Revisiting non-offspring nursing: allonursing evolves when the costs are low

Abstract

Allonursing, the nursing of another female's offspring, is commonly assumed to have evolved through the benefits of kin selection or reciprocity. The evolution of allonursing may also be influenced by variation in the possible costs to allonurses. The relative influence of costs and benefits on the incidence of allonursing in mammals remains unexplored. We show, using comparative analyses, that where females group with kin, the presence or the absence of allonursing is not associated with further variation in relatedness. Allonursing is most common where females produce litters; here the relative investment per offspring is low, and the costs of nursing additional young are likely to be reduced. Our results suggest that variation in the potential benefits is not associated with the distribution of allonursing, but that allonursing can quickly evolve when the costs to allonurses of nursing additional offspring are low.

1. Introduction

Allonursing, the nursing of non-descendant infants, occurs in a wide variety of mammals where females live in groups, including primates, cetaceans and canids [1]. Allonursing may confer substantial benefits to offspring in terms of growth, survival and the transfer of immune compounds [2,3]. These benefits come at a cost to the allonurse, as lactation is highly energetically demanding [4]. Several hypotheses have been suggested to explain the evolution of this costly behaviour, most focusing on potential adaptive benefits to be derived from allonursing: females may nurse to gain experience of maternal care; nursing may be a means of evacuating excess milk which may be painful, or impede mobility; nursing may increase the likelihood of reciprocity; or, nursing may provide indirect benefits where females are able to preferentially nurse related offspring [5,6].

Contrasts in the relative frequency of allonursing across taxonomic groups are also likely to reflect variation in the costs of nursing. For example, Packer et al. [1] showed that in wild mammals, non-offspring nursing is most common in species where females produce multiple offspring in litters (polytocous) relative to where females produce single offspring (monotocous), probably because in litters the milk provided to each offspring is a lower proportion of the total, reducing the cost of nursing additional young. In addition, in litter-bearing species females may have an increased likelihood of having an excess of milk due to loss of offspring or producing a relatively small litter [7]. Comparative models show that across mammals, for a given mass, females produce the same amount of offspring mass regardless of litter size, suggesting that total reproductive investment per offspring might be lower in polytocous species [8]. Producing milk for an additional offspring is therefore likely to be less costly in polytocous species, possibly allowing allonursing to evolve.

Here, we extend the work of Packer et al. [1] by analysing the ecological correlates of non-offspring nursing in line with a new question: has non-offspring nursing evolved only where it is likely to generate substantial fitness benefits or has it been more constrained by costs? Where females are related, indirect benefits are guaranteed—but variation in the magnitude of probable returns exists. If the evolution of allonursing is driven by variation in potential benefits, it should be most common where the benefits are likely to be highest: where individuals are closely related. If allonursing is constrained by cost, it should occur where costs are likely to be lowest: where food resources are reliable and cheaply obtained; and where relative investment per offspring is likely to be low (multiple offspring produced per breeding attempt, several breeding attempts a year). Allonursing might also be influenced by opportunism costs, and may therefore be more likely to occur when several females breed concurrently in close proximity over a short breeding season [5], or by offspring development and ecology, which may be reflected in milk composition [9]. Differences in group structure (for example, whether species breed cooperatively) may further influence the benefits and costs of allonursing.

2. Material and methods

Using the criteria of social system classification specified by Lukas & Clutton-Brock [10], we defined 119 wild mammalian species as group-living. Thirty-four of these were classed as singular cooperative breeders [10]: one female is the primary breeder, subordinate individuals help to rear her offspring, and allonursing is commonly a result of pseudopregnancy [11] or failed subordinate pregnancies [12]. In non-cooperatively breeding group-living species (n = 85), on the other hand, allonurses also nurse their own young. We classed species as allonursing if females regularly allonurse in wild populations. A lack of evidence of allonursing was taken as evidence of its absence only where sufficient behavioural studies exist. We expect this protocol to be sufficiently rigorous as allonursing is likely to be reported, but we acknowledge a potential bias against species in which allonursing may not be easily observed, for example in small burrowing or nesting rodents. To ensure that we did not incorrectly classify a species because of insufficient study, we excluded species for which insufficient behavioural or wild data were available (n = 14, indicated in the electronic supplementary material).

We compiled data on group structure (cooperative breeders, non-cooperative group-living species), mean litter size, number of litters produced per year, average relatedness within groups, milk composition (sum of percentage protein, fat and sugars) and diet. We defined species as seasonal breeders if breeding is restricted to a period of six months or less. All continuous variables were log-transformed before analyses. Data and references are provided in the electronic supplementary material.

A model was constructed to test the effect of group structure (whether species breed cooperatively, or not) on allonursing incidence using the package ‘MCMCglmm’ [13], specifying whether allonursing occurs or not as a binary dependent variable. The updated mammalian supertree [14] was used as the basis for phylogenetic analyses. We ran this model with and without the phylogenetic tree specified, and compared models using DIC [15]. Including phylogeny did not improve model fit, suggesting that phylogenetic similarity does not explain residual variance. We did not account for phylogeny in subsequent models.

Singular cooperative breeders and group-living species that do not breed cooperatively differ in the likelihood that allonurses also nurse young of their own [10]; the costs and benefits of allonursing are consequently likely to differ between these groups. We therefore split the data into cooperative and non-cooperative breeders, and in each subset constructed preliminary models using the lme4 package in R [16] to look at factors that might influence: (i) costs that might prevent females from allonursing (diet, litters produced per year, mono/polytocy); (ii) the probability that other females in the group will have offspring (litters produced per year, number of adult females, breeding season); and (iii) the potential benefits of allonursing (average within-group relatedness). As all cooperative species were polytocous, we used litter size instead of mono/polytocy in analyses of cooperative species. We tested the effect of milk composition on allonursing incidence in separate models owing to low sample size. Where there was collinearity between variables, the variable that explained most variation (tested using AIC comparison of single-parameter models) was included and the other discarded. Global models were then defined for each subset by taking any significant variables in the preliminary models, and setting them as explanatory variables in a global model.

3. Results

Proportionally fewer cooperatively breeding species allonurse (29%, n = 24) than non-cooperative group-living species (45%, n = 83). This difference was not statistically significant (MCMCglmm p = 0.13).

In cooperative breeders, carnivorous species were significantly more likely to allonurse than omnivorous species (table 1a). No herbivorous cooperative breeders allonursed. Preliminary models also suggested a positive correlation with litter size (model I): litter size correlated significantly with diet, which was a better predictor of allonursing according to AIC. There was a non-significant trend for species with higher within-group relatedness to allonurse (model III).

Table 1.

Generalized linear models testing factors associated with costs, benefits and the likelihood of other females also having offspring on the incidence of allonursing within (a) cooperatively breeding species and (b) non-cooperatively breeding species. Intercept estimates are italicized throughout; global models are reported in bold.

		estimate	s.e.	Z	p
(a) cooperatively breeding species
(i) costs predicted: diet + litter size (n = 20)	intercept	−6.87	7.22	−0.95
	litter size	6.36	5.56	1.14	0.07
	diet				<0.01
	herbivore	−21.49	4809.16	−0.004
	omnivore	−4.02	2.37	−1.70
(ii) concurrent offspring predicted: litters per year + number of adult females (n = 18)	intercept	−0.45	0.48	−0.94
(iii) benefits predicted: average relatedness (n = 13) global model predicted: diet (n = 24)	intercept	0.15	0.56	0.28
	intercept	1.10	0.82	1.35
	diet				<0.01
	herbivore	−19.66	3261.32
	omnivore	−3.50	1.33
(b) non-cooperatively breeding species
(i) costs predicted: diet + mono/polytocy (n = 69)	intercept	−0.97	0.35	−2.74
	mono/polytocy	1.61	0.53	3.01	<0.01
(ii) concurrent offspring predicted: litters per year + breeding seasonality (n = 47)	intercept	−0.15	0.31	−0.48
	litters per year	1.43	0.71	2.03	<0.05
(iii) benefits predicted: average relatedness (n = 31) global model predicted: mono/polytocy (n = 78)	intercept	0.60	0.38	1.59
	intercept	−0.93	0.33	−2.85
	mono/polytocy	1.58	0.50	3.18	<0.001

In non-cooperatively breeding species, polytocous species were more likely to allonurse than those producing single offspring (table 1b): allonursing has been recorded in 66% (21/32) of litter-bearing group-living species versus 31% of monotocous species (16/51). Preliminary models also suggested a positive correlation with litters produced per year (model II). Litters per year and mono/polytocy correlated significantly, and mono/polytocy was a better predictor of allonursing according to AIC. Variation in within-group relatedness is not related to the incidence of allonursing in group-living species.

Milk composition was not an important predictor of allonursing in cooperative species (GLMM Z₇ = 1.17, p = 0.13) or non-cooperatively breeding species (GLMM Z₂₃ = 1.13, p = 0.23).

4. Discussion

In species where helping behaviour between females is likely to lead to indirect fitness benefits, the evolution of allonursing appears to be constrained by costs rather than being explained by differences in the likely returns. Allonursing was very common in non-cooperatively breeding species that are polytocous, in agreement with the results of Packer et al. [1] which suggest that non-offspring nursing in monotocous species, where investment per offspring is high and diverting care to other young is likely to be prohibitively costly, is best understood as milk theft. Our results suggest that allonursing occurs in a wide range of species and can quickly evolve when relative investment per additional offspring is low [6]. A two-species study of social carnivores reached similar conclusions: lions are more likely to nurse non-offspring when they have excess milk and when the needs of their own young are reduced, possibly a low-cost by-product of communal defence against infanticide [7].

Allonursing was not more common in cooperative breeders, which had a higher mean within-group relatedness (x̄ ± s.d., r = 0.42 ± 0.09) than non-cooperatively breeding species (r = 0.17 ± 0.12). Within these groups, allonursing incidence also did not correlate with relatedness. Within-species studies show that females that nest in kin groups do not necessarily preferentially nurse close kin [17,18], suggesting that where females are likely to be related to some degree, directing care towards close kin may provide limited extra benefits. Similarly, variation in relatedness may generate little variation in the potential benefits of allonursing between species, and may therefore be unlikely to drive differences in allonursing incidence. However, data on relatedness within wild groups were not available for many species, notably the primates. Further study of patterns of relatedness in these groups may help to illuminate or confirm our results.

Among cooperative breeders, allonursing was most common in carnivores. In these species, reproduction may coincide with short periods of relative trophic abundance [19,20], which could potentially reduce the costs of subordinate reproduction for dominant breeders [21]. Overall, however, the incidence of allonursing in cooperatively breeding species was surprisingly low. The mean number of adult females per group in this sample was low (1.93, range 1–5), suggesting that allonursing may feasibly be restricted by opportunity costs and may only occur in species where groups regularly have more than one female of breeding age. In addition, interspecific differences among cooperative breeders in the likelihood of successful subordinate female breeding could result in further variation in the presence of potential allonurses; low incidences of allonursing in cooperative breeders may represent physiological constraints rather than adaptive behavioural decisions.

Our findings indicate that reproductive cooperation occurs in many species where females live with kin, across taxonomic groups and ecological conditions. However, our results do not permit us to estimate the potential benefits that reproductive cooperation might confer to females, and more detailed studies are necessary to investigate the role that allonursing might have in the evolution of female sociality.

Data accessibility

All data are available in the electronic supplementary material.

Funding statement

D.L. received funding from ERC grant no. 294494-THCB2011.