Female brain size affects the assessment of male attractiveness during mate choice

We quantified female preference in a standard dichotomous choice setup (see Methods for details). During the first 10 min of the trial, when overall preference levels were highest (Fig. 1B and information S1), we found an overall effect of brain size in the preference for attractive males [means ± SE: small-brained: −0.07 ± 0.08, large-brained: 0.24 ± 0.09, wild-type: 0.36 ± 0.12; LMM_preference: brain size: χ²(1) = 7.404, P = 0.006]. Both large-brained and wild-type females displayed significant preference for attractive males compared to unattractive males (Fig. 1). Small-brained females showed no preference for either attractive or unattractive males over the whole duration of the trial (Fig. 1). Preference for attractive males by large-brained and wild-type females differed significantly from small-brained [LMM_preference: small-brained versus large-brained: χ²(1) = 6.952, P = 0.008; LMM_preference: small-brained versus wild-type: χ²(1) = 8.660, P = 0.003]. There was no significant difference in average preference for attractive males between wild-type and large-brained females [LMM_preference: large-brained versus wild-type: χ²(1) = 0.662, P = 0.414].

We verified whether the observed differences could be attributed to differences in how females appraised information on the offered males. First, the observed differences could not be attributed to differences in sexual activity levels because females from all groups spent, on average, the same amount of time outside of the defined choice areas containing male pairs [means ± SE: small-brained: 203 ± 19 s, large-brained: 184 ± 26 s, wild-type: 243 ± 34 s; LMM_{no choice}: brain size: χ²(2) = 3.015, P = 0.221] and the total time spent in front of attractive males was significantly higher in large-brained females than in small-brained females [means ± SE: small-brained: 200 ± 28 s, large-brained: 285 ± 38 s; LMM_{total time}: small-brained versus large-brained: χ²(1) = 4.870, P = 0.027]. In addition, analyses of absolute preference irrespective of male attractiveness, controlling for individual activity and choice behavior, showed no differences between small-brained, large-brained, and wild-type females (Fig. 2). Second, we found no evidence that large- and small-brained females differed in their opportunity to gather information about unattractive and attractive males because the ratio of visits between choice areas (difference in visits to each choice area, standardized by the total visits) was not different between brain sizes [means ± SE: small-brained: −0.02 ± 0.06, large-brained: 0.08 ± 0.06; LMM_{visits ratio}: small-brained versus large-brained: χ²(2) = 1.981, P = 0.159].

We evaluated the possibility that the observed differences in female preference were caused by underlying differences in the way females of the selection lines perceive male color. We initially estimated female sensitivity to orange colors by measuring their optomotor response (see Methods for details). These experiments are based on the fact that fish orient their position using objects as reference and thus rotate to follow a rotating stimulus, usually bands of alternating color. By decreasing or increasing the saturation intensity of the color bands and tracking fish movement, it is possible to test whether fish are able to discern between the different color bands at different levels of contrast. In the first three rotating stimuli, with a higher saturation difference between red and green bands, of the optomotor response test, all female groups could perceive the contrast between bands and spent, on average, significantly more time performing optomotor behaviors during the rotational phase than during the static phase (mean of the differences: 39.130 s, t₅₀ = −27.689, P < 0.001) (Fig. 3). We then decreased the saturation intensity of the red and green stripes used as stimuli and observed no differences among the female groups in the average time spent performing optomotor behaviors during the rotational phase in relation to the static phase (mean of the differences: 0.655 s, t₅₀ = 0.454, P = 0.651) (Fig. 3). In the final rotating stimuli round, with maximum intensity of red and green bands, all female groups again spent significantly more time performing optomotor behaviors during the rotational phase, with similar magnitude as in the first round with high band contrasts (mean of the differences: 38.953 s, t₅₀ = −20.525, P < 0.001). Thus, the lack of difference observed in the rotating stimuli with lower saturation intensity in green and red bands was not likely a consequence of habituation to the test or caused by stress but rather indicates that the females were less able to perceive the contrast between the color stripes. We found that wild-type, small-brained, and large-brained females did not show differences in their sensitivity to color contrasts in either high saturation contrasts [means ± SE: small-brained: 46.72 ± 1.79 s, large-brained: 49.97 ± 1.73 s, wild-type: 47.54 ± 1.84 s; LMM_{optomotor response}: brain size: χ²(2) = 1.84, P = 0.398] or low saturation contrasts [means ± SE: small-brained: 20.28 ± 2.48 s, large-brained: 17.13 ± 2.40 s, wild-type: 19.68 ± 2.54 s; LMM_{optomotor response}: brain size: χ²(2) = 0.774, P = 0.679] (Fig. 3). Post hoc comparisons among groups did not show any differences for any of the multiple separate band contrasts tested (table S2).

To evaluate another component of the visual system that could have a strong effect on color perception, we measured the expression of opsin genes in females tested for male preference. Opsins are the proteins in the retina that mediate the initial steps of photon capture that lead to vision. Differences in opsin expression have been linked to differences in the light environment occupied by different species (35), and opsin expression has also been linked to female preferences in guppies (36). We found no significant expression differences between large- and small-brained females for any of the 10 guppy opsin genes (table S3A). We also measured the expression of opsin genes in males from each group and confirmed previous results showing significant sex differences in the expression of short- and long-wavelength opsins (table S3B) (37), indicating that our expression assay contains sufficient sensitivity. Together with the results from the optomotor response experiment, our data thus indicate that large- and small-brained females do not differ in these two aspects of the visual system and are therefore unlikely to differ in their physiological ability to perceive the attractiveness in males.

We studied the preference for colorful males in female guppies from laboratory-reared descendants of Trinidadian guppies from high predation areas of the Quare River. We used wild-type female guppies from this laboratory population and existing artificial selection lines (based on the same population) for small and large relative brain size, with >10% difference in relative brain size (17, 23). Briefly, the artificial selection experiment was based on indirect selection for parental brain weight data corrected for body size, which was used to generate replicated lines with large and small relative brain size (three replicates for large-brained and three replicates for small-brained; six populations in total). See Kotrschal et al. (17) for full details on the selection experiment. These selection lines showed an 11% difference in relative brain size in the third generation (42) and an up to 13.6% difference in the fourth generation (23). After the fourth generation, 30 non-sib males and females from each population were paired to generate a fifth generation of brain size–selected offspring.

Here, we used a total of 16 wild-type females and 36 large-brained and 36 small-brained females from the fifth generation of the brain size selection lines (12 individuals from each of the three up- and down-selected lines). These fish had not been used in any experiments before these behavioral tests. The sample sizes were chosen on the basis of previous experience in effect sizes of the different selection lines and due to constraints in the availability of wild-type females. All offspring were removed from their parental tanks after birth. Males were isolated from females as soon as their gonopodia began to develop and before any color pattern appeared on the males. Females were then placed in 12-liter tanks in groups of 10 fish. One week before the experiment, females were isolated in individual tanks, but we allowed visual contact between tanks containing females to avoid social stress from isolation. However, females did not have visual contact with any mature male before the experiment. All virgin females were tested approximately 6 months after birth. The laboratory was maintained at 26°C with a 12-hour light/12-hour dark schedule, which resulted in a water temperature of 25°C. Fish were fed an alternating daily diet of flake food and live brine shrimp.

To present males with large differences in coloration, we used 60 males for which we quantified coloration using image analysis. For this, we first anesthetized the males with a low dose of benzocaine and photographed both sides using a Nikon D5300 camera. We quantified body length, tail area, and area of orange, black, and iridescence coloration using ImageJ software version 1.44 (58). To establish that all males were sexually mature, we first housed the males together with nonparticipant females in the experiment for 2 hours. The two males that did not perform sexual displays toward the female were discarded. Of the 58 remaining males, we selected the 8 males with the highest and lowest combined scores of orange, black, and iridescence area, and then we size-matched them to create eight pairs consisting of one highly attractive male and one unattractive male. Males within pairs did not differ in body length (Welch t = −1.16; df = 13.79; P = 0.26) but presented a large difference in average total coloration, with 26% more area of total coloration in attractive males (Welch t = −8.66; df = 13.79; P < 0.001), 10% more orange coloration in attractive males (Welch t = −8.76; df = 13.79; P < 0.001), and 27% larger tails in more attractive males (Welch t = −2.91; df = 13.79; P < 0.001). To ensure that the differences in attractiveness persisted throughout the mate choice trials, we also quantified the differences in male attractiveness after the experiment and found that total and orange coloration differed by 18 and 6%, respectively (Welch t = −4.95; df = 14; P < 0.001 and Welch t = −3.73; df = 14; P = 0.002, respectively).

We tested 36 large-brained and 36 small-brained females for their mating preference for attractive males by measuring side-association data in a dichotomous choice setup (34, 59). Previous studies in this species have shown consistent results between side-association data and direct measurements of mating preference [reviewed by Houde (34)]. The setup consisted of eight plain glass tanks of 42 cm × 20 cm × 20 cm, where females were allowed to observe an attractive versus unattractive male pair for a 15-min period. Females could not observe both males simultaneously because males were presented in adjoining plain glass tanks of 11 cm × 10 cm × 20 cm with a 10-cm separation (fig. S4). Tests took place over nine consecutive days. All fish were netted and placed in their respective experimental tank 24 hours before the test to allow for acclimation and hence avoid potential differences in stress response between the lines. A nontransparent plastic film was placed on the side walls of the male tanks to avoid visual interactions between males. Equal numbers of attractive and unattractive males were presented to the right and left position toward the female tank for every treatment in a randomized order. We found no significant side bias in preference between the wild-type, large-brained, and small-brained fish (table S5). Every trial was broadcasted live using Logitech HD Webcam C615 from a top position and viewed from a distance on a laptop to avoid disturbance. The position of the female was scored by a single observer using the live observation mode in JWatcher software version 1.0 (60). The female experimental tank was divided into three zones to determine female position: (i) left choice zone, the area adjacent to the left male tank up to a maximum distance of 10 cm from it; (ii) right choice zone, the area adjacent to the right male tank up to a maximum distance of 10 cm from it; and (iii) no choice zone, the area between the left and right choice zones and all areas further away than 10 cm from the male tanks (fig. S4). Quantification of behaviors was performed blind to the treatment because only running numbers identified females and male position. To control for the total time during the test that every female spent associating with the offered males, we used a preference ratio widely used in dichotomous choice preference tests (34). This preference ratio was obtained as the difference in time spent with each male, standardized by the total amount of time in any of the choice areas [(time with colorful male – time with dull male)/(time with colorful male + time with dull male)].

After completion of the preference analyses in large- and small-brained females, we studied the preference for attractive versus unattractive males in 16 wild-type females of the same age. We repeated the procedure used for large- and small-brained females with a new set of four male pairs over two consecutive days.

To test whether females from all groups demonstrated partner preference irrespective of male attractiveness (color traits and tail length), we studied the absolute value of the preference ratio. The absolute preference could take values between 0 and 1 where 0 indicates that a female divided her time evenly between both males, whereas a score of 1 indicates that a female spent time exclusively with one of the males. To determine whether females significantly preferred one male over the other, regardless of their attractiveness levels based on coloration and tail size, we calculated whether their absolute preference was closer to 1 than expected by chance. The expected value in the case of no preference was affected by the activity and choice dynamics of the female. If she only moved a few times between choice zones, the absolute preference was expected to be closer to 1 than if she moved many times. To control for this, we performed Monte Carlo simulations that constrain the number of zone changes. For each female, we took the sequence of zone changes, randomly replaced the identities for the males, and then calculated the absolute preferences after random assignment. For each group (brain size by replicate), we calculated the mean absolute preference and compared it to 10,000 simulated mean absolute preferences. We calculated two-sided P values by counting how many simulations had an equal or more extreme result than our observation. For the small- and large-brained females, we combined the separate results for the three replicated selection lines into a single P value using Stouffer’s z score method (information S6) (61).

Data analyses on absolute preference values (information S1) showed a significant decay of absolute preference over the time of the trial [LMM_{abs. preference}: time: χ²(14) = 36.215, P < 0.001]. Because of this, we separated the data into three time periods of 5 min and independently tested these for potential differences in female preference for attractive males using the standardized preference ratio (Fig. 1B). We chose 5-min periods to ensure that enough preference data were available for all females in the trials. Standardized preference ratio indicates the relative proportion of time that a female spent with the attractive male based on coloration and tail size. Standardized preference ratio could take values between −1 (all time spent with the noncolorful/small-tailed unattractive male) and 1 (all time spent with the colorful/large-tailed attractive male). We tested for potential differences in preference for attractive males between the wild-type, large-brained, and small-brained females with a linear mixed-effects model (LMM). Models included brain size treatment as a fixed effect. The random effects portion of the model included a random intercept for each replicate selection line and a random slope for brain size within each replicate. Also, we included male pair and experimental tank as random factors (information S6). Preference differed between female groups in the initial 5 min of the trial and in the second 5-min period [LMM_preference: brain size: χ²(2) = 6.985, P = 0.030 and LMM_preference: brain size: χ²(2) = 9.124, P = 0.010, respectively]. We observed no difference in the third 5-min period [LMM_preference: brain size: χ²(2) = 1.664, P = 0.435] (Fig. 1B). Hence, to investigate potential differences in female preference for attractive males, we focused on the data obtained during the first 10 min of each trial. We then tested for differences in preference between wild-type females and large- or small-brained females with a similar LMM approach. In the case of tests for potential differences in preference between large- and small-brained females, we also included the day of the experiment as a random factor in the LMM (information S6). All statistical analyses were performed in R version 3.2.2 (62). The assumptions of normality and equality of variances were confirmed by visual inspection of the residuals.

We quantified the visual capacity of sensitivity of females to male coloration using optomotor response tests (63). Optomotor responses of several fish species have been studied by rotating an array of stripes to a chosen threshold speed (64). This test has also previously been used to assess the visual performance of guppies (65), which exhibit several stereotypical behaviors when subjected to a rotational stimulation: (i) optomotor circling, that is, swimming along to follow the stripes in a rotating circular pattern; (ii) compass reaction, that is, rotating of the body around its own midpoint to follow the stripes in a rotating circular pattern; (iii) nystagmic reaction, that is, turning of the head to follow the stripes in a rotating circular pattern followed by a rapid return to the initial position; and (iv) zero motion, that is, swimming without a clear circular motion or remaining static (65).

We performed a variant of the optomotor test on the 16 wild-type and on 18 large-brained and 18 small-brained female guppies randomly selected from the previous preference test (six individuals per replicated relative brain size–selected line). We generated a rotating stimulus with stripes of maximum peak transmittance of 550 and 630 nm. We selected these wavelengths to test the sensitivity of these guppies to a color relevant in male nuptial traits. Orange spots of male guppies in our laboratory population have peak transmittances around 630 nm (66). Because of this, we generated a video recording with six different rotating stimuli consisting of 16 alternating bands of red and green. These colors showed peak reflectance around 550 and 630 nm, respectively, when projected to a white background with an InFocus IN114 projector. To generate a gradual shift in the visibility of the successive rotating stimulus, we gradually lowered the intensity in red and green bands by 20% until both bands appeared the same (fig. S7). In addition, we added a seventh rotating stimulus with the initial maximum saturation contrast between bands. This final change in the rotating stimulus was added to control for the effect of stress and/or habituation to the stimuli. For each rotating stimulus, a minute of acclimation time was included before the start of the rotation in the video. This acclimation time consisted of a static image of the contrasting bands that would rotate next.

In a restricted randomized order (one female per treatment), females were individually placed in a circular white tank with a diameter of 25 cm. After 2 min of acclimation, the video recording was projected on the inside of the walls of the tank using an InFocus IN114 projector (fig. S7). Behavior of the fish was recorded using a Sony HDR-SR11E camcorder and later scored by a single observer using BORIS version 2.72. We scored the time that each fish spent performing the four stereotypical rotating behaviors for each acclimatization and rotation stimulus phase. Quantification was performed blind to the treatment. We found no difference between treatment groups in the time spent performing optomotor circling, compass reaction, or nystagmic reaction for any analyzed moving or acclimating phase (table S8). Therefore, we created a new variable, total rotating behavior, which combined the time of all three behaviors described above [(i) to (iii)]. We then tested each rotational stimulus independently for potential differences in total rotating behavior between the large-brained, small-brained, and wild-type females with an LMM. Models included brain size treatment as a fixed effect. The random effects portion of the model included a random intercept for each replicate selection line and a random slope for brain size within each replicate. Also, we included optomotor response during the static phase as a covariate in the model (information S6). All statistical analyses were performed in R version 3.2.2 (62). The assumptions of normality and equality of variances were confirmed by visual inspection of the residuals.

We measured the opsin expression of nine wild-type, nine large-brained, and nine small-brained females previously tested on their preference for attractive males. After dispatch, eyes were removed and immediately placed in RNAlater. Total RNA was purified using RNeasy kits (Qiagen) following standard manufacturer protocols. Expression level was quantified using the NanoString nCounter assay (67), which involves hybridizing target sequences in the sample by complementary base pairing to both a reporter probe and a capture probe. We designed probes for the cone opsin genes (LWS A180, LWS P180, LWS1, LWS S180, SWS1, SWS2A, SWS2B, RH2-1, and RH2-2) as well as rhodopsin (RH1). Two additional probes were designed in conserved regions shared by LWS duplicates to verify the specificity of probes. All probe sequences can be found in table S9. Hybridizations were carried out according to NanoString standard protocols. After hybridization, expression data were normalized on the basis of manufacturer spike-in controls to account for technical variance and four housekeeping genes (β-actin, COI, Gapdh, and Myosin HC) to control for sample input. We also sequentially removed each housekeeping gene from the normalization to test for outlier normalization effect. No single housekeeping gene substantially affects normalization of target loci.

A given target gene was considered to be significantly expressed if its absolute count exceeded the mean spike-in negative controls by at least 2 SDs, and all of our target genes met this threshold criterion. We analyzed the differences in expression independently for each measured opsin. Models using an LMM approach with a random intercept and slope for brain size for each replicate did not converge because of low sample size for each replicate/brain size combination. Therefore, we determined differential expression after log₂ transformation of expression data for our target genes using paired t tests in combined values for each relative brain size treatment.