Abstract
In an individualized animal society, social bonds can foster cooperation and enhance survival and reproduction. Cooperative bonds often exist among kin, but nonkin can also develop high-investment cooperative bonds that share similarities with human friendship. How do such bonds form? One theory suggests that strangers should ‘test the waters’ of a new relationship by making small initial cooperative investments and gradually escalating them with good partners. This ‘raising-the-stakes’ strategy is demonstrated by human strangers in short-term economic games, but it remains unclear whether it applies to helping in a natural long-term social bond. Here we show evidence that unfamiliar vampire bats (Desmodus rotundus) selectively escalate low-cost investments in allogrooming before developing higher-cost food-sharing relationships. We introduced females from geographically distant sites in pairs or groups and observed that bats established new reciprocal grooming relationships, and that increasing grooming rates predicted the occurrence of first food donations, at which point grooming rates no longer increased. New food-sharing relationships emerged reciprocally in 14% of female pairs, typically over 10-15 months, and developed faster when strangers lacked alternative familiar partners. A gradual grooming-to-sharing transition among past strangers suggests that ‘raising the stakes’ might be more evident when tracking multiple cooperative behaviours as new relationships form, rather than measuring a single behavior in an established relationship. ‘Raising the stakes’ could play a similar underappreciated role across a broader spectrum of social decisions with long-term consequences, such as joining a new social group or forming a long-term pair-bond.
Significance statement Vampire bats form long-term cooperative social bonds that involve reciprocal food sharing. How do two unrelated bats go from being strangers to having a high-investment food-sharing relationship? We introduced unfamiliar bats and found evidence that low-cost grooming paves the way for higher-cost food donations. Food sharing emerged in a reciprocal fashion and it emerged faster when two strangers did not have access to their original groupmates. The bats that formed new food-sharing relationships had a history of escalating reciprocal grooming up until the food sharing began. Our finding that unfamiliar nonkin vampire bats appear to gradually and selectively transition from low-cost to high-cost cooperative behaviors is the first evidence that nonhuman individuals ‘raise the stakes’ when forming new cooperative relationships.
Text
Animal societies are fundamentally shaped by repeated interactions among individuals over time, and repeated interactions allow for individuals to choose to cooperate based on their social history across different partners (1–3). Experiments have shown that organisms as diverse as animals, plants, and fungi can prevent exploitation by shifting cooperative investments towards partners that provide better reciprocal returns (4–7). However, a greater challenge has been testing if and how individuals prevent exploitation while forming stable social bonds. Although long-term bonds are known to enhance survival and reproduction (8–12), it remains unclear how they form. How do complete strangers form a long-term cooperative relationship?
A key idea is that individuals can reduce the costs of defection by first spreading out smaller cooperative investments across time (‘parceling’ (13)) or across different partners (‘social bet-hedging’ (14)), and then gradually increasing investments in the most cooperative partnerships (‘raising the stakes’ (15)). For example, one might first assess a potential partner’s tolerance by clustering for warmth, then gain feedback by investing in low-cost social grooming, and then use the partner’s response to make higher-cost decisions about sharing food or providing coalitionary support (16). Despite its intuitive appeal for explaining how new cooperative relationships develop, evidence supporting the 20-year-old ‘raising-the-stakes’ model (15) is surprisingly scarce. An early test using the cleaner and client fish mutualism suggested that the model does not apply well to situations with severe asymmetries in partner payoffs or options (17). Studies with nonhuman primates have only tested snapshots of established relationships (18–21), rather than the formation of new ones. Human strangers do ‘raise the stakes’ when playing economic cooperation games (e.g. 22, 23), but the strategy has never been properly tested in the more ecologically-relevant context of long-term relationship formation. Doing so requires measuring the emergence of natural helping behaviors between randomly introduced strangers.
We tracked the development of cooperative relationships between previously unfamiliar wild-caught vampire bats (Desmodus rotundus). Cooperative relationships in vampire bats involve low-cost social grooming (hereafter grooming) and higher-cost regurgitations of ingested blood (food sharing)(14, 24, 25), and we show evidence that vampire bats use reciprocal grooming to gradually establish bonds that entail food donations. Our data support four key predictions: (i) new food-sharing relationships should form faster when bats have fewer alternative partners, (ii) grooming rates should predict the probability that the grooming recipient later donates food to the groomer, (iii) low initial grooming rates should increase over time only up until this first reciprocal food donation, and (iv) the emergence of new food-sharing should be reciprocal.
Female vampire bats can live for 16 years in the wild (26), and new social bonds form when an unrelated female joins a social network about once every two years (25, 27). To observe how new adult food-sharing relationships form, we captured adult females from two distant sites, Tolé (n=19) and Las Pavas (n=8), in Panamá, and we then ran 638 fasting trials in which an overnight-fasted subject could be fed by a past stranger. To test the prediction that new sharing relationships would form faster when strangers have fewer options of alternative partners (2), we compared the occurrence of new sharing when wild-caught strangers were introduced in isolated pairs (one Las Pavas and one Tolé bat), in small groups (one Las Pavas and three Tolé bats), or in one large group (see Methods, Supplementary Information (SI) Figure S1). New bonds can also form when individuals are born into a group, and these relationships might form differently. In the large mixed group, we therefore also measured the development of non-maternal cooperative relationships between 26 female adults and 13 younger captive-born bats (6 males, 7 females, 11 to 21 months old). To analyze non-independent social network data, we compared observed coefficients from general and generalized linear models (slopes β, and odds ratios OR, respectively) to expected distributions of null coefficient values generated using network and data permutations (see Methods). For clarity, we use the term ‘relationships’ for the observed network edges (directed), and the term ‘bonds’ to discuss the underlying construct that we inferred from the measured relationships (see glossary, SI Table S1).
Over 424 days and 17,621 donation opportunities, there was an 11% probability that two bats would form a new food-sharing relationship (all bats: 111 of 1036 potential relationships; females: 14% of 648 potential relationships; wild-caught adult females: 16% of 243 potential relationships; see SI). The probability of new relationships depended on availability of alternative familiar partners. Per prediction (i), when strangers were introduced as isolated pairs, new relationships emerged much faster (see SI), and we observed higher rates of grooming (β = 1.09, p = 0.04) and food sharing (β = 1.14, p = 0.004), compared to when one Las Pavas bat was introduced to three Tolé bats. In the large mixed group, bats preferentially fed and groomed their original familiar groupmates, and new sharing emerged even more gradually than in isolated pairs or small groups (SI, Fig. S2-S3).
Under the ‘raising-the-stakes’ hypothesis, strangers should make low-cost investments and use the return rates to inform future investments. Per prediction (ii), the grooming rate given by actor A to recipient B predicted the later formation of new food sharing from B back to A (OR = 2.24, p<0.0002, n = 960). The trajectory of grooming rates over time clearly differed between pairs that developed new food-sharing relationships versus pairs that did not (interaction: OR = 1.57, p<0.0002, Fig. 1).
If reciprocal grooming serves to ‘test the waters’ for more costly cooperative investments, then the slope of the increase in grooming given should be greater before the food-sharing relationship begins. Per prediction (iii), grooming rates from bat A to B increased only before the first day that B fed A (Fig. 2). As expected, grooming rates were highest between bats that formed reciprocal food-sharing relationships, intermediate in one-way sharing relationships, and lowest among bats that never shared food (SI Fig. S4), and grooming rates towards previously familiar donors were greater than towards either previously unfamiliar donors or previously familiar non-donors (SI Fig. S5).
Finally, per prediction (iv), the formation of new food-sharing relationships was reciprocal. Directed food-sharing networks were more symmetrical than expected if new donations were random, even when controlling for kinship (estimated via maternal pedigree and microsatellite genotyping, see Methods; MRQAP-DSP; reciprocal sharing: β = 0.40, p<0.0002, kinship: β = 0.07, p = 0.10; SI).
Individuals can ‘raise the stakes’ quantitatively, by increasing the magnitude of cooperative investments, and qualitatively by performing higher-cost cooperative behaviors. Female vampire bats appeared to both increase grooming rates and transition from low-cost grooming to high-cost sharing. The rarity of new food-sharing relationships corroborates past evidence that food regurgitations are energetically costly and that social bonds require investments of time and energy (14, 25, 28–30). Our finding that the development of new food-sharing relationships between wild-caught adults is influenced by partner behavior and availability provides the clearest evidence to date that nonkin food sharing in vampire bats is not a mere byproduct of kin selection (30, 31).
Prior to this study, some previous evidence for ‘raising-the-stakes’ in nonhuman social relationships came from observations of grooming among familiar male chimpanzees after the death of an alpha male (18). The authors suggested that, during this period of social instability, these groupmates may have needed to re-establish their relationships, and that a diminishing threat of violence led to the increasing rates of grooming (18). Although the increase in grooming rates is consistent with each male ‘raising the stakes’ to assess the risk of aggression from their grooming partner, it might have also resulted from a general decline in vigilance against possible aggression from any other groupmate.
The empirical evidence for the ‘raise-the-stakes’ model (15) has been scarce in part because it is a variation on the classic ‘tit-for-tat’ strategy in the iterated prisoner’s dilemma (1), a model which is difficult to test unambiguously using natural forms of cooperation (4, 32). Contingently alternating forms of reciprocity similar to ‘tit-for-tat’ have been demonstrated using experiments with trained instrumental tasks (e.g. pulling a lever to deliver food) and payoffs accruing in distinct rounds (4–6). However, the ‘tit-for-tat’ model of reciprocity must not be operationalized too literally. It excludes many factors crucial to real-world cooperation, such as partner choice, social bonding, different service types, and the cost-benefit asymmetries resulting from demography, market effects, or social dominance (4, 7, 32). Moreover, long-term cooperative relationships do not produce strict ‘tit-for-tat’ helping. For example, primates with stronger bonds show less evidence for short-term contingencies in grooming (9). Models of cooperative relationships as a ‘social exchange’ must therefore consider the cognition underpinning the helping decisions. Social bonds and ‘trust’ result from heuristics that reduce many cooperative interactions to a single positive association, just as similar generalizations in economic choices lead to ‘brand loyalty’ (33). Even in humans, which possess the most extensive suite of cognitive specializations for social interactions (34), many of the neural processes that underlie simple operant conditioning also underlie learning from people (35) or about them (36).
The relevance of our findings extends beyond high-cost cooperative behaviors. For example, in some species, courtship behaviors could be seen as a short-term investment in the formation of longer-term pair bonds with substantial fitness consequences (e.g. 37). Similarly, the role of mere physical contact as a low-cost method for building tolerance and trust might be more general than currently recognized. The key role of grooming for relationship maintenance in primates is well established, but growing evidence suggests that tactile behaviors can reduce fear and encourage cooperation in many other species of mammals, birds, and fish (e.g. 5, 38–44). New methods for tracking formation of social bonds at fine temporal scales (45) will provide new opportunities to test whether gradual escalation of proximity and body contact is a widespread mechanism for ‘testing the water’ of a potential relationship.
Methods
Animals
We used 41 common vampire bats (Desmodus rotundus) as subjects, including 19 female bats captured exiting a roost in Tolé, Panamá; 8 female bats captured foraging at a cattle pasture in Las Pavas, Panamá about ~215 km from Tolé; and 14 captive-born bats (8 females, 6 males). We studied adult females and their young, because these individuals form the basis of food-sharing networks in the wild, whereas adult males compete for access to territories and females and do not form stable bonds as often. To ensure familiarity within groups and unfamiliarity between groups, we housed the groups separately (Tolé bats for 6 months and Las Pavas bats for 2 weeks) before the study began. Bats were marked with subcutaneous passive integrated transponders (Trovan Ltd. USA) and a visually unique combination of forearm bands (Porzana, National Tag, and birdbands.com). To feed bats, we provided refrigerated or thawed cattle or pig blood defibrinated with sodium citrate and citric acid.
We used a 3-4 mm biopsy punch to collect tissue samples in 80% or 95% ethanol, then used a salt–chloroform procedure for DNA isolation, and a LI–COR Biosciences® DNA Analyser 4300 and the SAGA GT allele scoring software to genotype individuals at 17 polymorphic microsatellite loci. Allele frequencies were based on 100 bats from Tolé and 9 bats from Las Pavas, respectively. Genotypes were 99.9% complete. To estimate genetic relatedness, we used the Wang estimator in the R package ‘related’. To estimate kinship, we assigned a zero kinship to known unrelated individuals from different sites and to individuals with negative pairwise relatedness, and we assigned a kinship of 0.5 for known mother-offspring pairs or pairs with genetic relatedness estimates greater than 0.5. For all other pairs, we used genetic relatedness as the estimate for kinship.
Experimental design
We induced allogrooming and regurgitated food sharing using a fasting trial, in which a focal subject was isolated from the group without food for a night and a day, then released back to the group of fed bats for 1 hour the following night. During the hour, all grooming or sharing interactions with the subject were recorded using an infrared (IR) light and an IR-sensitive video camera. Each donation bout was estimated by the number of seconds that the unfed subject spent licking the mouth of a particular groupmate, starting at 5 seconds of uninterrupted licking. The dyadic donation size for that trial was estimated as the sum of these bouts. Bats were weighed before and after trials, and total observed mouth-licking durations predicted weight gain. Fasted subjects gained 51 mg of mass per minute of mouth-licking (R2= 0.75, bootstrapped 95% CI: 45 to 57 mg/min, n=619, p<0.001).
We conducted fasting trials in each group during three experimental phases (SI Fig. S1). First, we conducted 57 “baseline” trials to assess preliminary sharing rates between the 19 Tolé bats housed in a 1.7 × 2.1 × 2.3 m outdoor flight cage in Gamboa, Panama (1 group, 3420 possible sharing interactions). Second, we conducted 106 “controlled introduction” trials to assess possible formation of new food-sharing bonds between past strangers introduced as either an isolated pair (one Las Pavas bat and one Tolé bat) or a quartet (one Las Pavas bat and three Tolé bats), housed in a 28 × 28 × 40 cm clear plastic observation cage (10 pairs and 8 quartets). These controlled introductions provided for 162 opportunities for new sharing between previous strangers (SI Table S2). Finally, we conducted 532 “mixed-group” trials to assess the formation of new sharing relationships when all bats were housed together in the flight cage described above (19 Tolé, 7 Las Pavas, and 14 captive-born bats). These introductions in this combined group provided 11,823 more opportunities for new sharing.
Statistical analyses
During the baseline and mixed-group trials, we estimated food sharing as the number of seconds that a fasted subject spent mouth-licking a fed groupmate. During the controlled introduction trials, however, when bats were forced in close proximity, we saw a greater frequency of begging, defined as mouth-licking that is clearly not food-sharing because the partner is turning away from the mouth-licking bat and the mouth-licking bat does not gain the weight that would be expected from food-sharing. To be conservative when measuring sharing, we therefore did not count mouth-licking as food sharing during the controlled introduction trials unless the subject weighed more than expected based on the average weight change for bats that did not perform any mouth-licking.
Durations of sharing and grooming were lognormal. To create a standard index of grooming rates, we therefore transformed the total duration of directed dyadic interactions in each trial using natural log (x+1). We call these measures of the log duration per hour “rates”. When interaction bout length and probability had different meanings, we decomposed rates into two separate response variables: amounts (the magnitude of nonzero rates in a trial) and probabilities (the presence or absence of a nonzero rate in a trial). We used permutation tests with 5,000 permutations for all hypothesis testing (p-values) and bootstrapping for all 95% confidence intervals. To approximate two-sided p-values, we doubled any one-sided permutation p-values that were below 0.05. Null distributions were not always centered on zero due to structure in the data, so caution must be taken when considering the observed coefficients.
To test for ingroup-outgroup biases in sharing for each site, we calculated observed coefficients for the effect of being from the same capture site on actor grooming rates, then calculated expected coefficients by permuting the grooming rates within each actor. To test the effects of kinship and reciprocal grooming on the formation of new food-sharing relationships in the mixed-group trials, we used multiple regression quadratic assignment procedure with double semi-partialing (MRQAP-DSP) via the netlogit function in the sna R package. This procedure uses generalized linear models via the glm function in lme4 package to calculate observed coefficients and uses network-level permutations to get expected coefficients. Since MRQAP-DSP cannot test interaction effects, we compared observed and expected interaction coefficients using permutations in which we shuffled trial rates given by the actor among different possible receivers and then shuffled the trial rates received by the receiver among different possible actors. If the interaction coefficients were significant (p<0.05), we conducted separate MRQAP-DSP tests within each group.
To test whether interaction rates changed over time, we generated expected coefficients for general or generalized linear models by permuting the order of interactions within each potential relationship (actor-receiver dyad). One captive-born bat died for unknown reasons during the mixed-group trials, and we removed it from all temporal analyses. To test for evidence of reciprocal sharing, we used MRQAP-DSP to test if the matrix of new sharing in the mixed-group trials was predicted by reciprocal sharing when controlling for kinship. As an additional test, we also counted the occurrence of both novel sharing and reciprocal sharing for all new potential relationships, then counted the same number after randomizing the presence of sharing across potential relationships.
Data availability
Behavioral data, genotypes, and R code are available as supplementary information.
Author contributions
Conceptualization, GC; Methodology, GC; Investigation, GC, RC, JV; Genotyping, SR; Statistical analysis, GC; Original draft, GC; Review & editing, GC, RC, JV, SR, DF, RP; Funding acquisition, GC, SR, RP; Resources, RP; Supervision, GC, DF, RP;
Summary of supplied data
Data S1. genotypes.csv
Microsatellite genotypes used to assess relatedness.
Data S2. vampire_maternal_kinship.csv
Maternal pedigree data
Data S3. new_bonds_data.Rdata
Food sharing and allogrooming data
Script S1. new_bonds_analysis15.R
R script for analyzing data
Temporary link for reviewers: https://www.dropbox.com/sh/5z1tb1zvj5v7fqt/AAA4D7fkeVlT7V_SArvy8D51a?dl=0
Acknowledgments
We thank the Smithsonian Tropical Research Institute for logistical support. Isabelle Waurick conducted the molecular lab work. Work was supported by a Smithsonian Postdoctoral Fellowship, a Humboldt Research Fellowship, a Smithsonian Institution Scholarly Studies Grant and a grant from the National Geographic Society Committee for Research and Exploration (WW-057R-17).