Aggressiveness , reliable signaling and survival in a wild songbird 3

22 The evolution and maintenance of honest or reliable signaling has been a major question 23 in evolutionary biology. The question is especially puzzling for a particular class of signals used 24 in aggressive interactions: threat signals. Here we report a study on song sparrows (Melospiza 25 melodia) in which we assayed males with playbacks in their territories to quantify their 26 aggressiveness and aggressive signaling levels and asked whether these affect their survival on 27 territory. We found that the effect on survival of residual signaling (signaling above or below the 28 level predicted by their aggressiveness) depended on aggression levels such that among males 29 with low aggression, those with higher residual signaling scores had higher survival. The 30 residual signaling did not have a strong effect among high aggression males. Aggressiveness by 31 itself did not have an effect on survival. These results present a first step in understanding the 32 fitness consequences of honest signaling in aggressive contexts. 33


Introduction
In signaling interactions between two unrelated individuals, the interests of signalers and receivers rarely coincide perfectly, which poses the question of how animal signals can evolve to be reliable (Maynard Smith and Harper 2003;Searcy and Nowicki 2005).This question is particularly prominent in the case of threat signals used in aggressive situations, when the interests of signalers and receivers are opposed to each other.Some threat signals reveal traits like physical size or strength of the signaler in a way that cannot be cheated (e.g.Clutton-Brock et al. 1979) but many threat signals are not intrinsically tied to an easily observable and costly trait.These signals, termed "conventional signals" because of the arbitrary relationship between the signal structure and signal message (Guilford and Dawkins 1995), are seemingly easy to cheat and therefore their reliability has been an empirical and theoretical puzzle (Maynard Smith and Harper 2003).Although initially thought to be unreliable and non-predictive in contests over all but trivial resources (Maynard Smith 1974;Dawkins and Krebs 1978;Caryl 1979;Maynard Smith 1979;Maynard Smith et al. 1988), many conventional signals have been shown to be at least partially reliable in predicting further escalation in recent research (Waas 1991;Searcy et al. 2006;Laidre 2009;Akçay et al. 2013).
Although conventional signals can be statistically reliable in predicting an escalation, the reliability often is imperfect (Searcy et al. 2013).Furthermore, signaling strategies of at least some species appear to be individually consistent over time, i.e., some individuals signaling consistently at high levels and others consistently at low levels, even though conventional signals are generally assumed to be flexible behaviors.For example we recently showed that male song sparrows (Melospiza melodia) show individually consistent signaling strategies in repeated aggression assays even after accounting for their aggression levels (Akçay et al. 2014a

PrePrints
results suggest that at least part of the variation present in imperfectly reliable signals can be explained by consistent individual differences (Botero et al. 2010).
The last decade has seen a proliferation of studies focused on consistent individual differences, sometimes termed animal personality (Gosling 2001;Sih et al. 2004;Bell et al. 2009;Dingemanse and Wolf 2010).These consistent individual differences often affect fitness (Dingemanse et al. 2004;Dingemanse and Réale 2005;Smith and Blumstein 2008;Seyfarth et al. 2012).Aggressiveness in particular has been subject of a number of studies (Smith and Blumstein 2008;Bell et al. 2009).The costs and benefits of aggressive signaling however has only been studied in the short-term by looking at immediate receiver responses (e.g.Templeton et al. 2012;Anderson et al. 2013).To the best of our knowledge no prior study has examined the long-term fitness consequences of individual differences in variation in aggressive signaling together with aggressiveness in a wild animal.
In this paper we report aggression and aggressive signaling measures from a population of male song sparrows that were assayed multiple times over a period of several years.Our previous studies found that both aggressive behaviors and aggressive signaling were individually repeatable over this time period and that once aggressive behavior levels were controlled for, residual signaling was also repeatable, suggesting individually consistent signaling strategies (Akçay et al. 2014a).Here we ask whether these individually consistent differences in aggressiveness and aggressive signaling strategies are correlated with survival in the wild, an important component of fitness.

Methods
Study site and subjects: We studied male song sparrows breeding in Discovery Park, Seattle, WA.The song sparrow population has been subject of a long term field study since 1986 (Stoddard et al. 1988;Beecher et al. 1994).Each male was banded with a US Fish and Wildlife Service aluminum band and a unique combination of 3 color bands for individual identification in the field.Subjects were 69 males that held territories in June 2009 which was a little over half the breeding males in our study area (n=123, Akçay et al. 2014b).
Aggression and aggressive signaling assays: We carried out repeated playbacks to simulate territorial intrusions and assay aggressive behaviors and aggressive signaling.The details of the protocol were reported by Akçay and colleagues (2014a).Briefly, we placed a speaker (Pignose Inc.) at the territory center connected to an iPod (Apple Inc.) with a 20 m audio cable.We played each subject two of his own songs (self song) that had been recorded earlier using a Marantz PMD 660 digital recorder and a Sennheiser ME66/K6 directional microphone.Each song was played for 5 minutes at a rate of 1 song every 15 seconds for a total 10-minute trial The amplitude song amplitude was approximately 80 dB SPL measured at 1 m (Radio Shack 33-2055 sound meter).In previous research, song sparrows have shown no behavioral differences in response to stranger song and self song (Searcy et al. 1982;Stoddard et al. 1992).
The playbacks were carried out in September and October 2009 and January, February and May 2010, and we attempted to test each male once in each of these months for a total of During each trial two investigators stood at about 20 m from the speaker and observed the subject.One of the observers also recorded the trial using the same recording gear as above.
We noted verbally the following behavioral measures by narrating the behaviors as they occurred: flights, distance from the speaker at each flight, loud songs, soft songs, wing waves.
Although the amplitude variation in song sparrow songs are continuous, an experienced observer can reliably classify loud songs and soft songs in the field (Anderson et al. 2008).The trial recordings were later viewed and annotated using Syrinx (John Burt, Seattle, WA; www.syrinxpc.com).
Behavioral measures: We extracted the following behavioral measures as our measure of aggression from each trial: rate of flights (per minute), proportion of time spent within 5m of the speaker, and closest approach to the speaker during the trial.These three measures were highly correlated with each other and therefore entered into a principal component analysis (PCA).The first component of the PCA explained 67.9% of the variance and was taken as the aggression score.Akçay and colleagues (2014a) found that the average aggression scores from these trials in 2009-2010 predicted whether the birds would attack a taxidermic mount in Spring 2011.
We use the definition of a signal proposed by Otte (1974): "behavioral, physiological, or morphological characteristics fashioned or maintained by natural selection because they convey information to other organisms" (p.738).Although flying close to a receiver may convey information to him, under this definition it doesn't qualify as an aggressive signal, since flying towards the receiver has evolved for physically dealing with an intruder rather than because of its  (Akçay et al. 2013;Searcy et al. 2014).
We extracted the number of soft songs and wing waves and converted these to rates per minute to account for unequal durations of observation due to different latencies of first response.Rates of soft songs and wing waves were correlated with each other and so we entered them into a PCA.The first component of this PCA explained 68.3% of variance and was taken as the aggressive signaling scores (reported in Akçay et al. 2014a).Note that although we also counted loud songs, loud songs in song sparrows do not seem to be a signal of aggression and do not predict attack (Searcy and Beecher 2009;Akçay et al. 2013;Searcy et al. 2014).Therefore loud songs were not included as an aggressive signal.The same goes for type matching (replying to the playback with the same song type), which has been shown to predict attack in a sequence of escalation in this population (Akçay et al. 2013) but not when playbacks are only done at the territory center (Akçay et al. 2011;Searcy et al. 2014).Thus we focus only on soft songs and wing waves which are the most reliable close-range threat signals in song sparrows.
Survival: We censused the study area as least once every two weeks in between January 2010 and February 2015 except in November and December of each year to determine whether subjects were still alive on territory.Each territory was searched extensively, using playback as necessary.In cases, when a new male was detected, we continued to check the territory as well as the neighboring territories for the former owner.We considered a subject to have disappeared permanently if the subject was not seen on his territory or any of the nearby territories and the area was being actively defended by a new male.Two of the subjects held territories that were adjacent to an army base where we could not engage in extensive searching and so were PrePrints excluded from the analyses, as we could not be sure whether they had disappeared or simply had moved (short moves are not uncommon).In the analyses below we consider years survived past 2009 as our response variable.Of our 67 subjects, 7 did not survive past 2009 (were gone in January 2010), 34 survived only to 2010, 10 survived to 2011, 4 survived to 2012, 7 survived to 2013 and 5 survived to 2014.All of the original subjects had disappeared by January 2015.

Data Analysis:
The repeatabilities for aggression scores (r= 0.48, 95% CI: 0.35, 0.62; p< 0.0001, n=219 trials, 69 subjects), aggressive signaling scores (r=0.57,95% CI: 0.45, 0.70, p< 0.0001, n=219 trials, 69 subjects) and residual signaling scores after controlling for aggression (r=0.36,95% CI: 0.214, 0.51, p < 0.00001, n= 219 trials) were all highly significant, as reported in detail elsewhere (Akçay et al. 2014a).Here we focus on average aggression scores and average residual signaling scores across all trials for a given subject.Residual signaling scores were obtained from a linear regression as described in Akçay et al. (2014a).We took the average of the residual signaling scores for each individual for the analyses reported below.
Our main analysis was a Cox proportional-hazards regression model (Cox 1972) where the response variable was number of years survived past 2009.The Cox regression estimates the change in hazard rate (i.e. the likelihood of death) that comes about with a change in the covariate.The predictor variables were aggression scores, residual signaling scores, the quadratic terms of each and an interaction term.The linear and quadratic terms for aggression scores estimate the strength of directional and stabilizing selection on survival based on aggression, respectively.Similarly, linear and quadratic residual signaling term estimates the strength of directional selection and stabilizing selection on residual signaling scores, respectively.Finally, the interaction term estimates the correlative selection between residual and aggression scores.

PrePrints
We report the hazard ratios, their 95% confidence intervals and the p-values from Wald tests for each predictor variable.Hazard ratios less than 1 indicate decreasing hazard rate (increasing survival) with increasing values of the predictor.Conversely, hazard ratios higher than 1 indicate increasing hazard rate (decreasing survival) with increasing values of the predictor.We report the full models in Table 1, but model averaging based in AICc scores gives very similar results (see the R code in the supplementary materials) For 33 of the 67 subjects for whom we have survival information, we also have information on the exact age of the male because they were banded in the nest, in their juvenile plumage (before their first September) or singing plastic song during their first fall.In previous analyses, reported in Akçay et al. (2014a), we did not find an effect of age on aggression scores or aggressive signaling scores.Nevertheless, we repeated the above analyses with the subset of subjects for whom we had age data by adding the age as a covariate in the Cox regression.All analyses were carried out in R (R Core Team 2012).The Cox regressions were carried out with the package 'survival' (Therneau and Lumley 2014).

Results
In the Cox regression with the entire dataset, there was a significant effect of residual signaling scores on survival (Table 1), indicating that individuals with higher residual signaling scores were more likely to survive longer.However, this effect was modified by an interaction effect of residual signaling with aggression scores: males with higher residual signaling scores generally survived longer among the less aggressive males, whereas residual signaling did not have a strong effect among the more aggressive males (Figure 1b).Aggressiveness did not have any effect on survival rates.The same pattern of results held in the subset of subjects with age PrePrints data.Additionally, in this model, the age of the subject had a significant effect on survival: not surprisingly, older individuals were less likely to survive longer.Finally, all of the results held true when the one subject with the highest residual signaling score was treated as an outlier and removed from the analyses (see supplementary R code).

Discussion
In this study we examined the effects of aggression and aggressive signaling on an important component of fitness, survival on territory.We found that aggressiveness was not correlated with survival, but residual signaling (the component of signaling not correlated with aggression) was, such that males with higher residual signaling scores were favored, specifically among the low aggression individuals.To our knowledge, this is the first demonstration of a correlation between aggressive signaling and fitness independent of aggression.
The finding that aggressiveness was not correlated with survival is consistent with other studies that have shown aggression to have a small and non-significant effect on survival in wild animals (Smith and Blumstein 2008).Note that although aggressiveness may not predict survival once on territory, there is evidence in some species that high aggressiveness may lead to obtaining higher quality territories or nesting sites in the first place (Duckworth 2006;Rosvall 2008;Scales et al. 2013).It is also possible that aggressiveness may confer benefits in terms of social and extra-pair mating.For instance in nightingales (Luscinia megarhynchos), aggressive males have higher subsequent pairing success (Schmidt et al. 2008).Similarly, although there is less evidence on this point, aggressive males may be better able to pursue and obtain extra-pair fertilizations, particularly in a system like song sparrows where extra-pair matings are driven

PrePrints
primarily by the male strategies (Akçay et al. 2012).These possibilities are worth pursuing in future studies.
The main and unexpected result of our study is that there was an effect of residual signaling on survival, but the direction of the effect depended on the aggressiveness of the individual, as indicated by the strong interaction effect: Among males with low aggression scores those with higher residual signaling had higher survival, but the effect disappeared among high aggression individuals (Figure 1b).One way to interpret these results is that aggressive signaling benefits only those individuals who don't at the same time display other aggressive (nonsignaling) behaviors like flights and close approach.Although these behaviors are not considered signals (sensu Otte 1974), they nevertheless give information to receivers, and can be used to predict attack (Searcy et al. 2006;Akçay et al. 2013).The presence of high rates of aggressive behaviors may thus explain why residual signaling does not have significant benefits or costs in terms of survival in high aggression males.
If only low aggression males benefit by high residual signaling, this will flatten the regression line between aggressive signaling and aggression scores (Figure 1a).and weaken the reliability of this signaling system.We think it is unlikely however, that it would cause this signaling system to break down, for several reasons.First, we note that the variation in residual signaling among low aggression males (where the effect of residual signaling is strongest) is small compared to variation in signaling across high and low aggression individuals.Thus, although selection on residual signaling can make the regression line flatter, the effect of it will be limited by the magnitude of the variation in residual signaling.

PrePrints
Second, the individuals with low aggression scores but high signaling scores are not necessarily deceptive signalers.Instead these individuals may simply be using an alternative strategy for dealing with an intruder, engaging in relatively more aggressive signaling (wing waving and soft singing) and relatively less aggressive behavior (getting close to the intruder and flying around it).These individuals may nevertheless attack an intruder, making good on their threat signals.Indeed, when we tested 3 of the 7 individuals (the rest were not tested) in the topleft quadrant of Figure 1a in Spring 2011 with a taxidermic mount, all 3 of them attacked the mount (Akçay et al. 2013).The arguments presented above are consistent with the fact that the song sparrow signaling system has been found to be reliable in two different subspecies of song sparrows (Searcy et al. 2014; see also Figure 1a).Furthermore, reliable aggressive signaling via soft songs and wing waves also occurs in the congeneric swamp sparrows (Ballentine et al. 2008), suggesting the reliability of these signals likely has been conserved since their common ancestor.
In summary our study indicates that there may be opposing selective forces on reliable signaling just as theory to date has suggested (Maynard Smith and Harper 2003).Given the correlational nature of the current dataset, and the current lack of information on heritability, it is five trials.The achieved sample sizes varied due to disappearance of subjects, either temporarily or permanently.Eleven subjects were tested once, 13 subjects were tested twice, 12 subjects 3 PeerJ PrePrints | https://dx.doi.org/10.7287/peerj.preprints.1195v1| CC-BY 4.0 Open Access | rec: 23 Jun 2015, publ: 23 Jun 2015 PrePrints times, 19 subjects 4 times, and 14 subjects all 5 times (mean+SD trials per subject: 3.17+1.38,total number of trials: 219).

PrePrints 20 Figure 2 .
Figure 2. Residual signaling scores and years survived past 2009.The subjects are