Competitive social feedback amplifies the role of early life contingency in male mice

Abstract

Contingency (or “luck”) in early life plays an important role in shaping individuals’ development. By comparing the developmental trajectories of functionally genetically identical free-living mice who either experienced high levels of resource competition (males) or did not (females), we show that competition magnifies early contingency. Male resource competition results in a feedback loop that magnifies the importance of early contingency and pushes individuals onto divergent, self-reinforcing life trajectories, while the same process appears absent in females. Our results indicate that the strength of sexual selection may be self-limiting, and they highlight the potential for contingency to lead to differences in life outcomes, even in the absence of any underlying differences in ability (“merit”).

Contingency (colloquially, “luck” or “chance”) has long been recognized as an important determinant of outcomes in both biological and social sciences (1–15). The contingency hypothesis posits that an individual’s behavior, health, social position, and fitness are strongly dependent on events and experiences that it can neither control nor predict (5, 16–20). The outcomes of contingent events in early life are often especially important, as they can set individuals onto divergent, self-reinforcing trajectories (1, 2, 5, 17, 18, 20). Recent evolutionary theory has argued that luck in an individual’s life, particularly in early life, can outweigh individual quality in determining lifetime reproductive success (1, 2).

Many animals naturally live within larger social groups, such that contingency in event outcomes is inextricably tied to individuals’ relationship to the behavior of others within societies (5, 21–25). Through repeated social interactions, individuals adopt a consistent set of social phenotypes [i.e., their “social niche” (22, 24, 26)]. We hypothesize that competitive social processes magnify the importance of contingency in early life. For example, animals that begin with zero or small differences in competitive ability may differ in their access to resources owing to variation in contingent dominance or territorial interactions (19, 27–30). The resulting increased resource access for a subset of the population then improves those animals’ condition relative to those with reduced resource access, further entrenching the initial differences and magnifying the importance of early contingency (19, 21, 30–33). This process is analogous to the “Matthew effect” in the social sciences, a phenomenon in which individuals or institutions that achieve early success tend to achieve ever greater success in the future (34–36).

Experimentally studying the role of contingency in individual outcomes is achievable with the use of “replicate individuals” that allow researchers to effectively “replay the tape of life” for a single genotype under different circumstances (37, 38). Within the lab, studies of functionally genetically identical animals indicate that contingent microenvironmental differences during development cause between-individual differences in early behavior, which then increase in magnitude over time (39–45). Yet assessing the ways in which competitive social processes interact with early contingency requires the study of replicate individuals living under realistic, complex, dynamic social conditions—requirements that cannot be readily met under standardized laboratory conditions (46–48). We overcame this limitation by studying the development of individuality in ecologically relevant spatial and social behaviors from infancy through adulthood in age-matched, isogenic (hereafter, “genetically identical”) mice living outside in a shared, seminatural field environment (figs. S1 and S2).

Males develop individual adult behavioral phenotypes earlier than females

In free-living C57BL/6J lab mice, males compete for territorial control and resource access, while females do not (49, 50). Females appear unconstrained in their movement, whereas males repeatedly visit only a small subset of available space (49, 50) (fig. S2). Males also spend much less time engaged in spatiotemporal overlap with each other than do females (49, 50) (fig. S2). And territory-less males have higher mortality and attain less access to females than do territory holders (49, 50) (fig. S2). We hypothesized that this difference in competitive experiences causes males to diverge onto self-reinforcing developmental trajectories as some win competitive interactions and others lose. We expected this same dynamic to be absent in females.

In this study, we monitored 16 litters of the C57BL/6J inbred mouse strain (n = 104 pups, 90 of which survived to adulthood) from infancy through adulthood outside in a large (~560 m²) enclosure that emulated the natural foraging and social environment of commensal house mice (fig. S1, A to C). We placed 2-week-old litters and their mothers in the enclosures within one of 16 identical “resource zones” containing food and shelter, monitored with radio-frequency identification (RFID) antennae. Our goal was to minimize the impacts of genetic variation, maternal effects, and physical microenvironmental differences, thereby restricting environmental variation to social dynamics as much as possible. We inferred periods of social overlap using an established workflow to translate RFID positional data into estimates of the duration of social aggregations within each of the monitored zones (fig. S1D) (49). On the basis of 7.4 million RFID reads, we traced the development of 17 social and spatial phenotypes from infancy through adulthood (14 to 58 days; table S1).

We first showed that genetically identical animals display distinct, individually repeatable social behaviors under ecologically relevant contexts, an open question in behavioral ecology (37, 38, 44, 45, 51, 52). We measured repeatability as the proportion of a phenotype’s total variation in each sex that was explained by individual identity (53) over a sliding 5-day age window, after controlling for maternal and litter identity. We detected significant repeatability across 5-day periods in all measured phenotypes, emerging as early as age 21 days (fig. S3), roughly 1 month earlier than reported for spatial behavior of female populations of this strain in enriched lab vivaria (11, 27). Although essentially all phenotypes that we measured became highly repeatable in both sexes over these short time frames, many more phenotypes became significantly predictive over longer time periods (15 to 25 days) in males than in females (fig. S3I).

We next assessed the developmental timing at which males and females assumed their individually distinct adult behavioral phenotypes. Many of the phenotypes that we measured covaried, so interpreting our data required us to flatten our phenotypes into orthogonal measures of behavior. We used principal components analysis to reduce the dimensionality of 16 of our 17 behavioral phenotypes into two principal components (PCs) that accounted for a majority of the total variation in our dataset (57% across PC1 and PC2; “time of first nightly transition” values were missing in young animals who did not move between zones). PC1 is a course metric of social and spatial exploration, being inversely related to both spatial and social fidelity. PC2 is a metric of social connectivity, being positively correlated with all prosocial traits, including number of social partners and the strength of social connections (table S2).

We identified animals’ final adult behavioral phenotypes by taking the average of each individual’s PC1 and PC2 scores during the last 3 days of the experiment (age 56 to 58 days; Fig. 1A). For each sex, we then assessed the relationship between individuals’ phenotypes at earlier time points and these final adult behavioral phenotypes by building linear regressions between final adult phenotype and individuals’ average phenotypes over 5-day, nonoverlapping windows (e.g., age 21 to 25 days).

Fig. 1. Males adopt their adult phenotypes earlier than females.

(A) Traces of observed individual behavioral PC1 and PC2 values, smoothed over 5 days, across animals’ development. (B) Traces of the development of individual differentiation from a single run of agent-based simulations in which individuals’ phenotypes develop either in the presence or absence of competitive feedback mechanisms. (C and D) Quantification of the patterns in (A) and (B). (C) (Top row) The correlation between earlier and final adult behavior (y axis) is stronger in males for PC1 (left column) and PC2 (right column). Asterisks denote significance of the correlations depicted in each point (linear model, Bonferroni corrected for 16 comparisons in each sex, *P < 0.05, **P < 0.01, ***P < 0.001). (C) (Bottom row) The slope of the relationship between earlier and final adult behavior (y axis). The slope of this relationship is consistently closer to 1 for males than for females. (D) Results from 1000 iterations of the agent-based simulation.

For both PC1 and PC2, males assumed their individual adult behavioral phenotypes earlier than females, consistent with our analysis of single behaviors (Fig. 1C and fig. S3I). Male behavior became predictive (P < 0.05, Bonferroni corrected for 16 hypothesis tests) of final adult behavior earlier than did female behavior (PC1: 26 days versus 26 days, PC2: 31 days versus 46 days; Fig. 1C). The strength of the correlation between earlier and later behavior for the same individual was substantially higher for males than for females across development (PC1: days 31 to 50; PC2: days 16 to 50; Fig. 1C). Moreover, the slope of the linear relationship between earlier behavior and adult behavior was closer to 1 for males than for females, and this strong relationship developed at an earlier age for both PC1 and PC2 (Fig. 1C and fig. S4). The above results were unchanged when we generated separate sex-specific PC values instead of including both sexes in a single analysis (fig. S5).

These differences in the developmental timing of behavioral individuality are unlikely to be explained by sex differences in the timing of sexual maturation, as males and females of this strain display similar timings of the onset of puberty and successful mating (54–59). Differences in final adult values of PC1 and PC2 were also not predicted in either sex by maternal identity or the area of the enclosure in which animals started the experiment (table S3).

We next assessed whether these sex differences in the developmental timing of individuality could be explained by sex-specific differences in the importance of competitive feedback in amplifying the impact of early contingency. We built a quantitative agent-based model to generate predictions of how competitive processes shape the long-term phenotypic impacts of early contingency (Fig. 1, B and D). We assumed that all individuals began with the same value of a phenotype, which then changed at discrete timesteps across their lives. Without competitive feedback, phenotype values increased or decreased randomly. With competitive feedback present, an individual’s phenotype increased in value if it won a competitive interaction with a randomly chosen individual and declined if it lost. The probability of winning the interaction depended on the relative phenotypic values of the two interactants. With competition present, event outcomes are therefore least predictable at the beginning of life and become more predictable as phenotypic differences emerge and are reinforced through competitive feedback (see supplementary materials).

The qualitative results of the simulation closely mirrored our observations of differences in the development of individuality in males and females in our system (compare Fig. 1A to Fig. 1B and Fig. 1C to Fig. 1D). When competitive feedback loops are present, (i) the correlation between behavior at any given time and behavior at the end of the modeled period is stronger, and (ii) the slope of the relationship between earlier and later behavior is closer to 1.0 (Fig. 1, C and D). Thus, the sex difference that we observed in the development of behavioral individuality could be entirely explained by differences in sex-specific competitive processes that amplify contingent early life differences in phenotype.

Competition for resource access acts as a sex-specific competitive feedback loop

We next assessed whether males and females displayed differences in the strength of resource competition in a fashion that would support this putative sex-biased competitive feedback loop. We estimated individuals’ nightly resource access by calculating a nightly “resource competition score” for each animal (see methods in the supplementary materials). Consistent with males facing higher levels of competition than females, the resource competition score varied more among males than it did among females (Fig. 2A). This difference emerged concurrently with the onset of sexual maturity, the period when we expect intrasexual competition to increase in intensity among male mice (fig. S2C).

Fig. 2. Competition for resource access shapes males’ phenotypes through a sex-specific competitive feedback loop.

(A) Traces of individual resource competition scores, smoothed over 5 days. Black lines indicate sex-specific means. Vertical dotted lines indicate approximate ages of weaning (“juvenility,” 21 days), sexual maturation (“adolescence,” 35 days), and onset of conceptive mating (“adulthood,” 46 days). (B) Male resource competition scores are more variable across individuals than female scores (see also fig. S2). Horizontal dashed line segments indicate the average coefficient of variation across each developmental stage. (C) Male adult resource competition scores are predicted by small differences in body mass in early life, a difference that is magnified over time. The relationship is absent in females. The y axes represent deviations from age-predicted body mass, points represent averages ± SEM. (D) Adult resource competition score strongly predicts an integrative measure (PC1) of the 16 other spatial and social phenotypes in males but not in females.

Two additional pieces of evidence are consistent with males, but not females, experiencing strong competitive feedback that set them on self-reinforcing divergent life trajectories. First, small individual differences in early body mass (days 21 to 27) predicted adult (days 46 to 58) resource competition scores for males (P < 0.05; Fig. 2C and fig. S6) but not for females. Consistent with individuals starting out on an approximately “even playing field,” very minor differences in infant body condition before release into the field (days 12 to 14) did not predict future resource access in either sex (P > 0.05; fig. S6). Differences in body mass between males with differential resource access then increased over the course of development (age × adult resource competition score interaction: P = 0.004; Fig. 2C and fig. S6), consistent with males experiencing a competitive feedback loop that increased the condition of winners relative to losers. This developmental pattern was absent in females (P = 0.19; Fig. 2C).

Second, male resource competition scores strongly predicted the other male behavioral phenotypes, whereas the same was not true of females, indicating that competition in males has major impacts on males’ daily behaviors and access to mates. Specifically, we performed a second principal components analysis using the 16 phenotypes other than resource competition score that we measured in adulthood (age 46 to 58 days) to obtain a single integrative measure of animals’ other adult behavior (PC1 explained 38% of the total variation in this dataset). Males’ average adult resource competition scores strongly predicted this adult PC1 value [P < 0.0001, coefficient of determination (R²) = 0.38; Fig. 2D], whereas females’ scores did not (P = 0.86, R² = 0.00; Fig. 2D). The same conclusion holds if we assess individual adult phenotypes, including access to mates (table S4).

Discussion

Our results provide empirical support for the hypothesis that contingency (or “luck”) in early life can have a major and competition-dependent impact on the development of animals’ individual differences in social and fitness-relevant phenotypes. The outcomes we observed are not explained by genotype or macroenvironment (which we controlled), maternal identity, or initial location in the arena. To the extent that early success or failure in social competition was determined by individual quality, the factors that generated individual differences were contingent experiences. Competitive feedback loops magnify the importance of contingency experienced early in life, such that young free-living male lab mice enter onto divergent, self-reinforcing developmental trajectories. Our results match an agent-based simulation of expected differences in the developmental timing of individuality in the presence and absence of competitive feedback.

We expect the sex-specificity and intensity of competitive feedback loops to vary across different contexts and species, depending on social behavioral ecology. Here, the most relevant differences in social competition happen to be related to sex, but this need not be the case [e.g., in wild house mice, females compete for space and mates under some conditions (60, 61)]. In female-dominant species, the reverse pattern may be present, such that females’ outcomes would be more dependent on early luck than males’ outcomes (62, 63). In other settings, perhaps including some or many human populations, both males and females may experience high or low levels of competition, or this competition may vary across time or space.

Our results suggest an inevitable limitation of sexual selection’s ability to shape behaviors. Intrasexual selection relies on within-sex competition resulting in differential reproductive success, and for variation in this success to be heritable (64–66). But here we have shown that intrasexual competition also magnifies the importance of early contingency in later life outcomes in the sex expressing that competition. As the importance of luck in determining individual phenotypic outcomes increases, selection’s ability to cause evolution declines (1, 2). Thus, intrasexual selection may be self-limiting, as an increase in the importance of competition in a single sex leads, in turn, to an increase in the importance of contingency in determining individual outcomes in that same sex. This increased importance of luck may help to explain why sexual selection fails to fully deplete genetic variation [i.e., the lek paradox (67–69)].

Our results provide a strong biological analog to the Matthew effect, an often-observed phenomenon in social science whereby small individual advantages earlier in life are correlated with ever larger advantages over time (34–36). Such processes are understood to be the result of social feedback mechanisms, by which an individual’s initial success improves their opportunities for future success as well as the perception by other members of society of the individual’s potential for success (36, 70). Our results suggest that the Matthew effect (i) may have a biological origin, (ii) is especially likely to occur in highly competitive environments or among groups that face high levels of competition, and (iii) may emerge even in the absence of any variation in underlying individual quality or ability. In populations of humans and nonhuman animals, the additional amplifying impacts of competition and contingency exist against the backdrop of unequal starting position and likely magnify early inequalities that result from structural or environmental adversity or advantage (71–74).

The sources of inequality in human society are of central interest to both moral philosophy and public policy (75–77). As with reproductive success in nonhuman animals, human outcomes are likely to be partially explained by differences in genotypes (78). But we show here that even among isogenic animals, individuals still attain drastically different phenotypic outcomes. Our results add to sociological and biological literature that underscores the potential importance of unpredictable, uncontrollable experiences in generating differences in outcomes even when differences in underlying quality (or “talent”) are small or nonexistent (12, 15, 16).

Supplementary Material

science.org/doi/10.1126/science.adq0579

Materials and Methods

Supplementary Text

Figs. S1 to S6

Tables S1 to S4

References (80–83)

MDAR Reproducibility Checklist

Data and materials availability:

All data and scripts supporting the analyses in this paper can be found in the Cornell eCommons repository (79).