Using the novel FishSim Animation Toolchain, we present a protocol for non-invasive visual manipulation of public information in the context of mate-choice copying in sailfin mollies. FishSim Animation Toolchain provides an easy-to-use framework for the design, animation and presentation of computer-animated fish stimuli for behavioral experiments with live test fish.
Over the last decade, employing computer animations for animal behavior research has increased due to its ability to non-invasively manipulate the appearance and behavior of visual stimuli, compared to manipulating live animals. Here, we present the FishSim Animation Toolchain, a software framework developed to provide researchers with an easy-to-use method for implementing 3D computer animations in behavioral experiments with fish. The toolchain offers templates to create virtual 3D stimuli of five different fish species. Stimuli are customizable in both appearance and size, based on photographs taken of live fish. Multiple stimuli can be animated by recording swimming paths in a virtual environment using a video game controller. To increase standardization of the simulated behavior, the prerecorded swimming path may be replayed with different stimuli. Multiple animations can later be organized into playlists and presented on monitors during experiments with live fish.
In a case study with sailfin mollies (Poecilia latipinna), we provide a protocol on how to conduct a mate-choice copying experiment with FishSim. We utilized this method to create and animate virtual males and virtual model females, and then presented these to live focal females in a binary choice experiment. Our results demonstrate that computer animation may be used to simulate virtual fish in a mate-choice copying experiment to investigate the role of female gravid spots as an indication of quality for a model female in mate-choice copying.
Applying this method is not limited to mate-choice copying experiments but can be used in various experimental designs. Still, its usability depends on the visual capabilities of the study species and first needs validation. Overall, computer animations offer a high degree of control and standardization in experiments and bear the potential to 'reduce' and 'replace' live stimulus animals as well as to 'refine' experimental procedures.
Recently, utilizing modern techniques for the creation of artificial stimuli, such as computer animations and virtual reality, has garnered popularity in research1. These methods provide several advantages compared to classic experimental approaches with live stimulus animals1,2. Computer animation enables non-invasive manipulation of the appearance (size, color) and behavior of virtual stimulus animals used in experiments. For example, the surgical removal of the sword in male green swordtails (Xiphophorus helleri) to test mate preferences in females3 was rendered unnecessary by using computer animation in a later study on this species4. Furthermore, computer animations can create phenotypes that are only rarely encountered in nature5. Morphological features of virtual animals may even be altered beyond the natural range of that species4. Particularly, the possible systematic manipulation of behavior is one major advantage of computer animation, since it is almost impossible with live animals6,7.
Various techniques exist to date for creating computer animations. Simple two-dimensional (2D) animations typically derive from a picture of a stimulus moving in only two dimensions and can be created with common software like MS PowerPoint8 or Adobe After Effects9. Three-dimensional (3D) animations, which require more sophisticated 3D graphics modelling software, enable the stimulus to be moved in three-dimensions, increasing possibilities for realistic and complex physical movement6,7,10,11,12. Even virtual reality designs that simulate a 3D environment where live animals navigate have been used13,14. In a recent review Chouinard-Thuly et al. 2 discuss these techniques one by one and highlight advantages and disadvantages on their implementation in research, which notably depends on the scope of the study and the visual capacities of the test animal (see “Discussion”). Additionally, Powell and Rosenthal15 give advice on appropriate experimental design and what questions may be addressed by employing artificial stimuli in animal behavior research.
Since creating computer animation may be difficult and time consuming, the need for software to facilitate and standardize the process of animation design arose. In this study, we introduce the free and open-source FishSim Animation Toolchain16 (short: FishSim; https://bitbucket.org/EZLS/fish_animation_toolchain/), a multidisciplinary approach combining biology and computer science to address these needs. Similar to the earlier published tool anyFish17,18, the development of the toolchain followed the goal to provide researchers with an easy-to-use method for implementing animated 3D stimuli in experiments with fish. Our software consists of a set of tools that can be used to: (1) create 3D virtual fish (FishCreator), (2) animate the swimming paths of the virtual fish with a video game controller (FishSteering), and (3) organize and present prerecorded animations on monitors to live focal fish (FishPlayer). Our toolchain provides various features that are especially useful for testing in a binary choice situation but also applicable to other experimental designs. Moreover, the possible animation of two or more virtual fish enables the simulation of shoaling or courtship. Animations are not bound to a specific stimulus but may be replayed with other stimuli making it possible to change the appearance of a stimulus but keep its behavior constant. The open-source nature of the toolchain, as well as the fact that it is based on the robot operation system ROS (www.ros.org), provide high modularity of the system and offer nearly endless possibilities to include external feedback devices (as the controller or a tracking system) and to adapt the toolchain to one’s own needs in research. In addition to the sailfin molly, four other species are currently usable: the Atlantic molly Poecilia mexicana, the guppy Poecilia reticulata, the three-spined stickleback Gasterosteus aculeatus and a cichlid Haplochromis spp. New species can be created in a 3D graphics modelling tool (e.g., Blender, www.blender.org). To exemplify the workflow with FishSim and to provide a protocol on how to conduct a mate-choice copying experiment with computer animation, we performed a case study with sailfin mollies.
Mate choice is one of the most important decisions animals make in their life history. Animals have evolved different strategies for finding the best mating partners. They may rely on personal information when evaluating potential mating partners independently, possibly according to predetermined genetic preferences for a certain phenotypic trait19, 20. However, they may also observe the mate choice of conspecifics and thereby utilize public information21. If the observer then decides to choose the same mate (or the same phenotype) as the observed conspecific — the “model” — chosen previously, this is termed mate-choice copying (hereafter abbreviated as MCC)22,23. Mate-choice copying is a form of social learning and, hence, a non-independent mate-choice strategy24, which has been observed in both vertebrates25,26,27,28,29 and invertebrates30,31,32. So far, MCC was predominantly studied in fish and is found both under laboratory conditions33,34,35,36,37,38 and in the wild39,40,41,42. Mate-choice copying is especially valuable for an individual if two or more potential mating partners are apparently similar in quality, and a “good” mate choice — in terms of maximizing fitness — is difficult to make43. The quality of a model female herself can affect whether focal females copy her choice or not44,45,46,47. Respectively, “good” or “bad” model female quality has been attributed to her being more or less experienced in mate choice, for example with regard to size and age44,45,46, or by her being a conspecific or a heterospecific47. In sailfin mollies that copy the mate choice of conspecifics39,48,49,50,51, it was found that focal females even copy the rejection of a male52. Since MCC is considered to play an important role in the evolution of phenotypic traits as well as speciation and hybridization21,23,53,54, the consequences of copying a “false” choice may be tremendous in reducing the fitness of the copier55. If an individual decides to copy the choice of another individual, it is important to evaluate if the observed model is a reliable source of information, i.e., that the model itself is making a “good” choice due to him or her being well experienced in mate choice. Here the question arises: what visual features may characterize a reliable model to copy from in sailfin molly females?
A distinct visual feature in female sailfin mollies and other Poeciliids is the gravid spot (also known as ‘anal spot’, ‘brood patch’ or ‘pregnancy spot’). This darkly pigmented area in their anal region derives from melanization of the tissue lining the ovarian sac56. The size and presence of the gravid spot are variable across conspecific females and may further individually change during the progression of ovarian cycles56,57. Gravid spots may serve to attract males and facilitate gonopodial orientation for internal insemination58 or as a means of advertising fertility59,60. Considering the link between the gravid spot and a female’s reproductive status, we predicted that the gravid spot serves as a sign of model female quality by providing information on her current reproductive state to observing focal females. We investigated two alternate hypotheses. First, if the gravid spot is a general sign for maturity, as predicted by Farr and Travis59, it denotes a presumably reliable and experienced model compared to an immature model (without the spot). Here, focal females are more likely to copy the choice of a model with a spot but not that of a model without a spot. Second, if the gravid spot marks non-receptivity due to already developing broods, as predicted by Sumner et al.60, the model is presumably less reliable since non-receptive females would be considered less choosy. In this case, focal females will not copy their choice but that of models without spot. So far, the role of the gravid spot for MCC in sailfin molly females has never been tested, nor experimentally manipulated.
We used FishSim to perform an MCC experiment by presenting virtual stimulus males and virtual model females on computer monitors instead of using live stimulus and model fish as used in the classic experimental procedure49,50,51,61. The general usability of our software has previously been validated for testing hypotheses about mate choice in sailfin mollies12. Here, we tested whether the absence or presence of a gravid spot in virtual model females affects the mate choice of observing live focal females. We first let focal females acclimate to the test tank (Figure 1.1) and let them choose between two different virtual stimulus males in a first mate-choice test (Figure 1.2). Afterwards, during the observation period, the prior non-preferred virtual male was presented together with a virtual model female (Figure 1.3). In a subsequent second mate-choice test, focal females chose again between the same males (Figure 1.4). We analyzed whether focal females had copied the mate choice of the observed model female by comparing her mate-choice decision in the first and second mate-choice test. We performed two different experimental treatments in which we visually manipulated the quality of the virtual model female. During the observation period, we either presented the prior non-preferred virtual male (1) together with a virtual model female with a gravid spot (“spot” treatment); or (2) together with a virtual model female without a gravid spot (“no spot” treatment). Additionally, in a control without any model female, we tested whether focal females chose consistently when no public information was provided.
Figure 1. General overview of the most important experimental steps for a MCC experiment using virtual fish stimuli. (1) Acclimatization period. (2) First mate-choice test: live focal female chooses between virtual stimulus males. (3) Observation period: focal female watches the prior non-preferred male together with a virtual model female with gravid spot. (4) Second mate-choice test: the focal female again chooses between virtual stimulus males. In this example, she copies the choice of the model. Please click here to view a larger version of this figure.
The performed experiments and handling of the fish were in line with the German Animal Welfare legislation (Deutsches Tierschutzgesetz), and approved by the internal animal welfare officer Dr. Urs Gießelmann, University of Siegen, and the regional authorities (Kreisveterinäramt Siegen-Wittgenstein; Permit number: 53.6 55-05).
1. Virtual Fish Design
Note: Find a list of the required hardware and software in the supplementary materials list. A detailed description of the general functionality of FishSim and additional tips and tricks can be found in the User Manual (https://bitbucket.org/EZLS/fish_animation_toolchain/).
Figure 2: Exemplar pictures of female body textures prior to (original) and after manipulation for the "spot" and "no spot" treatment using the picture editing tool GIMP. The dotted circle marks the area that was manipulated. Please click here to view a larger version of this figure.
Figure 3: Screenshots of a scene in FishSim. (A) The empty default scene without a fish, (B) a scene showing a male alone, (C) a scene showing that same male together with a model female with a spot, and (D) a scene showing the identical male and the identical model female without a spot. Please click here to view a larger version of this figure.
2. Animation of Virtual Fish Stimuli
Note: Each type of animation needed for the experiment needs to be prepared only once using one exemplary male scene and one exemplary observation scene (male and female animated together). During the animating process, a swimming path for each fish is created which can later be replayed by any fish, as long as the name is identical (see step 1.3.4).
3. Preparing Animation Playlists for the MCC Experiment
Note: Use FishPlayer to present animations on two monitors to live focal females. Arrange the playlist for each monitor separately to simulate the procedure of the MCC experiment (Figure 1). The tool consists of a main window showing the record playlist for each monitor (Figure 4) and a separate animation window for each presentation monitor.
Figure 4: Screenshot showing the FishPlayer playlists for the left and right monitors in the first part (i. e., the first mate-choice test) of the MCC experiment. Playlist entries are ordered as needed for the first mate-choice test in Treatment 1. Please click here to view a larger version of this figure.
Figure 5: Screenshot showing the FishPlayer playlists for the left and right monitors in the second part (observation period and second mate-choice test) of the MCC experiment. Playlist entries are ordered as needed for the observation period and the second mate-choice test in Treatment 1. Please click here to view a larger version of this figure.
4. Experimental Setup
Figure 6: Experimental setup for the MCC experiment with computer animation. The operating computer connects to two presentation monitors (Monitor 1 and 2) which replay animations to live focal females inside the test tank. For illustration, both LCD monitors are angled to show an animated scene. Please click here to view a larger version of this figure.
5. Running the MCC Experiment
Note: Follow the experimental procedure below to perform one trial of Treatment 1, Treatment 2 or the control MCC experiment using one live focal female (see Figure 1).
6. Data measurement
Following the protocol, we used FishSim to create computer animations of virtual sailfin molly males and females. We further used the toolchain to present animations to live focal females in a binary choice situation to perform an MCC experiment according to the experimental procedure described in Figure 1 and step 5 of the protocol.
In order to determine whether focal females copied the choice of the virtual model female, we measured a focal female's association time for each male within the first and second mate-choice test during the experiments. Several parameters are typically analyzed using association time obtained in the first and the second mate-choice test for each treatment and the control for choice consistency. How the data are being analyzed is not bound to a specific statistical test but can be done in various ways (e.g., parametric/nonparametric tests, repeated measures ANOVA, statistical models) and may depend on the final data structure. For our data analysis, we used R 3.2.262. We uploaded the raw data we obtained in our experiment as well as the R-code we used for our analysis to Figshare (doi: 10.6084/m9.figshare.6792347).
In the current study, we created 15 different virtual model females with a gravid spot for Treatment 1 and identical 15 model females without a spot for Treatment 2. All model females had a virtual standard length (SL) of 50 mm. The relative gravid spot area was 4.7% of the total body surface (excluding fins; as measured with ImageJ v1.51j8) for all females in Treatment 1. Further, we created 30 different virtual stimulus males presented during mate-choice tests, allowing for 15 unique male stimuli pairings. Stimulus males had a virtual SL between 41-45 mm. We performed 15 trials for each treatment and the control for choice consistency. We tested a total number of 55 live focal females descendant from wild sailfin mollies caught on Mustang Island near Corpus Christi, TX, USA in 2014. All focal females were mature adults and were only tested once. Two females had to be excluded due to technical problems during testing. One female was excluded due to stress since she did not acclimate to the test situation and was too afraid to enter either choice zone. The control for side bias in focal fish (protocol step 6.3) required that we further exclude seven females from the final analysis due to their side bias in the first mate-choice test. Altogether, we analyzed a total of n = 15 focal females for each treatment and the control. Focal females had a mean SL of 32 ± 5 mm in Treatment 1, 33 ± 5 mm in Treatment 2 and 33 ± 3 mm in the control for choice consistency. We compared the standard length (SL) of focal females across treatments and the control using a Kruskal Wallis rank sum test for independent data revealing that SL did not differ between treatments and the control (Kruskal Wallis rank sum test: n = 45, df = 2, χ2 = 0.329, p = 0.848).
The most important parameter measured in an MCC experiment is the focal female's association time for each male (protocol step 6.1). Association time is an indirect measure of female mate preference in fish63,64,65,66 and a well-established measure to determine mate choice in sailfin mollies when no direct contact is provided12,48,61,67,68. For each treatment and the control, we first used association time to analyze whether the choosing motivation differed between mate-choice tests. Choosing motivation is defined as the total time a focal female spent in both choice zones within a mate-choice test. However, a change in choosing motivation does not necessarily reflect a change in preference for either male. If choosing motivation is significantly different between the two mate-choice tests it is obligatory to use preference scores instead of absolute association time, for further analysis to ensure comparability within and between treatments (see protocol step 6.2). In our study, choosing motivation of focal females before and after observation of a virtual model female sexually interacting with a male did not differ in Treatment 1 (Wilcoxon signed rank test: n = 15, V = 44, p = 0.379) and in the control for choice consistency (Wilcoxon signed rank test: n = 15, V = 42, p = 0.33). However, choosing motivation was significantly higher after observation of a virtual model female without gravid spot sexually interacting with a male in Treatment 2 (Wilcoxon signed rank test: n = 15, V = 22, p = 0.03).
The most important determinant of whether MCC occurred is a significant increase in time spent/preference scores for the prior non-preferred male from the first to the second mate-choice test22. Transferred to a natural situation, an increase in time spent with the prior non-preferred male, consequently increases the probability that a female will mate with that male. Therefore, the main analysis compared either the absolute times or the preference scores for the prior non-preferred male between the two mate-choice tests. This analysis has to be done for each treatment and the control separately. Since in our study, choosing motivation differed in Treatment 2, we used preference scores for the initially non-preferred stimulus male, instead of absolute association time, to determine whether these scores changed between the first and second mate-choice test when public information was provided, compared to the control treatment in which public information was absent.
For this, we fit a linear mixed effect (LME) model with the lme function from the 'nlme' package69 with preference score for the prior non-preferred male (pref_NP) as the dependent variable. We included mate-choice test (Mtest: M1, M2) and treatment (treatment: spot, no spot, control) as fixed factors as well as focal female's standard length (SL) as a covariate. To account for the repeated measures design, focal female identity (ID) was included as a random factor. We were especially interested in whether the effect of mate-choice test was different among treatments; therefore, we included an interaction between mate-choice test and the treatment in our model. We conducted two orthogonal comparisons for "treatment" using the function contrasts70. We set the contrasts of the model (1) to compare the control against the mean of all treatments in which any virtual model female was presented during observation [control >> (spot, no spot)], and (2) to compare the treatment showing a virtual model female with spot against that without a spot (spot >> no spot). A plot of the standardized residuals of a factor against the fitted values revealed heteroscedasticity of the residual variances for "Mtest". Therefore, we included a weights function using the varIdent class of the lme function to allow for different variances for each level of "Mtest"71,72. We used the R package 'phia'73 for a post hoc analysis with Holm-Bonferroni correction of significant interaction terms. We inspected model assumptions (Q/Q-plots, residuals, residuals against fitted values) for all models visually74. We further compared the distribution of the residuals against a normal distribution using Shapiro-Wilk normality tests. The given p-values were considered significant if p ≤ 0.05.
The results of this analysis are given in Figure 7A and Tables 1 and 2. We found a significant interaction between M2 and the contrast "[control >> (spot, no spot)]" for preference scores of the prior non-preferred male (LME: df = 42, t = -2.74, p = 0.009). However, preference scores were not affected by focal female SL. Further post hoc analysis of the interaction term revealed a significant difference of preference scores of the prior non-preferred male in M2 in Treatment 1 (spot: df = 1, χ2 = 30.986, p < 0.001) and Treatment 2 (no spot: df = 1, χ2 = 19.957, p < 0.001) but not for the control (χ2-test: df = 1, χ2 = 2.747, p = 0.097). Here, our results demonstrate that, as predicted for MCC, preference scores for the prior non-preferred virtual male significantly increased from M1 to M2 after focal females had been presented with the simulated mate choice of a virtual model female. We found this effect in both treatments but not in the control for choice consistency. Instead, in the control where no model female was present during the observation period, focal females were consistent in their mate choice for a male.
Factors | Lower | Estimate | Upper | SE | df | t-value | p-value |
(Intercept) | 0.046 | 0.339 | 0.632 | 0.145 | 42 | 2,336 | 0.024 |
M2 | 0.207 | 0.296 | 0.384 | 0.044 | 42 | 6.750 | < 0.001 |
Control → (spot, no spot) | -0.041 | -0.012 | 0.017 | 0.014 | 41 | -0.852 | 0.4 |
Spot → no spot | -0.093 | -0.043 | 0.008 | 0.025 | 41 | -1,715 | 0.094 |
SL | -0.012 | -0.003 | 0.006 | 0.004 | 41 | -0.747 | 0.459 |
M2 x [control → (spot, no spot)] | -0.148 | -0.085 | -0.023 | 0.031 | 42 | -2,743 | 0.009 |
M2 x (spot → no spot) | -0.067 | 0.042 | 0.15 | 0.054 | 42 | 0.777 | 0.441 |
Table 1. LME estimates for effects on preference scores for the prior non-preferred virtual male. Preference score for the prior non-preferred virtual male stimulus was the dependent variable throughout. Given are estimates ± standard error and lower/upper confidence intervals, degrees of freedom, t-values and p-values for each fixed factor. Intercept estimates represent the grand mean of all treatments. Orthogonal comparisons of treatments are given. If treatments are combined in parentheses, mean values of these treatments are used in the comparisons. The intercept reference category for factor "M2" is "M1". Significant p-values (p < 0.05) are printed in bold. M1 = first mate-choice test, M2 = second mate-choice test, SL = standard length of focal females. 90 observations with n = 15 focal females per treatment.
Random factor | Variance | SD |
ID ((Intercept)) | 1.464×10-10 | 1.21×10-5 |
Residual | 1.859×10-2 | 0.1364 |
Table 2. LME variance components for focal female ID. Variance and standard deviation for the random effect "focal female ID" and the residuals are given.
Figure 7: Results of the virtual MCC experiment manipulating model female quality by the visual absence or presence of a gravid spot. (A) Preference scores for the (prior) non-preferred virtual stimulus male in M1 and M2 for both treatments and the control. (B) Change of preference from M1 to M2 (copying score) for the prior non-preferred virtual male in the treatments and in the control. The dotted line depicts no change in preference, positive values show an increase in preference and negative values show a decrease in preference. Grey dots in A and B depict raw data of each focal female. (C) Number of mate-choice reversals in M2 for each treatment and the control. M1 = first mate-choice test, M2 = second mate-choice test, ns = not significant, * = p < 0.05, ** = p < 0.01, *** = p < 0.001. N = 15 for both treatments and the control. Please click here to view a larger version of this figure.
To obtain additional information about whether a copying effect might be more or less strong depending on the respective treatment, a comparison between copying scores and the number of mate-choice reversals between the different treatments and the control was conducted. Therefore, we further analyzed whether copying scores for the prior non-preferred male were different across treatments. The copying score for a male describes the change in female preference for a male from the first to the second mate-choice test. The copying score is defined by the preference score of a male in the second mate-choice test minus the score of that same male in the first mate-choice test. Copying scores range between -1 and +1 and can either be positive or negative values in which negative values describe a decrease in preference and positive values an increase in preference for that male. Here, we fit an LME with copying score (copy_NP) as the dependent variable, treatment as a fixed factor, focal female SL as a covariate and focal female's spot area as a random factor. Here, we conducted the same two orthogonal comparisons for "treatment" as described above.
As we show in Figure 7B and Tables 3 and 4, we found a significantly higher copying score for the prior non-preferred male in treatments with a virtual model female compared to the control (LME: df = 20, t = -2.833, p = 0.01) but no significant difference between treatments (LME: df = 20, t = 0.618, p = 0.544). Copying scores were not affected by focal female SL.
Fixed factors | Lower | Estimate | Upper | SE | df | t-value | p-value |
(Intercept) | -0.889 | -0.081 | 0.727 | 0.389 | 21 | -0.208 | 0.837 |
Control → (spot, no spot) | -0.153 | -0.088 | -0.023 | 0.031 | 20 | -2.833 | 0.01 |
Spot → no spot | -0.079 | 0.033 | 0.146 | 0.054 | 20 | 0.618 | 0.544 |
SL | -0.013 | 0.011 | 0.035 | 0.011 | 20 | 0.991 | 0.333 |
Table 3. LME estimates for effects on copying scores for the prior non-preferred virtual male. Copying score for the prior non-preferred virtual male stimulus was the dependent variable throughout. Given are estimates ± standard error and lower/upper confidence intervals, degrees of freedom, t-values and p-values for each fixed factor. Intercept estimates represent the grand mean of all treatments. Orthogonal comparisons of treatments are given. If treatments are combined in parentheses, mean values of these treatments are used in the comparisons. Significant p-values (p ≤ 0.05) are printed in bold. SL = standard length of focal females. 45 observations with n = 15 focal females per treatment.
Random factor | Variance | SD |
spot_area ((Intercept)) | 0.028 | 0.166 |
Residual | 0.075 | 0.275 |
Table 4. LME variance components for focal female spot area. Variance and standard deviation for the random effect "spot_area" and the residuals are given.
Additionally, we tested whether the number of focal females that reversed their initial mate preference in M2 differed across treatments. Mate-choice reversal is defined as whether there is a change in the preference for a male (from less than 50% to more than 50% of the time in both choice zones) from the first to the second mate-choice test. Mate-choice reversal is counted as a "Yes" (preference for a male has changed) or a "No" (preference for a male did not change). Here, we performed a post hoc pairwise G-test using the R package 'RVAideMemoire'75 with correction for multiple testing. As we show in Figure 7C, eleven out of 15 focal females reversed their mate choice in Treatment 1 and ten females reversed their mate choice in Treatment 2. On the other hand, only two reversals were observed in the control. Thereby, the number of focal females that reversed their initial mate choice in favor of the prior non-preferred male in M2 was significantly larger in both treatments compared to the control (post hoc pairwise G-test: Spot vs. control, p = 0.002; no spot vs. control, p = 0.003) but not significantly different between Treatments 1 and 2 (Post-hoc pairwise G-test: spot vs. no spot, p = 0.69).
The gravid spot in sailfin molly females was previously described to serve as a means of fertility advertisement towards conspecific males59,60. Whether a gravid spot may also provide information to conspecific females in the context of mate choice had not been tested so far. In the present case study, we investigated the potential role of a gravid spot as a source of public information for observing conspecific females in the context of MCC. Our study shows that the gravid spot seems to not be a sign of model female quality for live focal females when deciding to copy the mate choice of a virtual model female for a virtual male. Focal females copied the choice of a virtual model female for a prior non-preferred virtual male regardless of whether the model female had a gravid spot or not. We found no difference in copying scores nor the number of mate-choice reversals between the two treatments, indicating that the copying effect was also equally strong whether the model female had a gravid spot or not. When no public information was provided in the control (no model female present), focal females were consistent in their mate choice. This supports that the observed change of preference within treatments can be explained by the presence of the virtual model female only, providing sufficient public information for copying the mate choice of others.
Even though the general presence and extent of the gravid spot are considered to be linked to the female reproductive cycle, with the spot being largest prior to parturition and smallest or absent after giving birth60, systematic visual observations of the development of gravid spots in individual females are still missing. Moreover, variation in gravid spot size can be high between individual females with spots also being completely absent in mature, gravid females60. Even though sailfin molly females are most receptive short after parturition59,76, they are able to store sperm for several months57. Therefore, females should always be choosy for the best quality mate. With regards to our case study on MCC and the tested hypotheses, we conclude that a gravid spot may not be a valid indicator of model female quality to observing conspecific females. Information on the reproductive state of a model female that an observing female might possibly gain seems to not be important in the decision to copy her choice or not, at least among sailfin mollies.
Notably, our study demonstrates a highly standardized procedure for visual manipulation of public information provided in MCC experiments by using computer animated fish. In contrast to an earlier study by Benson77, who injected live fish with tattoo ink to manipulate gravid spots, our method provides a completely non-invasive alternative for visual manipulation. We described in detail the procedure on how to create and animate virtual sailfin mollies in FishSim. We further showed how computer animation can be used to adopt the experimental procedure of a classic MCC experiment with virtual fish for the presentation towards live test fish in a binary choice situation.
Following the protocol, we identify several critical steps that need specific attention to ensure the correct handling of our toolchain and the success of the experiment. Since computer animations are created and presented using computers and display devices such as computer monitors, the technical equipment should always be good enough to ensure a smooth processing of the general workflow and, most importantly, the playback of the animation (steps 2, 3, and 5). When using two or more monitors for presentation of stimuli, their technical specifications should be identical. When using our software, the set monitor resolution should always be that of the presentation monitors (see step 1.2.1.). Setting the scene (step 1.2.) as well as the design (steps 1.3. and 1.4.) and animation (step 2) of virtual stimuli should always be done on a monitor later used for stimulus presentation during experiments to ensure the correct dimensions.
In this protocol, we concentrate on the necessary steps to create one set of fish stimuli (steps 1.3. and 1.4.) for the use in one trial of a treatment (step 5). Here, we would like to point out that it is important to create several different fish stimuli and/or animations to account for pseudoreplication15,78,79 which affects the possible interpretation of the data obtained during experiments. With our toolchain, it is easy to create various fish stimuli offering possibilities to use a unique set of stimuli for each experimental trial. Overall the total number of stimuli needed depends on the intended sample size for each treatment (see "Representative results" section for information on our case study).
With our toolchain, we wanted to provide a fast and easy-to-create animation process by using a video game controller (step 2). Thereby, the general swimming behavior of the virtual fish is automatically generated, based on videos of swimming live sailfin mollies80. Swimming behavior (including movement of fins and gonopodium) is, therefore, tuned to the use with virtual stimuli of sailfin mollies in particular and live-bearing fish in general. Apart from live-bearing fish, an additional template for a three-spined stickleback provides additional functions for species-specific movement, such as raising/lowering of dorsal and ventral spines.
Animation functions currently provided by our toolchain might not be sufficient for every behavioral pattern and fish species. This, however, is up to the user and depends on the tested research question. Further, animation with FishSteering (step 2) needs a little practice beforehand to get accustomed to the functions of the gaming controller. Therefore, the animation process is probably the most time-consuming step of the protocol. A controller of a different brand may be used here but the functionality might not be that smooth and the button functions (as given in the user manual) may be different or completely absent. During the animation process, only one feature of a virtual stimulus (e.g., position, fins, gonopodium) can be animated at a time. First, the swimming movement (position) and afterwards additional features (e.g., fins) may be added independently. We recommend saving each step separately. This offers the advantage that the user has the possibility to come back to an earlier version of the animation to change a specific feature, for example keeping the swimming path constant but changing the dorsal fin movement compared to a previous version. Especially when animating more than one fish (step 2.2.), the order in which fish stimuli are animated is very important and needs to be determined beforehand. Here, it might be helpful to refer to the biology of the tested species. In our case study, we simulated the courtship behavior of sailfin mollies in which a male is generally following a female81. Hence, we first created the swimming path of the virtual female and added the path for the virtual male by following the female.
When running the experimental procedure (step 5) the timing is crucial for the success of the experiment. The times/durations we referred to in the protocol (step 5) derived from previous studies with sailfin mollies. They should be regarded as suggestions and are not obligated for the general success of the experimental but should, nevertheless, be tightly followed during the procedure. Acclimatization time may vary between fish species and even individuals and should generally be as long as the focal fish needs to explore the whole test tank and acclimate to its new surroundings. We determined the appropriate pause duration length in training runs of the experimental procedure. The pause should be at least as long as the time needed for catching the fish with the cylinder, as well as walking to and from the test tank and operational computer to release the fish from the cylinder. Here, times possibly vary depending on the specific experimental situation in each lab and the tested fish species. In any case, the experimenter may individually change times/durations either by setting a different time in FishPlayer (see step 3.1. 3.) or by creating animation sequences with a different length (see step 2.1).
The experimenter can improve the measuring of association time for each mate-choice test by implementing an automated tracking system, though it needs to be capable of tracking in real-time. Here, we also want to point out that there is no possibility of having a blind observer and, hence, blind analysis when following the procedure for testing MCC. Since the experimenter cannot know which virtual male stimulus will be preferred by the focal female prior to testing, he or she needs to be aware of the focal fish's choice to rearrange the order of animation sequences accordingly (see steps 3.2 and 5.10).
The protocol we describe here is specific to our study design on MCC in sailfin mollies. However, the toolchain can also be used in combination with other experimental designs with up to four monitors for presentation. In general, computer animation tools offer a wide variety of solutions to study various questions on fish behavior like mate choice, shoaling decisions or predator-prey detection, using artificial visual stimuli. General technical and conceptual considerations for the use of computer animation in animal behavior research should be carefully evaluated before using it in experiments2,15. Most important for the decision whether computer animation approaches can be implemented in research regards the visual capabilities of the tested fish species and whether it responds naturally towards virtual stimuli presented on monitor screens. Especially, when testing the effect of color aspects, it should be noted that monitor screens only depict colors as RGB values and that this might impede or limit research possibilities, although RGB color output may indeed be adjusted82. For some fish species, a limitation might certainly be that monitors do not emit UV wavelengths or that, on the other hand, certain monitor types are highly polarized which might be a limitation with fish being sensitive to polarized light for example in questions of mate choice83. Therefore, a validation for the effectiveness of presented stimuli as computer animations is necessary before testing any hypotheses2,12,15,84,85.
In the future, new developments in animal tracking and action recognition might make it possible to create interactive virtual stimuli that react in real-time towards live fish and simulate corresponding behavior to massively increase realism for observing fish86. Thanks to the modularity of the underlying ROS, external devices such as cameras may be integrated into the toolchain, provided that the user has adequate programming skills. A first successful attempt showed that FishSim can generally be used to simulate interactive virtual fish stimuli by extension of a 3D real-time tracking system87,88,89. During the science communication event "Molly knows best" (https://virtualfishproject.wixsite.com/em2016-fisch-orakel), we were able to demonstrate that virtual fish can be programmed to follow live focal fish on screen and perform courtship behavior according to a predefined algorithm. Further, such real-time tracking systems could be used to measure association time automatically to enhance experimental procedure. This feature is not yet included in the current version of FishSim but is subject to future development.
In conclusion, the use of computer animation in animal behavior research is a promising approach when conventional methods would require invasive treatment of live animals to manipulate the expression of a visual trait or behavioral pattern. Manipulating computer animations allows for a high degree of control and standardization compared to using live test fish, especially, since it also offers solutions to manipulate behavior which is very limited or even impossible in live fish. Further, in line with the 3Rs-principle and similar guidelines for the use of animals in research and teaching90,91, this technique bears the potential to 'reduce' and 'replace' live test animals as well as to 'refine' experimental procedures in research.
The authors have nothing to disclose.
This work was supported by the Deutsche Forschungsgemeinschaft DFG (WI 1531/12-1 to KW and SG and KU 689/11-1 to KDK, KM and JMH). We sincerely thank the DAAD RISE Germany program for providing and organizing an undergraduate research internship between SG and DB (Funding-ID: 57346313). We are grateful to Mitacs for funding DB with a RISE-Globalink Research Internship Award (FR21213). We kindly thank Aaron Berard for inviting us to introduce FishSim to the JoVE readership and Alisha DSouza as well as three anonymous reviewers for their valuable comments on a previous version of the manuscript.
Hardware | |||
2x 19" Belinea LCD displays | Belinea GmbH, Germany | Model 1970 S1-P | 1280 x 1024 pixels resolution |
1x 24" Fujitsu LCD display | Fujitsu Technology Solutions GmbH, Germany | Model B24-8 TS Pro | 1920 x 1080 pixels resolution |
Computer | Intel Core 2 Quad CPU Q9400 @ 2.66GHz x 4, GeForce GTX 750 Ti/PCIe/SSE2, 7.8 GiB memory, 64-bit, 1TB; keyboard and mouse | ||
SONY Playstation 3 Wireless Controller | Sony Computer Entertainment Inc., Japan | Model No. CECHZC2E | USB-cable for connection to computer |
Glass aquarium | 100 cm x 40 cm x 40 cm (L x H x W) | ||
Plexiglass cylinder | custom-made | 49.5 cm height, 0.5 cm thickness, 12 cm diameter; eight small holes (approx. 5 mm diameter) drillt close to the end of the cylinder lower the amount of water disturbance while releasing the fish | |
Gravel | |||
2x OSRAM L58W/965 | OSRAM GmbH, Germany | Illumination of the experimental setup | |
2x Stopwatches | |||
Name | Company | Catalog Number | コメント |
Software | |||
ubuntu 16.04 LTS | Computer operating system; Download from: https://www.ubuntu.com/ | ||
FishSim Animation Toolchain v.0.9 | Software download and user manual (PDF) from: https://bitbucket.org/EZLS/fish_animation_toolchain | ||
GIMP Gnu Image Manipulation Program (version 2.8.22) | Download from: https://www.gimp.org/ |