Dynamically honest displays: courtship locomotor performance indicates survival in guppies


  • Shyril O’Steen,

    Corresponding author
    1. Biology Department, Seattle University, 901 12th Ave, Seattle, Washington 98122, USA;
      Correspondence author. E-mail: osteens@seattleu.edu
    Search for more papers by this author
  • Stephanie L. Eby,

    1. Biology Department, Bates College, 44 Campus Ave, Lewiston, Maine 04240, USA
    Search for more papers by this author
    • Present address: Department of Biology, Syracuse University, 107 College Place, Life Sciences Complex, Syracuse, New York 13244, USA.

  • John A. Bunce

    1. Biology Department, Bates College, 44 Campus Ave, Lewiston, Maine 04240, USA
    Search for more papers by this author
    • Present address: Department of Anthropology, University of California at Davis, 1 Shields Avenue, Davis, California 95616, USA.

Correspondence author. E-mail: osteens@seattleu.edu


1. The indicator theory of sexual selection suggests that mating displays honestly signal aspects of fitness. While rarely studied, kinematic (locomotor) performance is an excellent candidate for an honest indicator, as mating displays of many animals include rapid or extended locomotion that may be physiologically correlated with performance traits that impact survival.

2. We investigate the indicator value of display locomotion of wild-caught male guppies, Poecilia reticulata, by examining relationships between mating display kinematic traits, anti-predator kinematic traits, and survival during a subsequent staged encounter with a natural predator, the pike cichlid Crenicichla alta.

3. We first compared guppy display kinematics with subsequent survival, and found that display body angle and angular speed positively predicted survival. We next compared anti-predator kinematic and tactical traits with survival, to identify traits that might link mating displays to survival. We measured anti-predator traits in two tests, first in response to a standardized stimulus (fast start test), and second in response to the live predator (encounter test). Guppy fast start speed and encounter speed, time in refuges, and approach distance (response distance) all positively predicted survival, while encounter swim duration negatively predicted survival. These data provided our final hypothesis, that these particular anti-predator traits would be correlated with mating display kinematics. However, we detected only one of eight predicted correlations, a negative relationship between display body angle and encounter swim duration that may reflect an energy trade-off.

4. We conclude that courtship locomotor performance can be an honest survival indicator in guppies, and that the mechanism linking courtship to survival merits further study. These results suggest that courtship locomotion may contribute to viability impacts on the evolution of animal mate choice, and support others in suggesting that these traits may reward greater attention in sexual and natural selection studies.


Sexually selected traits increase fitness by increasing mating success, but they may also affect viability (survival), with total fitness resulting from the combination of mating and viability effects. Viability consequences of sexual selection hold broad importance in that they can influence the rate of adaptive evolution and the probability of extinction (Proulx 1999; Whitlock 2000; Kokko & Brooks 2003a). We currently lack the volume of research needed to determine the frequency of viability impacts in sexual selection systems (Andersson 1994; Moller & Alatalo 1999; Jennions, Moller & Petrie 2001; Kokko 2001; Kokko et al. 2003b). Our study aims to add to the tools for assessing viability selection by examining kinematic elements of mating displays, as these traits are common but rarely analysed (Irschick et al. 2007, 2008b). We propose that display kinematics may reliably predict viability, and that this relationship may result from functional links between displays and kinematic skills that directly influence survival.

Mating display traits that reliably predict viability are considered honest indicator traits, and as such should be either costly or constrained (Zahavi 1975, 1977; Grafen 1990; Cotton, Fowler & Pomiankowski 2004; Tomkins et al. 2004; Vanhooydonck et al. 2007). Display costs include metabolic demands, for example, acoustic mating calls often require large fractions of daily energy budgets (Halliday 1987; Ryan 1988; Prestwich 1994; Sullivan & Kwiatkowski 2007). Constraints include developmental and mechanical factors. For example, colour pattern, body shape and size are developmentally linked in Danio fish (McClure & McCune 2003), and thus displays featuring one of these traits could indicate the linked traits and underlying regulatory genes. Similarly, jaw muscles of male collared lizards (Crotaphytus collaris), displayed during male–male contests, appear mechanically constrained to indicate bite force, a sexually selected performance trait (Lappin et al. 2006).

Kinematic elements of mating displays should be prime candidates for honest viability indicators due to physiological and biomechanical constraints (Irschick et al. 2007, 2008b). A key constraint could result from functional links between display kinematics and anti-predator skills, such as maximum speed or turning ability. Our study examines the prediction that display kinematics will reliably indicate ability to escape from predators. We use Trinidadian guppies (P. reticulata), a model system for evolutionary studies of mate choice and anti-predator behaviour. Guppies have a promiscuous mating system with opportunity for high variance in male mating success and thus sexual selection (Houde 1997). Females show distinct preferences for male chromatic and morphological traits, contributing to the rapid evolution of these traits in natural and experimental environments (Endler 1980, 1983; Houde & Endler 1990; Houde 1997). The fitness benefits of these preferences apparently accrue mainly from producing sexy sons, as male colour and morphology are subject to negative viability selection; preferred males show reduced survival (Godin & Dugatkin 1996) and reduced offspring viability (Brooks 2000).

However, male guppy displays include unassayed kinematic elements that may provide different information on viability. Males advertise using sigmoid displays, in which the male bends laterally into an S or C shape and holds the shape while moving slowly through an arc in front of the female. The display bears an interesting resemblance to a freeze-frame of a standard escape response in teleost fish. A startled fish contracts all the muscles on one side of the body, bending laterally into a C shape, then accelerates with a propulsive stroke of the tail (Webb 1994; Dommenici & Blake Robert 1997; Hale et al. 2002). Guppy mating displays are stiff and slow relative to the escape response, but the common C shape suggests the behaviours could be phenotypically linked and indicative of skill escaping from predators (Walker et al. 2005). Such skills are highly relevant to guppies as they fall prey to multiple piscivores in nature, which influences many aspects of their evolution (Endler 1980; Reznick, Bryga & Endler 1990; Houde 1997; O’Steen, Cullum & Bennett 2002; Magurran 2005).

We examined three aspects of the phenotypic relationship between guppy mating display kinematics and viability. First, we tested the hypothesis that male display kinematics would reliably indicate male survival during a later staged encounter with a natural predator, the pike cichlid Crenicichla alta. Second, we tested the hypothesis that male anti-predator kinematics would influence survival, a relationship widely assumed but little tested (Walker et al. 2005). Third, we examined correlations between male mating display and anti-predator kinematics, to test the hypothesis that the latter would provide a functional link between displays and survival.

Materials and methods


The focal animals of this study were wild caught guppies, Poecilia reticulata, captured as adults in the Aripo River, Trinidad. We used fish from three populations with the goal of obtaining a range of male display and anti-predator behaviours. Two of the populations, downstream and midstream Aripo, are considered high-predation (high mortality) sites and contain the primary guppy predator, the pike cichlid Crenicichla alta (Liley & Seghers 1975; Endler 1978; Reznick, Butler & Rodd 2001; O’Steen, Cullum & Bennett 2002). The third site, upstream Aripo, is a low-predation site where cichlids do not occur (ibid). Further descriptions of capture sites can be found in O’Steen, Cullum & Bennett (2002). The fish were housed in our laboratory in 25 × 50 × 30 cm high glass aquaria at 24–26 °C and exposed to a 12 : 12 light : dark cycle. Guppies from different populations were maintained in separate aquaria at similar densities and demographic levels, in a total of 4–5 aquaria per population, and were fed commercial tropical fish food 1–3 times daily. We also maintained an outbred population of domestic P. reticulata that was morphologically similar to wild guppies and was used to provide stimulus females in the mating display experiments. At the time of the wild guppy collection, we captured four adult pike cichlids from the downstream Aripo River. Cichlids were maintained as above, except that they were housed individually and fed live invertebrates or guppies.

Overview of experimental design

We used high-speed video to record kinematic and tactical behaviours of adult male guppies during three separate events: (1) mating displays, (2) standardized fast starts, and (3) encounters with a pike cichlid. Individual males were recorded during each event in sequence over the course of several hours. We consider one such record of these three events to constitute one trial. One male from each of the three populations was measured every third trial in randomized order. The predator encounter phase of the experiment was conducted in a random block design, with each of the four predators used in 9–12 trials, and each predator encountering each guppy population in 3–4 trials in random order. The final dataset contained 42 trials conducted over a 4-month period. Sample sizes for the three trial events individually ranged from 34 to 38 due to the imperfect cooperation of subjects and equipment.

Trial event 1: mating displays

On the evening before a trial we selected a mature male that we endeavoured to size match within and among populations [standard length = 1·85 ± 0·14 cm (mean ± 1 SD), n = 42]. The male was anaesthetized in water containing 230 mg L−1 tricaine methanesulfonate (MS 222) and 115 mg L−1 sodium bicarbonate and photographed to document length. All males recovered immediately on being returned to clean water. The male was left undisturbed overnight in a display arena measuring 20·5 cm square and filled to 12 cm with water kept at 25 °C. The next morning the trial was initiated by placing a stimulus female in the display arena. Stimulus females were size matched (2·56 ± 0·38 cm, n = 42) non-virgin domestic guppies, chosen as they were equally unfamiliar to each population of Trinidadian males and based on daily observations were morphologically and behaviourally similar to Trinidadian females.

For each male, we recorded two sigmoid displays in top view using a Redlake PCI 500S high-speed video system recording at 250 frames/s (fps). Displays were only recorded when the two guppies were within 6 cm apart and the female was within 90° of directly facing and not moving away from the male. We defined one display as beginning when the male began to bend into the sigmoid position and ending when he returned to a straight-bodied position and paused or swam normally for 0·1 s. Displays lasted on average 3 s with a range of 1–6 s.

We quantified kinematic display variables using a customized digitizing program (Didge 2.3 created by A. Cullum, http://biology.creighton.edu/faculty/cullum/Didge/index.html) to define the positions of the male and female. We defined the male with six equidistant points placed along his midline from snout to base of the tail, and defined the female with one point on the snout and one at mid-body. We digitized every tenth frame (25 fps) for the entire display, and digitized every frame (250 fps) for 20 consecutive frames at the midpoint of the display. We used a customized Microsoft Excel spreadsheet (available on request from S. O’Steen) to calculated variables from the digitized data. Male variables are described in Table 1. Female variables were used to test for possible influences on male display (Houde 1997). Pearson’s and partial correlation matrices showed no significant correlations between male variables and female size or behaviour (Table S2 in Supporting Information), and we excluded female data from further analysis. We used both of the male displays in further analyses to examine the consistency of display variables and of discriminant functions analyses. For some males we had data for one display only, as during digitizing we rejected incompletely recorded displays and those too far from the female. Thus we composed a larger dataset, display one, containing one display from every male, and a smaller dataset, display two, containing the second display from males with two complete displays.

Table 1.   Descriptions of male guppy variables
Trial event 1: Mating displays
Duration: Total duration of one sigmoid display
Body angle: Maximum body angle of displaying male, as averaged over any given frame interval, below. Angle calculated as the sum of deflections from 180° of the four body angles described by six digitized points placed along the body midline in top view
Angular speed: Maximum averaged rate of change in the angle formed between the tip of the snout and the third digitized point, the latter being the usual the centre of mass during fast starts
Linear speed: Maximum averaged velocity of the third digitized point
 The latter three variables were measured using: Frame rate = 25 fps; Averaging = 10 intervals (400 ms); Segment analysed = entire display
 The latter three variables were also measured at 250 fps. The resulting data failed to predict survival and are presented in Table S1 in Supporting Information
Trial event 2: Fast starts
Angular speed: Maximum averaged rate of change in the angle formed between the tip of the snout and the centre of mass, the latter identified as the digitized point at the centre of bending during the fast start
Linear speed: Maximum averaged velocity of the centre of mass
Net distance: Net distance travelled by the centre of mass between the outset of the fast start and the end of the time segment
 All variables were measured using: Frame rate = 250 fps; Averaging = two intervals (8 ms). Two different time segments were used:
 20 ms: Maximum speeds and distance recorded from 0 to 20 ms
 100 ms: Maximum speeds and distance recorded from 0 to 100 ms
Trial event 3: Predator–prey encounters
Refuge time: Total time that guppy used refuges during the 6 min encounter
Approach distance: Distance between the guppy centre of mass and the snout of the cichlid at the outset of the guppy’s first fast-start of the encounter
Linear speed: Calculated as for event 2 above, using the first fast-start of the encounter
Net distance: Calculated as for event 2, using the first fast-start of the encounter
Swim time: Total time that guppy swam without pausing from the outset of the first fast-start through 400 ms (guppies captured in under 400 ms were dropped from the experiment)
Survival time: Time that the guppy survived during the 6 min encounter

Trial event 2: fast starts

Fast starts were recorded in the mating display arenas. Following the second display, the female was removed and the focal male left undisturbed for 20 min. A fast start response was then elicited with a standard stimulus provided by dropping small weight against the outside of the tank from a fixed pendulum. The males responded with a classical C-start, first bending the body into a C shape around the centre of mass and then accelerating with a propulsive stroke of the tail (Webb 1994). We recorded two fast starts with the high-speed video, leaving the male undisturbed for 10 min between starts. We digitized 25 consecutive video frames (100 ms at 250 fps) for each fast start, beginning with the frame before the first C-start movement, and using the same digitizing points as for displays. We retained the data from the fastest overall of the two starts for further analysis. Variables are described in Table 1.

Trial event 3: predator–prey encounters

Encounters between male guppies and cichlids were recorded in arenas designed to provide some natural behavioural options by allowing freedom of movement, refuges for the guppies, and acclimation time. The arenas measured 60 × 45 cm, were filled to a depth of 12 cm and maintained at 25 °C. They contained two refuges: guppies successfully hid under an artificial rock (12 × 6 × 4 cm high) and behind and above the cichlid’s holding box (21 × 13 × 12 cm, below). The cichlid was resident in the arena for a minimum of 10 days prior to an experiment, and during this time was acclimated to trial conditions by being fed one domestic guppy every other day using the trial protocol.

Encounters were initiated after the second fast start. To prepare, the male guppy was left undisturbed in the display/fast start arena for 20 min, during which the cichlid was moved to one corner of the encounter arena and trapped beneath the clear holding box. The guppy was then placed in the far corner of the encounter arena in a clear box, left undisturbed 20 min, then released and left undisturbed an additional 20 min. During the latter time the guppy typically swam several circuits around the arena including the area around the cichlid. The cichlid box was then raised via remote pulley and replaced after the cichlid swam out. When replaced the top of the box was 0·5 cm underwater and two sides of the box were 0·5–1 cm from the arena corner walls, providing guppy refuges. The encounter was considered to start when the cichlid left the holding box and was stopped 6 min later. At the start the guppy typically stopped swimming and faced the predator. If the cichlid approached, the guppy would perform a fast start, and this approach-escape sequence repeated until either the guppy was captured or 6 min elapsed.

We recorded encounters with two different cameras mounted directly above the arena. The high-speed camera system was capable of recording for 8 s at 250 fps with a refractory period of 60 s. As some guppies were captured within the first 60 s, the camera refractory period meant that we could record a single 8 s sequence for every guppy. Thus we used the high-speed system to record the first guppy fast start performed when the cichlid was within 20 cm and moving toward the guppy. To document the entire 6 min encounter, we used a Sony TRV230 digital video camera recording at 30 fps. We used the high-speed video to digitize 25 consecutive frames beginning with the frame before the first fast start movement. The large field of view required for the encounters produced lower resolution images than those in the display and fast start events, and we digitized guppy position using two consistently distinct points, the snout and the rear of the abdomen, and estimated centre of mass as the midpoint, which precluded measuring guppy angular speed (Table 1). Cichlid position was defined using one point on the snout. Variables are described in Table 1.

Statistical analyses

We used discriminant functions analyses to identify variable combinations within each trial event that predicted survival time during the predator encounter. Discriminant analyses use a nominal dependent variable, here guppy survival vs. non-survival at given time points. We had no a priori reason to use particular times and constructed models using a series of survival points. Models using similar times produced similar results, and we present six representative models using survival points of 1, 4, 10, 20, 100, and 360 s. To construct the models, we first examined all independent variables for normality and conducted transformations as required. Second, we examined possible influences of cichlid identity on the dependent variable (survival) by constructing a nominal logistic regression for each survival time point. A marginal influence was detected at 1 s (d.f. = 3, r2 = 0·19, chi square = 7·4, P = 0·06), but none at other time points (0 < r< 0·12, 0·1 < chi < 5·1, 0·99 > P > 0.14). Third, for predator encounter variables, we used one-way analysis of variance (anova) to examine possible influences of cichlid identity on guppy variables. Cichlids affected guppy linear speed−20 ms and swim time (F = 5·8 and 5·3, P = 0·002 and 0·004, respectively, d.f. = 3, 35), and we used anova residuals to remove cichlid influence from these variables. Cichlid identity did not significantly influence other encounter variables (0·2 < F < 2·4, 0·85 > P > 0·08, d.f. = 3, 35). Fourth, we conducted principal components (PC) analyses of locomotor variables within each trial event (Table 2), in order to retain principal variables, without bias, while eliminating correlations that could disrupt discriminant analyses. In the predator encounter PC analysis, the more tactical variables of approach distance and refuge time were excluded, as they were not significantly correlated with other encounter variables (Table S3) and this enabled us to limit PCs to locomotor traits. Last, discriminant models were constructed using linear analysis and stepwise variable selection in spss 17·0 (SPSS Inc., Chicago, IL, USA).

Table 2.   Principal components (PC) analyses of locomotor variables. PCs were constructed to identify major components of variance within datasets, reduce the number of variables without bias, and remove correlations among variables that might influence further analyses. PCs were included here and in further analyses where PC eigenvalue > 0·5 and percent > 15.
Trial event 1: Mating display one    
 PC 1PC 2PC 3PC 4
Eigenvalue 1·48 1·02 0·86 0·62
Rotated Factor Pattern    
  Duration 0·00 0·99 0·08 0·09
  Body Angle 0·17 0·09 0·05 0·98
  Angular Speed 0·99 0·00 0·05 0·17
  Linear Speed 0·05 0·08 0·99 0·05
Trial event 1: Mating display two
 PC 1PC 2PC 3 
Eigenvalue 1·53 1·34 0·64 
Rotated Factor Pattern    
  Duration 0·01 0·96 0·17 
  Body Angle 0·81−0·27−0·21 
  Angular Speed 0·86 0·24 0·12 
  Linear Speed−0·04 0·17 0·98 
Trial event 2: Fast start
 PC 1PC 2PC 3 
Eigenvalue 2·84 2·00 0·90 
Rotated Factor Pattern    
  Angular Speed 20ms−0·08 0·99 0·04 
  Angular Speed 100ms−0·09 0·99 0·08 
  Linear Speed 20ms 0·98−0·02 0·03 
  Linear Speed 100ms 0·87 0·02 0·40 
  Net Distance 20ms 0·93−0·24−0·03 
  Net Distance 100ms 0·12 0·09 0·98 
Trial event 3: Predator-prey encounter
 Guppy dataCichlid data
 PC 1PC 2PC 3PC 1
Eigenvalue 1·98 1·52 0·90 2·62
Rotated Factor Pattern    
  Linear Speed 20ms 0·95−0·05 0·01 0·38
  Linear Speed 100ms−0·07 0·85 0·30 0·93
  Net Distance 20ms 0·90 0·22−0·17
  Net Distance 100ms 0·22 0·88−0·16 0·75
  Guppy Swim Time−0·09 0·07 0·97

Possible differences between guppy populations in male standard length, trail locomotor PCs and trial tactical variables were examined with anova, and population differences in survival were examined with logistic regression. Only one variable, predator–prey encounter PC 1, differed among populations (Table S4), and this difference disappeared when populations were recoded by predation regime (F = 0·1, P = 0·71, d.f. = 1, 38). These results may in part reflect relatively low, per population sample size (n = 14), as measuring population differences was not a study goal. Hypothesized correlations between mating display, fast start and predator encounter PCs were examined with Pearson’s and partial correlations. anova, logistic regressions and correlations were constructed using jmp 7 (SAS Institute, Cary, NC, USA).


Mating display discriminant functions

Kinematic aspects of male mating displays successfully predicted male survival during subsequent encounters with predators. In particular, display body angle and angular speed were reliable and positive indicators of survival, while display duration and linear speed were not. Discriminant models of display one showed that PC 1 (angular speed) and PC 4 (body angle) positively predicted survival at several time points (Table 3 pt. 1). Display two confirmed this pattern, as the same traits positively predicted survival in all significant models (Table 3 pt. 1). In contrast, PC 3 (linear speed in both displays) was inconsistent across displays, being a strong negative predictor of survival in all models of display one, but contributing to no models of display two (Table 3 pt. 1). Display duration contributed to no models for either display. The concordance, and lack thereof, among the survival models is consistent with the repeatability of kinematic performance across displays. The PCs comprising body angle and angular speed were significantly repeatable (r > 0·58, P < 0·004), while those for linear speed were not (r = 0·02, P = 0·86, Table S5). Thus the consistency of display angle and angular speed may contribute to their reliability as survival indicators, while the reverse appears true for display linear speed.

Table 3.   Discriminant functions models of the ability of trial event variables to predict guppy survival. Standardized function coefficients are shown for variables retained in models; correlations between variables and discriminant functions are shown for all variables. Thumbnail image of

Standardized fast start and predator–prey encounter discriminant functions

Guppy kinematic performance during standardized fast starts was a relatively poor predictor of later survival, in that these data produced just one discriminant model. However, this model showed strong positive correlations between the first two PCs (primarily linear and angular speeds) and survival (Table 3 pt. 2). Guppy kinematic and tactical performance during predator-prey encounters was a better predictor of survival, in producing significant models at all survival time points (Table 3 pt. 3). Guppy tactical performance appeared in the most models. Time spent in refuges strongly positively predicted survival across the final four time points, and approach distance was a positive predictor across the time spectrum (Table 3 pt. 3). Of the guppy kinematic traits, PC 2 (linear speed and distance over 100 ms) positively predicted survival at two time points, a result that corresponds to that for standardized fast starts (Table 3 pts. 2 and 3). Interestingly, at the first two time points, guppy kinematic PC 3 (swim time) negatively predicted survival in combination with cichlid PC 1 (speed and distance) (Table 3 pt. 3). This result suggests that longer guppy swim times may have resulted from faster cichlid attacks. However, these two variables were not correlated across the dataset (Table S3), suggesting that faster cichlids only affected guppy swim times, or that long swim times were only a poor tactic, for guppies under immediate attack.

Correlations between mating display and anti-predator kinematic variables

The discriminant models identified four candidates for the phenotypic link between mating display kinematics and survival: standardized fast start PCs 1 and 2, and predator-prey encounter PCs 2 and 3. Table 4 presents a correlation matrix examining relationships between these four variables and the two mating display PCs that reliably predicted survival, PC 1 (angular speed) and PC 4 (body angle). Based on each variable’s relationship with survival, the phenotypic link hypothesis predicts that the two mating display PCs should be positively correlated with the two fast start PCs and encounter PC 2. However, none of these six correlations is either significant or strong (Table 4). For the final encounter variable, PC 3 (swim time), the link hypothesis predicts a negative correlation with the two mating display PCs. In this case, one of the two correlations is indeed strongly negative: that with display PC 4 (body angle) (Table 4). Thus while both mating display and anti-predator kinematics predicted survival, our analyses detected but one of eight potential links between the two types of kinematic traits.

Table 4.   Correlation matrix of guppy locomotor variables that significantly predicted survival in discriminant analyses. Pearson correlation coefficients appear above the diagonal, partial correlations below
 1. Mating display one2. Fast start3. Pred. encounter
PC 1PC 4PC 1PC 2PC 2PC 3
Angular speedBody angleLin. spd., Dist. 20Angular speedLin. spd., Dist. 100Swim time
  1. Values in bold are significant (Bonferroni corrected for n = 8 predicted correlations, P < 0·05; uncorrected P = 0·0004); no other correlations are significant regardless of correction. The matrix depicts PCs from display one due to the larger sample size and omits display PC 3 due to its unreliability as a survival indicator. See Supplementary Table S2 for correlation matrix of display one versus two. See Table S3 for correlation matrix of all variables used in discriminant analyses. N = 34 trials.

1. Mating display onePC 1−0·060·23−0·02−0·04
PC 40·24−0·25−0·15−0·57
2. Fast startPC 1−0·070·220·16−0·17
PC 20·24−0·190·120·15
3. Pred. encounterPC 2−0·04−0·210·190·09
PC 3−0·06−0·55−0·020·03


Kinematic display traits can honestly indicate viability

The results support our first hypothesis, that kinematic elements of guppy mating displays can reliably indicate viability. Body angle and angular speed of displaying males successfully predicted survival during later encounters with pike cichlids. Pike cichlids are key predators that influence diverse aspects of guppy evolution (Endler 1980; Reznick, Bryga & Endler 1990; Houde 1997; Magurran 2005), suggesting that information regarding a prospective mate’s ability to escape these predators would be valuable. In fact such information should be broadly valuable in animals, but we have found no prior studies of the relationship between display kinematics per se and viability. While few in number, studies of display rates have shown viability links. For example, elegant studies of drumming displays by male wolf spiders demonstrate that males that elect to drum at higher rates survive longer in the laboratory and the field (Kotiaho et al. 1996, 1999; Mappes et al. 1996). Together, these results suggest that kinematic elements of displays may be good candidates for honest indicators across diverse taxa.

A logical next step in assessing display kinematic function is to determine their influence on mating success. Studies of display rate suggest females may attend to male kinematics. Females of many species favour males with higher display rates, while females of fewer species show no or occasionally a negative response (reviews in Andersson 1994; Houde 1997; Sullivan & Kwiatkowski 2007). Male display rate can be confounded by female interest, as display rate often increases with female attention, probably contributing to the diversity of female responses (Houde 1997). Display kinematics may suffer the same problem, although video animation studies could circumvent this (e.g. Rowland 1995; Nicoletto & Kodric-Brown 1999), and in live studies kinematics could be studied while controlling for display rate. Questions clearly remain on the function of courtship movement in inter-sexual selection, and kinematic studies may prove a useful tool for addressing them.

Kinematic anti-predator traits can predict viability

The results support our second hypothesis, that kinematic elements of guppy anti-predator performance will affect survival. This result is consistent with reports in other vertebrates. A review by Irschick et al. (2008b) identified 16 studies of the relationship between locomotion and survival, 12 of which reported positive directional selection gradients acting on speed. However, selection on kinematic traits is poorly understood in guppies, despite extensive research on the ecological and evolutionary consequences of their living with predators (Magurran 2005). Guppies do exhibit evolved differences in survival skill. Guppies from high-predation populations are more likely to survive attacks by predators (Seghers 1973; O’Steen, Cullum & Bennett 2002), and female guppies from high-predation populations have greater fast start speeds and swim distances in early gestation, but interestingly not in late gestation, suggesting trade-offs between reproductive and defensive allocations (Ghalambor, Reznick & Walker 2004). It is interesting that our study did not detect population differences in male kinematics, but we feel interpretation awaits further study given our small per population samples (Materials and methods; Table S4).

Our results suggest that pike cichlids can create selection that favours guppies capable of faster linear and angular speeds. Further evidence is provided by Walker et al.’s (2005) study of guppy fast starts during active strikes by pike cichlids, where increased guppy linear and rotational velocity, acceleration, and distance travelled increased the probability of evading strikes. Our results complement these with several new insights. First, our data show that a guppy’s initial response to a cichlid, prior to an active strike, successfully predicts survival. This could suggest that fast initial speeds deter or delay cichlid pursuit by signalling unprofitability (review in Caro 2005), or simply that speed is repeatable. Second, our data indicate that speed is important even when a more powerful anti-predator tactic, refuge use, is available (Table 3 pt. 3). Third, in our encounter arenas, speed measured over 100 ms (PC 2) predicted survival while speed measured over 20 ms (PC 1) did not (Table 3 pt. 3). These data suggest that studying speed over longer intervals will be a meaningful complement to earlier work restricted to 20–30 ms intervals (Ghalambor, Reznick & Walker 2004; Walker et al. 2005). A possible explanation for our result is that if guppy speed acts to signal unprofitability, then longer intervals may better reflect the signal. Fourth, our kinematic data predicted survival using reduced resolution encounter data and simple-averaging variable calculations (Table 1). Higher resolution images and smoothing algorithm calculations can reduce the risk of type II errors (Walker 1998), suggesting our survival results are robust, and in turn that lower resolution methods may remain useful if interpreted with caution.

In addition to speed, we found that guppy tactics aided in escaping cichlids, as time in refuges and approach distances positively predicted survival. Ours is the only kinematic study to provide refuges during predator encounters, and guppies used these in several ways, including sheltering in the shallow water and screening refuges provided by the empty predator box, and leaping out of the water to hang suspended on the side of the arena for several minutes. These tactics are similar to those observed in wild guppies avoiding either natural predators or researchers in Trinidad (personal observations by O’Steen; review in Magurran 2005). Our results document the efficacy of these tactics, consistent with findings in other species (Krause, Hensor & Ruxton 2002; Caro 2005), and provide evidence of the selection assumed to underlie evolved differences in tactical behaviours among guppy populations (Magurran 2005).

Mechanism linking kinematic display traits to viability

Our results provided only limited support for our third hypothesis, that anti-predator kinematic traits would provide phenotypic links between mating display kinematics and survival. In the link that did fit the hypothesis, guppies with greater mating display angles had shorter predator encounter swim times, and both traits predicted longer survival (Table 3). An explanation could be that extreme displays fatigued guppies, resulting in a beneficial shift to shorter-burst strategies during predator encounters, or simply that guppies opted to allocate energy differently in the different situations. This possibility of an energy trade-off is consistent with the negative correlations we also found between display angular speed and encounter approach distance and PC 1 (Table S3).

This idea leaves open the questions of why we did not detect more correlations between display and anti-predator traits, and why the former actually appeared better at predicting survival. One possible reason is that the true correlations between traits were underestimated. Such underestimations are predicted for performance traits exhibiting intra-individual variability (Adolph & Hardin 2007), and could also have resulted from the lower resolution of our encounter variables. Thus we cannot rule out the possibility that our original hypothesis was correct. Our hypothesis assumed that display and anti-predator kinematics would be linked by anaerobic machinery. The anti-predator traits were almost certainly anaerobic due to their high speed and short duration (Webb 1994); guppies completed the fast starts with a mean linear speed of 0·93 m s−1, and we measured anti-predator traits in the initial 20–400 ms. Mating displays lasted considerably longer, although their 1–6 s length also places them in a swimming category generally governed by anaerobic processes (Beamish 1978). Nonetheless, displays may contain aerobic components due to their slow overall rate of movement (mean linear speed 0·047 m s−1). Aerobic processes probably also contributed to guppy survival in our study, as the predator encounters lasted up to 6 min and often involved multiple escape bursts by the guppies. Therefore, a second explanation for our results could be that the missing links between display kinematics and survival are aerobic anti-predator traits, such as the ability to recover between bursts, or endurance. Consistent with this possibility is the fact that display kinematics measured at 250 fps, presumably more likely to reflect anaerobic skills, failed to predict survival (Table S1).


Our results suggest that kinematic aspects of mating displays can honestly indicate viability in guppies. The results raise several questions, including: what mechanism regulates the relationship; do kinematic display elements indicate other aspects of fitness such viability in the wild or mating success; and, do kinematic display elements interact with male colour pattern, a primary subject of guppy sexual selection (Endler 1983; Houde 1997). Since typical mating displays showcase multiple traits, identifying which if any is an honest indicator can be tricky. Mating preferences may evolve to use the single most constrained (honest) trait for mate choice (Pryke, Andersson & Lawes 2001), or to use several traits either independently or in preferred combinations (Brooks & Couldridge 1999; Blows, Brooks & Kraft 2003). Given these scenarios, overlooking a possible indicator trait might result in underestimating the role of viability selection in mate choice. Our results support others that suggest that performance traits used in courtship are useful for addressing such broader questions in sexual selection, and similarly, that inter-sexual selection may contribute to the complex influences on the evolution of locomotor performance (Oufiero & Garland 2007; Husak & Fox 2008; Irschick & Le Galliard 2008a).


We gratefully acknowledge field support from R. Hernandez and the Asa Wright Nature Centre in Trinidad, and thank R. Rutherford, L. Whitlow and four anonymous reviewers for constructive comments on the manuscript. This research was approved by the Animal Care and Use Committee of Bates College, and funded by the Murdock College Research Program for Life Sciences 2006257 (O’Steen), Seattle University and Bates College.