 4D). An important future challenge is to address the orthogonality of various mechanisms.

For those dysfunctional mtDNAs/mitochondria that have escaped a cell’s PC capability, they can be further selected against via PFF between cells and mtDNAs/mitochondria (Figure 4E). For example, in at least several types of mammalian cells, certain mutations in mtDNA will lead host cells to undergo programmed cell death (apoptosis) (Zamzami et al., 1995; Chomyn and Attardi, 2003) or cell cycle arrest (Arnould et al., 2002; Owusu-Ansah et al., 2008). At least some of these PFF mechanisms can operate in germ cells. For example, in birds and mammals, the majority of oocytes die via apoptosis, which has been hypothesized as a mechanism to purge dysfunctional mtDNAs/mitochondria (Krakauer and Mira, 1999). Furthermore, oocytes harboring dysfunctional mtDNAs/mitochondria are less likely to be fertilized (Perez et al., 2000). Thus, dysfunctional mtDNAs/mitochondria can be selected against because they lower the fitness of their host cells.

Finally, when we move one level further up, dysfunctional mtDNAs/mitochondria that have survived PC and PFFs at lower levels (Figure 4C–E) can be selected against via PFF between host organisms and mtDNAs/mitochondria (Figure 4B).

A model based on the Price equation predicts that mtDNA bottleneck (inheriting a small number of mtDNAs per mitochondrion or per germ cell) should facilitate PFF at the corresponding level. Indeed, mtDNA bottleneck has been observed during fly and mammalian oogenesis (Koehler et al., 1991; Chinnery et al., 2000; Cree et al., 2008; Hill et al., 2014) (Appendix note 27). This mtDNA bottleneck means that oocytes from a mother heteroplasmic for functional and dysfunctional mtDNAs will exhibit high variability in the frequency of functional mtDNAs that they inherit. Oocytes dominated by dysfunctional mtDNAs can be subsequently purged via apoptosis (Appendix note 28), thus creating a germline mtDNA filter that improves the prognosis of offspring.

Conclusion

To reflect the hierarchical organization of biological systems, I introduce the concept of 'entity'. I have revised the definitions of PC and PFF so that they are orthogonal and so that they can operate on organismal or sub-organismal entities. PC requires that a focal cooperative entity can direct more benefits to cooperative over cheating partner entities despite their spatial equivalence. PFF suppresses cheaters that have escaped PC (if PC exists) via repeated interactions between entities in groups that, by chance, show high variability in cooperator frequency. By applying this revised definition, I show that sometimes, PFF at a lower level (e.g. between yucca flowers and moth offspring) may appear as PC at a higher level (between a yucca and moth offspring). PFF and PC can also act in conjunction at multiple levels to promote cooperation, as exemplified in the mitochondrion-eukaryote cooperation. This framework brings conceptual clarity for understanding the persistence of cooperation.

In the interests of transparency, eLife includes the editorial decision letter and accompanying author responses. A lightly edited version of the letter sent to the authors after peer review is shown, indicating the most substantive concerns; minor comments are not usually included.

Thank you for submitting your work entitled "A revised criterion for differentiating mechanisms that promote cooperation" for peer review at eLife. Your submission has been favorably evaluated by Diethard Tautz (Senior Editor and Reviewing Editor) and two reviewers: James Bull and Luke McNally.

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:

The paper addresses the ambiguity in distinguishing between two mechanisms for the evolution of cooperation: partner choice (PC) and partner fidelity feedback (PFF). The author suggests distinguishing cases of PC from PFF by specifying the requirement that PC allows cooperators to direct their cooperation to other cooperators instead of cheaters despite their "spatial equivalence".

We agree that the arguments made in this paper are topical and of broad interest, and think that with refinement this paper could help resolve ongoing debates that the author mentions and be an important contribution to the field. However, we have some issues with the paper as it currently stands, particularly with the clarity of presentation and how practically useful the criterion suggested would be.

The manuscript is very densely packed at present, jumping between multiple diverse examples and theoretical issues in individual paragraphs. We suggest that the author restructures the manuscript, moving many of the examples to tables or boxes so that they do not distract from the narrative of the paper. This would greatly increase the clarity of the paper. While keeping the main text short, you could make optimal use of the online presentation options through Lens.

Both referees address the clarity issue in similar, but also partly complementary ways. The relevant passages below are directly taken from the reports.

Essential revisions:

Reviewer #1:

As an example that presents just such a challenge, I'll offer a hypothetical case that Joel Sachs and I pondered some years ago. A plant allows nitrogen-fixing bacteria to colonize various (discrete) patches in its roots. Each colonized patch functions independently of others. If the bacteria provide sufficient nitrogen to the patch, the plant cells in the patch proliferate and automatically 'feed' the bacteria carbohydrates, and the bacteria then proliferate as well; the mix of bacteria and plant cells in the patch increase accordingly and become a functioning nodule that benefits the plant and the bacteria. However, if the bacteria do not provide nitrogen to the patch, the plant cells in that patch die and thus so do the bacteria that colonized the patch. Thus, at the level of the individual patch, the mechanism is partner fidelity feedback – cooperators proliferate because of the feedback from helping their partners. But at the level of the entire plant, the mechanism is (I hope) partner choice – because the plant has disproportionately rewarded cooperative bacteria. The attempt to discriminate mechanisms gets murky unless one acknowledges that different mechanisms are operating, depending on which 'entities' are being considered.

The paper acknowledges and offers such examples, and it suggests a resolution to the existing debate by modifying the existing 'definition' of partner choice. The formal definition appears to emphasize discrimination among spatially equivalent partners as the key. My sense is that the new definition in this paper does indeed 'work,' but it needs to be expanded/modified in a couple ways to be fully operational. First, there is only a definition of partner choice. Partner fidelity feedback should also be defined, even if to say it is everything except choice. Second, the definition is cryptic about its use of 'entity.' I think that the formal definition should be expanded to explain more about 'entities.' The text following the definition does so, but something needs to go in the definition, so that when the definition gets quoted in future work, 'entities' are formally explained. I emphasize this point about entities because I suspect much of the confusion and controversy in the field lies in the failure to appreciate the potential operation of different mechanisms at different levels in the individual (that multiple levels of entities may exist within one individual); thus one of the paper's main contributions may be to raise awareness about entities. Note that my hypothetical example above fits logically into the 'entity' paradigm – the root patches are one level of entity, the cells within the patches another level.

The foregoing comprises my main point – clear up the definition. However, I also think (strongly) that the paper is difficult to follow. It is of course challenging to write such a paper, because the topic is not intuitive, so a general audience needs to be given considerable background that is already well known within the field. Keeping the paper short limits the depth of this background before the paper dives into the new material. So such a paper will pose difficulties in the best of circumstances. But I nonetheless think the paper is not written to be widely accessible. Specifically, there is too much explanation embedded in details of biological systems to fully comprehend the (Abstract) points that constitute the substance of the paper. I think the paper would be more easily understood if it relied on hypothetical examples easily grasped by the reader: hypothetical examples can force the focus onto the key issues and avoid distracting details. The hypothetical examples can be borrowed from natural examples, but they should be structured to emphasize only the point being made. The description of natural examples could come late, once the main points are laid out with hypothetical examples.

Reviewer #2:

The author's example of cooperation between mitochondria and host cells illustrates how complicated the relationship between PFF and PC can be, particularly in hierarchically organised systems. However, this leads to the question of how useful is this distinction between PFF and PC for understanding the evolution of cooperation? Is this a meaningful debate? My general feeling is that the debate does matter and can help us identify mechanisms and genes underlying cooperation and mutualism. However, I think a general readership would greatly benefit from discussion of why the distinction between PFF and PC matters and isn't simply semantics. I would suggest adding this at the beginning of the paper to engage readers before the author outlines her revised criteria.

My final issue with the paper is that while the author has suggested a new criterion for distinguishing PC from PFF it is not immediately clear how this criterion could help design experiments to tell them apart. I would suggest that the author give at least one hypothetical example of how to design an experiment based on her suggested criteria in order to distinguish these two mechanisms. In particular, the author mentions the example of the conflicting views over whether the figure tree and figure wasp mutualism is maintained by PC or PFF. If the author could suggest a hypothetical experiment to distinguish between PC and PFF in this example it would greatly strengthen her case that this distinction is practically useful for understanding mechanisms of cooperation.

Both reviewers:

The supplementary material on the Price equation does not really contribute much. It would be useful if the Price equation could be written to separate partner choice from fidelity feedback. This paper by Kevin Foster uses a direct fitness model to do just this: http://zoo-kfoster.zoo.ox.ac.uk/sites/default/files/files/FosterWenseleersJEB2006.pdf. Reworking of this model into Price equation form could likely be used to address this point.

[Editors' note: further revisions were requested prior to acceptance, as described below.]

Thank you for resubmitting your work entitled "Redefine mechanisms that promote cooperation" for further consideration at eLife. Your revised article has been favorably evaluated by Diethard Tautz (Senior Editor and Reviewing Editor) and two reviewers. The manuscript has been improved but there are some remaining issues that need to be addressed before acceptance, as outlined below:

Reviewer 1 provides several comments on how to further improve the manuscript which you should address in your revised version. Reviewer 2 raises a further important point, namely the discussion of the question of intraspecific competition. The full comments of this reviewer are also attached. However, after discussion between the reviewers and the editor, we have come to the conclusion that including this into the current version of the manuscript would increase its length beyond its limits and would make it even more complex. Accordingly, we should like to ask you to drop this part entirely and reserve it for a possible follow-up paper in a more specialized journal. Hence, the comments of Reviewer 2 are only for your information.

Reviewer #1:

This is a revision; its predecessor was longer, far less focused and difficult to follow. The revision is now at a stage that is succinct and complete, but it still needs work. I think that the manuscript now needs to add structure and some clarification, as below. As before, I think the paper has a meaningful and necessary contribution to add to our understanding of the evolution of cooperation.

1) Make clear early what the paper is adding to the field. The paper should start with a clear statement of what is being done without going into detail about how it is being done. (The Abstract attempts to do so, but awkwardly.) Are PFF (partner fidelity feedback) and PC (partner choice) being redefined or just clarified? Is the result a 'reconciliation' or 'redefinition?' It seems that the big change is to recognize different organizational levels within an organism, such that PC can operate at one level and PFF operate at a different level; if so, then the paper recognizes PFF and PC as mechanisms that may operate within individuals as well as between them (which is new and useful).

2) The new definition of PFF is the complementary set to PC (i.e., PFF is everything that is not PC). This is not a good approach, because it runs the risk of including standard PFF with other mechanisms that are yet to be discovered. PFF should be defined as a true mechanism. Furthermore, if the new focus on entities is useful, it should lead to a clear definition of PFF.

3) The existing heading structure is good. I would add more. For each definition, I would have a separate heading: Definition: Entities; Definition: Partner Choice (etc.). Following the definitions of PC and PFF, I would have a heading 'How to demonstrate.' The paragraphs on how to demonstrate are useful and not something that would have occurred to me.

4) Something to contemplate: if PFF and PC work together within a system to maintain cooperation, do we want to recognize some additional basis for the evolution of cooperation above and beyond PFF and PC? I don't have an answer, but this might be something to think about.

Reviewer #2:

I think this revision of the manuscript has greatly improved its clarity, and all of my previous comments have been well addressed. However, I have one substantial revision that I feel is necessary before publication.

Reading the previous version of the manuscript I believed the author was primarily concerned with interspecies/heterotypic cooperation. However, from this revision and the author's responses I can now see that the ambition of this paper is wider, aiming to provide a framework for understanding both interspecies and intraspecies cooperation. I think this is fine, and the author generally makes a strong case. However, my major concern here is that while the relationship between the author's classification scheme and previous schemes for classifying interspecies cooperation is clear, there is insufficient discussion of the relationship between the author's classification scheme and previous classification schemes for the evolution of intraspecies cooperation. I hesitate to suggest new additions on a second round of review, but I think the author could address my points by adding a short subsection to the paper, and think that this is essential to support the lofty ambition of the paper.

The two most influential schemes for intraspecies cooperation are those provided by inclusive fitness theory and the "rule based" approach from Martin Nowak and other game theorists. In particular there have been two highly influential reviews classifying intraspecies cooperation from an inclusive fitness perspective that the author doesn't cite or discuss (West et al. 2007: http://www.zoo.ox.ac.uk/group/west/pdf/West_etal_07_CB.pdf & Lehmann & Keller 2006: http://onlinelibrary.wiley.com/doi/10.1111/j.1420-9101.2006.01119.x/abstract) and also one high-profile review from a game-theoretic perspective (http://ped.fas.harvard.edu/files/ped/files/science06_0.pdf?m=1425933699). I think the paper needs a short section comparing the utility of the authors scheme to these different perspectives to really convince the reader that this scheme makes progress relative to the classification schemes for intraspecies cooperation of inclusive fitness (classification based on direct vs indirect fitness effects) and game theoretic classifications (classification based on different games/rules). While I personally think that inclusive fitness theory is the most useful scheme, I think the author has a strong case that her scheme could be a strong competitor, but this needs to be made more explicit.

Related to this point, one thing that I think could help increase the readers intuition for the current scheme in the context of intraspecies cooperation would be adding an explanation (possibly as a table) of how previous classifications would fall within it. There is some of this throughout the manuscript, but it is not centralised in any way. I see the main strength of the author's scheme as being the separation of causes into two meaningful biological components, that of demography (partner fidelity feedback, which captures population structure, group augmentation, non-linear demographic feedbacks) and plasticity (partner choice, which captures direct, indirect and generalised reciprocity, punishment, policing, greenbeards, kin recognition). Stating this more explicitly would help relate the new scheme to previous schemes and highlight it's main strength – while more specific effects are difficult to partition (e.g. telling apart different forms of reciprocity), the author offers a simple way to separate the effects of demography and plasticity.

If the author can add a short section addressing these points I think it will help this paper have a wider-ranging impact on the field.

[…] The manuscript is very densely packed at present, jumping between multiple diverse examples and theoretical issues in individual paragraphs. We suggest that the author restructures the manuscript, moving many of the examples to tables or boxes so that they do not distract from the narrative of the paper. This would greatly increase the clarity of the paper. While keeping the main text short, you could make optimal use of the online presentation options through Lens.

I have chosen diverse examples to ensure that all examples can fit into this redefinition. Indeed, illustrating diverse examples and explaining conceptual nuances have made the writing challenging. In this revision, I started with abstract concepts, moved many examples to figures, and hid even more conceptual nuances into notes. I hope that I have succeeded in improving the flow.

Both referees address the clarity issue in similar, but also partly complementary ways. The relevant passages below are directly taken from the reports. Essential revisions:

Reviewer #1:

As an example that presents just such a challenge, I'll offer a hypothetical case that Joel Sachs and I pondered some years ago. A plant allows nitrogen-fixing bacteria to colonize various (discrete) patches in its roots. Each colonized patch functions independently of others. If the bacteria provide sufficient nitrogen to the patch, the plant cells in the patch proliferate and automatically 'feed' the bacteria carbohydrates, and the bacteria then proliferate as well; the mix of bacteria and plant cells in the patch increase accordingly and become a functioning nodule that benefits the plant and the bacteria. However, if the bacteria do not provide nitrogen to the patch, the plant cells in that patch die and thus so do the bacteria that colonized the patch. Thus, at the level of the individual patch, the mechanism is partner fidelity feedback – cooperators proliferate because of the feedback from helping their partners. But at the level of the entire plant, the mechanism is (I hope) partner choice – because the plant has disproportionately rewarded cooperative bacteria. The attempt to discriminate mechanisms gets murky unless one acknowledges that different mechanisms are operating, depending on which 'entities' are being considered.

You are exactly right. The same event (aborting non-functional nodules) can be interpreted as PC or PFF, depending on entities under consideration.

The paper acknowledges and offers such examples, and it suggests a resolution to the existing debate by modifying the existing 'definition' of partner choice. The formal definition appears to emphasize discrimination among spatially equivalent partners as the key. My sense is that the new definition in this paper does indeed 'work,' but it needs to be expanded/modified in a couple ways to be fully operational. First, there is only a definition of partner choice. Partner fidelity feedback should also be defined, even if to say it is everything except choice. Second, the definition is cryptic about its use of 'entity.' I think that the formal definition should be expanded to explain more about 'entities.' The text following the definition does so, but something needs to go in the definition, so that when the definition gets quoted in future work, 'entities' are formally explained. I emphasize this point about entities because I suspect much of the confusion and controversy in the field lies in the failure to appreciate the potential operation of different mechanisms at different levels in the individual (that multiple levels of entities may exist within one individual); thus one of the paper's main contributions may be to raise awareness about entities. Note that my hypothetical example above fits logically into the 'entity' paradigm – the root patches are one level of entity, the cells within the patches another level.

I am very grateful for these great points. I have expanded my definition and discussion on entity:

“A biological entity (entity) is a biological structure with a boundary such that the birth, or growth, or death of an entity is separable from that of other similar entities due to chemical or physical coupling within an entity and the lack of equivalent coupling between entities Note 10. […] Thus, an entity (e.g. a figure tree-figure wasp entity) can contain smaller entities (e.g. the wasp larvae; the figure tree) which contain even smaller entities (e.g. figs) which contain even smaller entities (e.g. cells, mitochondria, mtDNA), reflecting the hierarchical organization of multicellular organisms Note 13.”

I have added the following definition for partner fidelity feedback.

“I define PFF as the mechanism that promotes cooperation when PC does not exist or does not operate fully Note 16 to exclude all cheaters. […] PFF can operate within a population of entities (e.g. homotypic cooperation in Figure 3D) or between populations of entities (e.g. heterotypic cooperation in Figure 1D).”

The foregoing comprises my main point – clear up the definition. However, I also think (strongly) that the paper is difficult to follow. It is of course challenging to write such a paper, because the topic is not intuitive, so a general audience needs to be given considerable background that is already well known within the field. Keeping the paper short limits the depth of this background before the paper dives into the new material. So such a paper will pose difficulties in the best of circumstances. But I nonetheless think the paper is not written to be widely accessible. Specifically, there is too much explanation embedded in details of biological systems to fully comprehend the (Abstract) points that constitute the substance of the paper. I think the paper would be more easily understood if it relied on hypothetical examples easily grasped by the reader: hypothetical examples can force the focus onto the key issues and avoid distracting details. The hypothetical examples can be borrowed from natural examples, but they should be structured to emphasize only the point being made. The description of natural examples could come late, once the main points are laid out with hypothetical examples.

Thanks for this suggestion. See my summary comments above.

Reviewer #2:

The author's example of cooperation between mitochondria and host cells illustrates how complicated the relationship between PFF and PC can be, particularly in hierarchically organised systems. However, this leads to the question of how useful is this distinction between PFF and PC for understanding the evolution of cooperation? Is this a meaningful debate? My general feeling is that the debate does matter and can help us identify mechanisms and genes underlying cooperation and mutualism. However, I think a general readership would greatly benefit from discussion of why the distinction between PFF and PC matters and isn't simply semantics. I would suggest adding this at the beginning of the paper to engage readers before the author outlines her revised criteria.

This is a good point. I have revised the text to stress the importance of defining orthogonal mechanisms. For example, in the first paragraph of the Introduction:

“How might cooperation evolve and persist despite the threat of cheaters? Identifying orthogonal mechanisms that independently promote cooperation will help us understand (and quantify) contributions from different sources. This will in turn move us closer toward predicting how persistent cooperation might be in a natural or artificial system.”

My final issue with the paper is that while the author has suggested a new criterion for distinguishing PC from PFF it is not immediately clear how this criterion could help design experiments to tell them apart. I would suggest that the author give at least one hypothetical example of how to design an experiment based on her suggested criteria in order to distinguish these two mechanisms.

I have added the following two paragraphs:

“This redefinition of PC leads to a conceptually simple experimental test. […] One can then test whether the focal entity preferentially directs benefit to the cooperative instead of cheating partner entity. If so, then PC operates.”

“To experimentally demonstrate the contribution of PFF in promoting cooperation, one will need to compare cooperator frequency when both PFF and PC are operative (e.g. in a spatially-structured environment) with when only PC is operative (e.g. in a well-mixed environment). […] In this case, one can compare cooperator frequency when PFF acts alone (in a spatially-structured environment where PC has been mutationally inactivated) with when neither PFF nor PC is operative (in a well-mixed environment with PC mutationally inactivated).”

In particular, the author mentions the example of the conflicting views over whether the figure tree and figure wasp mutualism is maintained by PC or PFF. If the author could suggest a hypothetical experiment to distinguish between PC and PFF in this example it would greatly strengthen her case that this distinction is practically useful for understanding mechanisms of cooperation.

In mutualisms between figure trees and figure wasps, PC between a figure tree and wasp offspring is identical to PFF between figs and wasp offspring. I have revised the text to hopefully make this clearer:

“We now return to the cooperation between figure trees and figure wasps. Selection against cheating pollinators can occur via PC between a focal figure tree and its internal pollinator offspring (Figure 4E, left). […] Even though Sachs et al. and Weyl et al. respectively classify this cooperation as being stabilized by PC and PFF, my redefinition suggests that both are correct depending on entities under consideration.”

Both reviewers:

The supplementary material on the Price equation does not really contribute much. It would be useful if the Price equation could be written to separate partner choice from fidelity feedback. This paper by Kevin Foster uses a direct fitness model to do just this:

I have deleted the supplementary material on the Price equation. I did cite Kevin Foster’s work in my original version and this version. To potentially extend the Price equation, we note that , the change in community-wide cooperator frequency after one round of interactions (group formation, interaction, and reproduction), is determined by:

. Here, a group with initial cooperator frequency of P will have final size π and final cooperator frequency of P + ΔP after one round of interactions, and average group size will be . Var(P) is the variance of initial cooperator frequency P across all groups, and β is a correlation coefficient. We can simply equate the Price equation to contributions from PC and PFF:

, where is the expected change in community-wide cooperator frequency in the absence of PFF and PC. But this formulation seems a bit trivial.

If we look at each term of the Price equation, PC can contribute to Var(P) (e.g. yeast flocculation, Figure 1Ciii). PC can also contribute to by diminishing the advantage of cheaters over cooperators in an interaction group (e.g. toxin-antitoxin, Figure 3D). PFF relies on stochasticity during group formation to achieve Var(P). The duration of groups will also affect both terms in the Price equation. If group duration is short, or will be small due to limited amount of repeated interactions; if group duration is long, this term may also be small since groups that initially differed in cooperator frequency will eventually look alike as cheaters increase in frequency in all groups. Group duration will also affectby affecting final group size as well as . Currently, I don’t see a biologically meaningful way of neatly partitioning the two terms in the Price equation into PC and PFF. This can be an interesting future project, but is beyond the scope of this article.

[Editors' note: further revisions were requested prior to acceptance, as described below.]

Reviewer 1 provides several comments on how to further improve the manuscript which you should address in your revised version. Reviewer 2 raises a further important point, namely the discussion of the question of intraspecific competition. The full comments of this reviewer are also attached. However, after discussion between the reviewers and the editor, we have come to the conclusion that including this into the current version of the manuscript would increase its length beyond its limits and would make it even more complex. Accordingly, we should like to ask you to drop this part entirely and reserve it to a possible follow-up paper in a more specialized journal. Hence, the comments of Reviewer 2 are only for your information.

Thank you for your consideration. I did address comments from Dr. McNally, since he raised good points about how to reach a wider readership, and since adding a few sentences did not take much extra space.

Reviewer #1: 1) Make clear early what the paper is adding to the field. The paper should start with a clear statement of what is being done without going into detail about how it is being done. (The Abstract attempts to do so, but awkwardly.) Are PFF (partner fidelity feedback) and PC (partner choice) being redefined or just clarified? Is the result a 'reconciliation' or 'redefinition?' It seems that the big change is to recognize different organizational levels within an organism, such that PC can operate at one level and PFF operate at a different level; if so, then the paper recognizes PFF and PC as mechanisms that may operate within individuals as well as between them (which is new and useful).

This is a great point. I have revised the title and the Abstract to shift the focus as per your suggestion. I have also changed “redefinition” to “revised definition” throughout the text.

Acknowledging selection at sub-organismal levels resolves controversy on pro-cooperation mechanisms

“Cooperators who pay a cost to produce publically-available benefits can be exploited by cheaters who do not contribute fairly. […] I also show that cooperation between multicellular eukaryotes and mitochondria is promoted not only by the traditionally-thought PFF between hosts and symbionts, but also by PFF and PC between sub-organismal entities.”

2) The new definition of PFF is the complementary set to PC (i.e., PFF is everything that is not PC). This is not a good approach, because it runs the risk of including standard PFF with other mechanisms that are yet to be discovered. PFF should be defined as a true mechanism. Furthermore, if the new focus on entities is useful, it should lead to a clear definition of PFF.

I have revised the definition of PFF as:

“Let us now consider potential mechanisms that promote cooperation when PC does not exist or does not operate fully Note 17 to exclude all cheaters. One mechanism orthognal to PC is PFF. PFF requires the formation of interaction groups that vary significantly in cooperator frequency due to stochasticity and that last long enough so that cooperators benefit from their cooperative acts (Figure 3B right, Figure 3C).”

3) The existing heading structure is good. I would add more. For each definition, I would have a separate heading: Definition: Entities; Definition: Partner Choice (etc.). Following the definitions of PC and PFF, I would have a heading 'How to demonstrate.' The paragraphs on how to demonstrate are useful and not something that would have occurred to me.

I have done so.

4) Something to contemplate: if PFF and PC work together within a system to maintain cooperation, do we want to recognize some additional basis for the evolution of cooperation above and beyond PFF and PC? I don't have an answer, but this might be something to think about.

I have added the following to the main text:

“Ideally, one would quantify processes involved in PC and PFF, and mathematically model how they contribute to cooperator survival. […] Otherwise, we will need to look deeper into potential causes of this mismatch.”

and the following to Figure 3D legend:

“To establish a mathematical model, one would quantify in monocultures of cooperators and cheaters their rates of birth, non-suicial death, and suicidal death under favorable (low cell density) and unfavorable (high cell density) environments. […] In a spatially-structured environment (PFF), one would also quantify the diffusion of toxins. Such a model can be used to predict the dynamics of cooperator frequency.”

Reviewer #2: I think this revision of the manuscript has greatly improved its clarity, and all of my previous comments have been well addressed. However, I have one substantial revision that I feel is necessary before publication. Reading the previous version of the manuscript I believed the author was primarily concerned with interspecies/heterotypic cooperation. However, from this revision and the author's responses I can now see that the ambition of this paper is wider, aiming to provide a framework for understanding both interspecies and intraspecies cooperation. I think this is fine, and the author generally makes a strong case. However, my major concern here is that while the relationship between the author's classification scheme and previous schemes for classifying interspecies cooperation is clear, there is insufficient discussion of the relationship between the author's classification scheme and previous classification schemes for the evolution of intraspecies cooperation. I hesitate to suggest new additions on a second round of review, but I think the author could address my points by adding a short subsection to the paper, and think that this is essential to support the lofty ambition of the paper.

This is a good point. I have added:

“Several conceptual frameworks have attempted to classify pro-cooperation mechanisms (Sachs et al., 2004; Nowak, 2006; Lehmann and Keller, 2006; West, Griffin and Gardner, 2007). […] For example in kin cooperation, Hamilton’s “kin discrimination” and “limited dispersal” (Hamilton, 1964) would correspond to PC and PFF, respectively.”

I also explain how tit-for-tat links to PFF:

“As an example of PFF, let’s consider “proportional tit for tat (pTFT)”, a game theory stradegy. […] pTFT at group size two becomes the standard TFT, and to survive cheaters, both requirements of PFF must be satisfied: repeated interactions in groups and large variation in the frequency of TFT across groups (Axelrod and Hamilton, 1981).”

The two most influential schemes for intraspecies cooperation are those provided by inclusive fitness theory and the "rule based" approach from Martin Nowak and other game theorists. In particular there have been two highly influential reviews classifying intraspecies cooperation from an inclusive fitness perspective that the author doesn't cite or discuss (West et al. 2007:

I found it quite amusing that West et al. criticized Nowak’s review in the following manner:

“First, we do not need to keep reinventing the wheel with more theoretical models that incorrectly claim to provide a new mechanism for the evolution of cooperation [12,97,98][…] This is illustrated by a recent review which

suggests five mechanisms for the evolution of cooperation [104]—three of these were just different ways of modelling the same thing (kin selection) [97,105-107], two were different forms of reciprocity.”

Yet, many mechanisms listed in West et al. Figure 2 may be regarded as kin discrimination (PC). I have now cited all these papers (see comment above). I have also added to note 7:

“One potential difference between homotypic kin cooperation and heterotypic mutualistic cooperation is that in kin cooperation, a focal individual can directly benefit from its own cooperative act (e.g. obtaining siderophores-Fe3+ after releasing siderophores to scavenge environmental Fe3+, Figure 1A). […] However, direct benefit from cooperating with self may be considered as part of “basal fitness” (the fitness of an individual when alone).”

Related to this point, one thing that I think could help increase the readers intuition for the current scheme in the context of intraspecies cooperation would be adding an explanation (possibly as a table) of how previous classifications would fall within it. There is some of this throughout the manuscript, but it is not centralised in any way. I see the main strength of the author's scheme as being the separation of causes into two meaningful biological components, that of demography (partner fidelity feedback, which captures population structure, group augmentation, non-linear demographic feedbacks) and plasticity (partner choice, which captures direct, indirect and generalised reciprocity, punishment, policing, greenbeards, kin recognition). Stating this more explicitly would help relate the new scheme to previous schemes and highlight it's main strength –

while more specific effects are difficult to partition (e.g. telling apart different forms of reciprocity), the author offers a simple way to separate the effects of demography and plasticity.

I understand what you mean. However, if phenotypic plasticity is the ability of an organism to change its phenotype in response to changes in the environment, then plasticity is also seen in PFF: a focal cooperator will grow poorly near a cheating partner, and grow fast near a cooperative partner.

If the author can add a short section addressing these points I think it will help this paper have a wider-ranging impact on the field.

See comments to your first point.