The authors have declared that no competing interests exist.
Conceived and designed the experiments: DIB CDB. Performed the experiments: CDB KCS. Analyzed the data: DIB. Contributed reagents/materials/analysis tools: MS. Wrote the paper: DIB.
Multiple biological processes can generate sexual selection on male visual signals such as color. For example, females may prefer colorful males because those males are more readily detected (perceptual bias), or because male color conveys information about male quality and associated direct or indirect benefits to females. For example, male threespine stickleback often exhibit red throat coloration, which females prefer both because red is more visible in certain environments, and red color is correlated with male immune function and parasite load. However, not all light environments favor red nuptial coloration: more tannin-stained water tends to favor the evolution of a melanic male phenotype. Do such population differences in stickleback male color, driven by divergent light environments, lead to changes in the relationship between color and immunity? Here, we show that, within stickleback populations, multiple components of male color (brightness and hue of four body parts) are correlated with multiple immune variables (ROS production, phagocytosis rates, and lymphocyte:leukocyte ratios). Some of these color-immune associations persist across stickleback populations with very different male color patterns, whereas other color-immune associations are population-specific. Overall, lakes with red males exhibit stronger color-immune covariance while melanic male populations exhibit weak if any color-immune associations. Our finding that color-immunity relationships are labile implies that any evolution of male color traits (e.g., due to female perceptual bias in a given light environment), can alter the utility of color as an indicator of male quality.
Animal mating signals are among the most impressive facets of biological diversity. Consequently, biologists have proposed a variety of models of sexual selection to explain signal evolution [
Multiple forms of sexual selection can act concurrently on a single trait. For example, bright male colors can serve to attract female attention, by making the male more visible against background light [
Because multiple forms of sexual selection can concurrently act on male color, there is the potential for interactions between these forces. Specifically, evolution of male color in response to female perceptual biases may alter the relationship between color and immune traits, thereby undermining the potential for females to choose based on male quality. For example, Wong et al. [
In most populations throughout sticklebacks’ circumpolar range, breeding males exhibit a red throat, a translucent or dark back, and iridescent blue eye. The iconic red throat has been the subject of extensive research, showing that females generally prefer red males [
Multiple mechanisms appear to underlie female sticklebacks’ preference for males with redder throats. Visual modeling of light spectrum reflectance, water transmission, and female visual reception shows that the red throat increases male visibility against background light in certain light environments [
There is also evidence that female stickleback prefer redder males because brighter red pigments indicate higher quality mates [
We are thus left with apparently contradictory conclusions that red coloration compromises some measures of immune function, but nevertheless indicates low parasite load and higher resistance. This contradiction may reflect the complex relationship between parasite resistance and the few available measures of stickleback cellular immune function. However it is also possible that contradictory results reflect genuine among-population differences in color-immune relationships [
We therefore hypothesized that stickleback populations inhabiting different signaling environments exhibit different color-immunity relationships. Specifically, we predicted that relationships between immune function and male color will differ between mostly red and mostly melanic populations. To test this prediction, we measured color-immunity relationships in each of four populations of stickleback from lakes in northern Vancouver Island, British Columbia. Because we used wild-caught animals to capture natural variation in color and immunity, these relationships are strictly correlational. However, it is exactly this natural correlation, with all its confounding variables and sources of noise, which would be most relevant to female mate choice decisions.
During the breeding season (May-June 2011), we collected ≈40 nesting male stickleback from each of four lakes on Vancouver Island (Gosling, Lower Stella, Blackwater, and Farewell Lakes). The light environment differs between these lakes, in terms of both total irradiance and the relative composition of short and long wavelengths (
Two males are shown per lake to illustrate the range of color within populations, as well as between-population differences. All males shown here were actively nesting. Note that the very melanic male from Gosling Lake (bottom right) is exceptional for that population, most individuals from that lake resemble the red-chinned individual. Photographs were taken under standardized lighting conditions.
A snorkeler (CDB) searched for nesting or territorial males along the shoreline of each lake (nests were 0.25–2.5 meters deep). The snorkeler observed each male for 10 minutes to confirm nesting status, presence/absence of eggs or fry, and male behavior (aggression, courting). The male was then caught by dipnet. Collections were conducted with permission of the British Columbia Ministry of Forest, Lands and Natural Resource Operations. All collections and methods used in this study were approved by the University of Texas at Austin Institutional Animal Care and Use Committee (IACUC; Protocol AUP-2011-00044). Specimens were collected from public land, and did not entail collection of protected species. Details of collection location and sample sizes are provided in
Following capture, males were immediately placed alone in a black mesh container at 1m depth in the lake, to minimize subsequent color changes that might be induced by male-male interactions or shifts in the light environment. Males were held in the dark until enough males were captured for the day, at which point we measured all males’ color and immune traits. Typically, males were held for between 1.5 to 3.5 hours. Each male was removed from the lake in turn, and immediately anaesthetized (within ~ 1 minute) and measured for color, then immediately euthanized and head kidneys removed for immune measurements described below. Immune measurements were carried out in a mobile immunology laboratory (a modified RV trailer) parked at the lakes to minimize travel time. Thus, we designed our study to minimize, as much as possible, handling time effects on both male color and immunity.
Males were lightly anaesthetized in MS-222, and their color (reflectance spectra) was measured in triplicate at four body locations (lower eye, lower jaw, preoperculum, and abdomen), using an EPP200C UV-VIS spectrophotometer, SL-4 Xenon lamp, and R400-7 reflectance probe at a 3-mm distance from the fish (Stellarnet, Tampa, FL). Spectralon white standard measurements were taken between fish to account for amp drift. For each measurement, we binned reflectance data into 10 nm intervals from 300 to 700 nm, then averaged reflectance within each bin and across triplicate readings, to yield mean reflectance for each wavelength bin. Summing reflectance across all bins quantifies overall brightness, whereas the proportion of brightness in each bin gives a measure of color. Here, we focus on brightness and the proportion of red reflectance (571 to 700 nm as a proportion of total reflectance). Spectrophotometric data from Farewell Lake was corrupted by hardware failure, so we analyze color data from only three lakes, but immune data from four lakes. To compensate for the loss of data from this one lake, we also measured color from photographs of all fish. Methodological details of photographic color data, and results of color-immune associations, are presented in
Here we give a brief overview of immune measurement methods, which are modified from [
Head kidneys were held in culture medium in individual 0.5 ml centrifuge tubes on ice for up to 2.25 hours while the remaining head kidneys were acquired from all males captured that day. Holding time had no significant effect on subsequent immune measurements. We then created a single-cell suspension of each sample (see
First, we ran a cell suspension (with propidium iodide [PI] to identify dead cells) through an Accuri C6 flow cytometer to coarsely classify and count live cells, collecting data on 50,000 particles. These particles include cellular debris (too small to be cells) and dead cells (PI+), which were subtracted from the dataset. Live cells were classified as either as lymphocytes or granulocytes based on their combination of cell size (forward scatter) and internal complexity (side scatter). We calculated the total number of live cells sampled, and the proportion of granulocytes and lymphocytes. Note that these cell types are coarse categories, each of which includes heterogeneous populations of cell types that cannot presently be distinguished in stickleback.
We incubated aliquots of the cell suspension with Dihydrorhodamine 123 (DHR), which fluoresces green when exposed to oxygen radicals. The intensity of green flourescence per cell reflects the magnitude of ROS production by either unstimulated cells, or in cells stimulated by Phorbol myristate acetate (PMA) [
Finally, we incubated an aliquot of cells overnight with or without fluorescent beads at 18°C for 18 hours, then used flow cytometry detection of bead fluorescence to count granulocytes and lymphocytes with and without beads. We used forward- and side-scatter gating to focus on the fraction of granulocytes with beads, as lymphocytes phagocytize at a lower (but non-zero) rate [
We tested whether populations differ in immune function or color. We focused on three metrics of immune function: the proportion of granulocytes, ROS production by granulocytes, and the proportion of phagocytic granulocytes at 18°C. We then compared these three immunological metrics to spectrophotometric measures of male brightness and hue on each of four body parts (eye, throat, abdomen, preoperculum). Because spectrophotometric color data were corrupted from one lake, we also compared immune traits to photographic measures of color. Because these photographic data yielded comparable results to the more precise spectrophotometric data, we focus on the latter here, and relegate the photographic data to
We used canonical correlation analyses (CCA) to test for multivariate correlations between male color (total brightness and relative red reflectance of each body part) and immune phenotypes (proportion of granulocytes, ROS production, and granulocyte phagocytosis rates). We used an asymptotic test of significance (R package CCP) to evaluate the null hypothesis that color and immunity are unrelated. We performed the CCA for each population separately. We also used CCA on all 4 lakes together, after mean-standardizing each trait within each population to avoid correlations due to between-population trait differences. The CCA loadings indicate the relative contribution of particular color and immune traits towards such correlations. Comparable CCA results for photographic color measures are detailed in
We then statistically tested the role of particular combinations of color and immune traits, in generating the observed multivariate canonical correlations. Within each lake, we used linear regression to evaluate the associations between single immune traits (dependent variables: proportion of granulocytes, or ROS production, or phagocytosis rates), and single color traits (brightness or relative red reflectance of each of the four body parts). Because the CCA analyses demonstrated that immune-color correlations do exist, we do not apply P-value adjustments for multiple comparisons. Instead, we treat these numerous regressions as a way to identify the univariate trait combinations driving already-confirmed canonical correlations, rather than
A key limitation of CCA is that it is not effective at detecting group-specific (e.g., lake-specific) canonical axes. Therefore, to test whether color-immune relationships are parallel across lakes, or differ among lakes, we used all samples in an ANCOVA testing whether a given immune trait depends on lake, a main effect of color (indicating a shared color-immunity association), or a lake by color interaction (indicating lake-specific color-immune relationships). Again, because our CCA analyses had already identified significant color-immune relationships in some lakes but not others (see
Color-immunity relationships could be indirect results of their joint correlation with some other covariate. To account for such potentially spurious color-immune associations, we considered many possible confounding variables (see
We found significant among-population variation in all measured immunological traits (MANOVA, P < 0.0001), including the relative abundance of cell types, ROS intensity, and phagocytosis rates of granulocytes (
As another measure of repeatability and among-individual variation, we calculated the correlation between duplicate measures, mean-standardizing by lake to remove effects of among-lake variation and randomly picking two of the triplicate measures per fish. The correlations between duplicate measures of cell counts, proportion of granulocytes, ROS production, and phagocytosis rates were r = 0.95, 0.74, 0.83, and 0.90 respectively (all P < 0.0001). Among-individual variation in immune function is not driven simply by body size variation. With the exception of total cell count (which depends on the size of head kidneys used to generate cell cultures), none of the other immune measures were significantly correlated with fish mass (
Male nuptial color varies significantly both within and between populations (
Within populations, immune function is correlated with male color. We observed significant canonical correlations between color (8 variables) and immunity (3 variables) within each of the two red-male lakes (Gosling and Lower Stella, both P < 0.004) but not the one melanic population for which we had spectrophotometric data (Blackwater, P = 0.424). These Gosling and Lower Stella canonical correlations remain significant after Bonferroni correction for multiple comparisons. The canonical axis loadings indicate which traits drive these multivariate correlations (
All 4 lakes | Blackwater | Gosling | Lower Stella | |
---|---|---|---|---|
r | 0.540 | 0.484 | 0.745 | 0.670 |
P | 0.4240 | |||
Brightness | ||||
Abdomen | 0 | 0 | 0 | 0 |
Lower eye | 0 | 0 | 0 | 0 |
Preoperculum | 0 | 0 | 0 | 0 |
Throat | 0 | 0 | 0 | 0 |
Proportion red reflectance | ||||
Abdomen | -1.66 | 1.14 | -2.00 | 1.93 |
Lower eye | -0.62 | 1.56 | -1.63 | 1.52 |
Preoperculum | 0.93 | -2.93 | 1.23 | -0.42 |
Throat | 0.46 | -4.73 | 0.16 | -2.31 |
% granulocytes | -0.65 | 1.22 | -0.58 | 1.48 |
Median ROS production | 0.028 | 0.24 | -0.01 | -0.07 |
% phagocytic granulocytes | -0.469 | -0.03 | -0.73 | -0.471 |
CCA was run first for all four populations combined, after first mean-standardizing each trait within each population, to avoid correlations arising from between-population differences. CCA was then run separately for each lake (except Farewell, due to data corruption). For each CCA, we present the correlation coefficient (r), and an asymptotic approximation to an F statistic and P-value, obtained via the p.asym command in the R package CCP. We also present the loadings of each trait on the canonical x axis (for color) and y axis (for immune traits).
Although hue was significantly associated with granulocyte number and activity in both lakes, the particular effects differed between populations. In Gosling Lake, both phagocytosis and (to a lesser extent) granulocyte abundance tended to be higher in males with redder abdomens and eyes, and less red preopercula (
We used linear models to investigate particular combinations of color and immune traits, and to test for lake-specific effects (e.g., immunity depending on a lake*color interaction). The results of these linear models are summarized in
Consistent with the CCA loadings, linear regression confirmed that males with redder throats have fewer granulocytes (
While the trend is only significant within Lower Stella (r = 0.353, P = 0.009), the slope is very similar in the other two lakes (Blackwater, r = -0.181; Gosling, r = -0.185) leading to an overall effect of color (P = 0.002) and no color*lake interaction. Symbols and colors denote populations (see legend). A separate trendline is drawn for each population, using a thick solid line to denote a significant correlation, or thin dashed line when non-significant.
Consistent with the CCA results, ROS production showed few associations with male color. Across all lakes, ROS tends to be higher in fish with bluer eyes (color P = 0.017, no color*lake interaction), although within individual lakes this trend is only significant in Blackwater Lake (P = 0.015). This regression includes, as a covariate, infection intensity by the parasitic copepod
Granulocyte phagocytosis activity was higher in males with redder abdomens (
The trend is significant within Gosling Lake (r = 0.438) and in a similar direction in Lower Stella Lake (r = 0.327) but not Blackwater Lake (r = -0.009) leading to a marginally significant lake*color interaction. Symbols and colors denote populations, as in
Although the CCA did not give any weight to male brightness, univariate linear regressions suggest that there are associations between brightness of individual body parts and phagocytosis rates. For all four body parts examined, brighter males tend to have lower phagocytosis rates by granulocytes (
(A) For abdomen brightness r = -0.426, -0.347, -0.386 respectively for Blackwater, Gosling, and Lower Stella Lakes. (B) For preoperculum brightness, r = -0.231, 0.129, and -0.304 respectively. Symbols and colors denote populations, as in
The linear model results (supporting an effect of male brightness) disagreed with the CCA, which gave brightness negligible weight. This could occur if hue has a relatively stronger effect, so that brightness is relegated to higher-order x axes. We therefore repeated CCAs using only brightness of the four body parts, excluding male hue data. These new CCAs confirmed a general effect of brightness (across all three lakes, r = 0.421, P = 0.035). In lake-specific CCAs, brightness was significantly associated with immunity in Gosling (P = 0.028), Lower Stella (P = 0.021), but not Blackwater (P = 0.467). In each lake, and in the combined CCA, phagocytosis rates were by far the most strongly loaded immune trait driving these multivariate correlations.
In an observational correlative study such as this, correlations between traits could arise from those traits’ joint association with some other variable.
Prior research has clearly established that populations inhabiting different light environments are subject to divergent sexual selection on color signals [
Consistent with prior studies of male color and condition in stickleback [
Controlling for brightness, hue is also associated with immunity. In fact, canonical correlation analyses indicates that hue is the primary covariate of immune traits. Overall, we find that males with redder throats have proportionally fewer granulocytes, suggesting reduced capacity for innate immune responses by primary phagocytic cells. This is consistent with the previously demonstrated trade-offs between innate immune function and male red coloration [
The color-immune associations noted above involve multipe areas of the males’ bodies and different immune functions (e.g., red throat ~ granulocyte frequency; red abdomen ~ phagocytosis rates). This highlights an often-overlooked point that color-immune relationships entail highly multivariate association. Different body parts can be associated with different aspects of male immune function. Although there is evidence that stickleback females do use male color to evaluate immunity or parasite load [
Our over-arching question here was whether male color-immune associations are consistent across populations that exhibit very different male colors and light environments. For some pairs of traits (abdomen brightness and phagocytosis rate), the color-immunity relationship appears to be fairly consistent across populations (e.g., no lake*color interactions) despite very different male colors. For other combinations of traits (preoperculum brightness and phagocytosis; red abdomens and phagocytosis, red surface area and proportion of granulocytes), we find support for lake*color interactions that would imply lake-specific color-immunity relationships. The most dramatic example of lake-specific color-immune correlations came from our photographic measures of color (
Interestingly, although skin pigments convey no detectable information about immunity in Blackwater Lake, uniquely in this lake bluer eyes imply greater ROS production. This raises the intriguing question of whether, in melanic populations such as Blackwater, the loss of typical associations between immunity and red throat coloration leads to the evolution of female choice based on alternative characters, such as eye color, that may provide additional cues about male immune traits.
For evolution to alter color-immune relationships across lakes, both types of traits must be heritable. We have strong reason to believe this requirement is valid. First, several other studies have demonstrated that there is heritable variation in stickleback immune function [
A second caveat is that the results presented here are entirely correlative, as it is not feasible at present to experimentally manipulate color (or immunity) in wild-caught stickleback. The goal here is to document, as much as possible, color-immune associations as they exist in the wild where they will be relevant to female mate choice decisions. It remains to be seen whether such associations persist in laboratory settings where they might be studied experimentally without the many confounding (but perhaps mechanistically important) variables such as diet, parasite loads, and social stresses. A corollary is that we cannot determine, with certainty, the direction of causation of associations presented here. We consider it likely that the link between color and immunity is mediated by variation in some unmeasured trait(s) such as hormone levels [
To conclude, we have shown that male color is correlated with various aspects of immune function in wild threspine stickleback. Some combinations of color-immune traits (e.g., granulocyte abundance versus red throat color) are correlated in a comparable manner in multiple populations. The repeatability of this trend, despite very divergent mean male colors, suggests that there may be some relatively immutable trade-off between these color and immune traits consistent with prior studies [
Other color-immune combinations are inconsistently correlated across lakes. Melanic stickleback populations appear to have lost or mitigated the color-immune correlations that we observed in stereotypical red-male populations. Thus, consistent with our initial expectations, populations in different light environments, which are know to evolve different color patterns [
(DOCX)
(DOCX)
(DOCX)
(DOCX)
(DOCX)
(DOCX)
(DOCX)
This includes 3 pages of text, and Figs A-C.
(DOCX)
This includes 6 pages of text, Figs A—N, and Table A.
(DOCX)
This includes 9 pages of text, and Figs A—K, and Tables A-C.
(DOCX)
(DOCX)
(DOCX)
(DOCX)
We thank Natalie Steinel, Jesse Weber, and anonymous reviewers for feedback on the manuscript. Jacob Heiling, Gina Calabrese, Jay Falk, and Kim Ballare provided valuable help doing flow cytometry in the field. Mahmoud Irannezhad assisted with phenotypic measurements of specimens.