Structural colours reflect individual quality: a meta-analysis

1. Introduction

Colour is a ubiquitous channel of communication in nature and is showcased at an extreme in the service of mate choice [1,2]. A central hypothesis in evolutionary biology is that sexual selection has driven the elaboration of colourful ornaments into reliable indicators of individual quality [3], with empirical tests guided by indicator and handicap models [4,5]. These models argue that conspicuous displays are selectively favoured because their production is differentially costly (handicap) and/or constrained (index) between individuals of varying quality, and so encode honest information to potential mates. A prediction common to honesty-based models is that signals should show heightened condition-dependent expression, and the most robust support to date among ornamental coloration is found in carotenoid-based signals [6,7]. As pigments that cannot be synthesized de novo, all carotenoids must ultimately be acquired via diet before being incorporated into signals directly or following bioconversion. This offers ample opportunity for selection to favour mechanistic links between foraging, metabolic performance and sexual signal expression (that is, the combined perceptual features of hue, saturation and brightness), which is now well established, at least among birds [8,9]. Relative to our knowledge of pigment-based coloration, however, the potential for structural colours to signal individual quality remains both understudied and poorly resolved.

Unlike pigments, which are selectively absorbent, structural colours result from the selective reflectance of light by nano-structured tissues [10,11]. Accumulating evidence suggests that the development of these structures is driven by self-assembly—such as the phase separation of keratin and cytoplasm in nascent feather barbs [12–14]—rather than the active (and ‘expensive’) cellular processes that underlie some pigmentary colour production [8]. Three general arguments have been articulated around the potential for honesty among structural colours. One is that if sufficient material is required to produce nano-architectures, then it will establish a trade-off with other physiological needs that may be differentially met among individuals of varying ‘quality’, as consistent with a handicap explanation [4,6]. A non-exclusive alternative builds on the observation that features of the macro-scale expression of signals rely on the precision with which the underlying structures are organized [15,16]. If individuals heritably vary in their capacity to achieve such organization as a result of, for example, physiological constraints on signal production, or the behavioural acquisition of stable developmental environments, then signals may serve as an index of underlying genetic quality [5]. Finally, the lack of obvious ecologically relevant material to trade-off against during signal construction, together with the self-assembly inherent in structural colours noted above, has motivated arguments against any expectations for condition-dependence sensu lato [12]. Though experimental work is able to partition these hypotheses in some contexts [17], most empirical studies to date have focused on the overarching question of honesty by examining the predicted covariance between fitness-related traits and signal expression. This has provided valuable insight into the central question, but diversity in signal designs, measures of ‘quality’, and taxonomy have presented a challenge for qualitative synthesis. Modern quantitative methods, however, provide robust tools for controlling for and capitalizing on such variation (e.g. via mixed-effect models and meta-regression), and so can offer substantive answers to longstanding questions [18].

Here, I used phylogenetically controlled meta-analysis and meta-regression to examine whether structural colour signals encode salient information on individual quality. Specifically, I synthesized estimates of correlations between measures of individual quality and signal expression to test the prediction of condition-dependence, before examining methodological and theoretically derived mediators of effect-size variation among studies.

2. Methods

3. Results

The final dataset comprised 186 effect sizes, across 28 species, from 41 studies [6,17,31–69]. Of those 186 effects, 117 were drawn from birds, 22 from insects, 28 from reptiles, and 11 from arachnids (electronic supplementary material, table S1 and figure S2). As predicted, I found a positive overall correlation (i.e. meta-analytic mean effect) between individual quality and structural colour signal expression (Z = 0.1573, 95% CI = 0.084–0.231; figure 1 and table 1). The heterogeneity of effect sizes—that is, the variation in effect-size estimates after accounting for sampling error—was high (I² = 80.42%, 95% CI = 77.26–83.01), as is typical of meta-analytic data in ecology and evolutionary biology [70]. A small amount of heterogeneity was explained by among-study effects (I² = 14.21%, 95% CI = 8.97–20.20), and only a very weak phylogenetic signal was evident (I² = 2.17%, 95% CI = 1.18–3.47).

• Download figure
• Open in new tab
• Download PowerPoint

Of the measures of quality considered, body condition and immune function were reliably positively correlated with structural colour expression, while age and parasite resistance were not (table 1 for all corresponding numerical results henceforth). This varied slightly across taxa, however, with a robust effect of condition on coloration apparent across all groups, while the effects of age and parasite resistance were apparent among insects and birds, respectively (electronic supplementary material, table S2). Both the hue and brightness of signals were similarly informative channels on the whole, while chromaticity was not consistently associated with individual quality (though this varied by taxa; electronic supplementary material, table S2) nor was any correlation apparent when composite measures of coloration were used. Iridescent signals were subject to slightly stronger positive correlations than non-iridescent signals across all measures of quality. Signal honesty was apparent among males only though the weak, borderline effect and much smaller sample among females (n = 29/186 versus 146/186 for males) suggests a male bias in the literature similar to that in related fields [71], which may have partly driven this outcome. Experimental studies tended to report marginally stronger correlations than observational assays, which most likely reflects slightly exaggerated experimental manipulations of condition relative to natural variation [29]. Finally, the majority of studies (n = 36) did not include the measurements of non-sexual control traits in tests of heightened condition-dependence, though I found no clear difference in effect-size estimates between those that did and did not.

4. Discussion

Colourful ornaments may be reliable conduits of information on mate quality, though evidence for the predicted covariance between signal expression and mate quality among structural, as opposed to pigmentary, signals is equivocal. Here, I found meta-analytic support for this link in the form of a positive correlation between structural colour expression and individual quality (figure 1), consistent with honesty-based models of sexual signal evolution [4,5]. The strength of the overall correlation, though moderate [72], was commensurate with meta-analytic estimates from pigment-based sexual signals [8,9,73], and suggests that structural coloration may similarly serve a reliable indicator of individual quality.

Quality is a multivariate feature of individuals, and this is reflected in the effect-size variation between measures. Both condition (as narrowly defined above), and proxy measures of immune system integrity, were on average positively correlated with signal expression across all taxa in which those relationships have been examined. This is consistent with experimental work showing that body mass and immune function are responsive to ecologically salient stressors, with consequences for colour production. Among birds, for example, disease and dietary stress produce abnormalities in the keratin barbules that contribute to coloration [16,74,75], while in butterflies, the organization of wing-scale architectures is disrupted by nutritional and environmental stress during pupal (hence, wing-structure) development [7,56]. By contrast, neither age nor parasite resistance was consistently signalled, though this varied slightly across taxa. These latter measures are often predicated on, or susceptible to, the mechanical degradation of structures post-development. Thus, the inherently heightened variability of sexual signals combined with parasite-induced damage (ectoparasite, in particular) and/or accumulated wear with age, combined with varied mechanisms of colour production across animal classes, may compound to render the signals less accurate predictors on balance [57,76,77]. Curiously, the near inverse relationship was recently identified in a meta-analysis of carotenoid-based signalling. Weaver et al. [8] examined correlations across similar categories of quality as those used here but found no consistent relationship between signals and either of body condition or immune function. Given the fundamental optical and developmental differences between structural and pigmentary colour production (described above), the potential exists for each to signal unique aspects of individual mate quality, as is suggested by the totality of this work [63].

Colour is often assumed to be the central conduit of information exchange given its relative stability under variable natural illumination [78,79], though my results suggest both the colour and brightness of signals are similarly informative (figure 1; electronic supplementary material, table S2). Furthermore, I identified slightly stronger condition-dependence among iridescent, as opposed to non-iridescent, patches. While the underlying architecture varies across taxa, all iridescent coloration arises from coherent light interference and so may demand a level of architectural organization beyond that of incoherent scattering [11,26], as well as non-iridescent coherent scattering (though evidence for the latter possibility is limited; [14]). Iridescence also introduces temporal structure to signals since the colour appearance depends on the precise arrangement of signals, viewers and light sources. These combined features may render iridescent colours particularly suitable as bearers of information [28] and so contribute to the ubiquity of the phenomenon [80,81]. Though only indirectly considered here, as few studies quantify between-individual variation in iridescence itself, this idea has found more immediate support via condition-dependent variation in signal angularity [17], and a predictive relationship between iridescence and mating success [27,82]. These results affirm the view that the extended spectral and temporal repertoire available to structural colours may facilitate the exploration of distinct ‘signalling niches’, with tangible evolutionary consequences [1,27,52].

By integrating the development of signal structure and fitness-related traits, structural colours may serve as informative signals during mate choice. A holistic understanding, however, awaits progress on several fronts. Most significant is the inclusion of appropriate non-sexual controls since the ultimate evidence for handicap models lies in the demonstration of heightened condition-dependence among sexual traits [29]. Partitioning indicator and handicap models of signal evolution and understanding the nature of direct and/or indirect benefits being signalled are key challenges [17]. Understanding the development of structural colours during ontogeny is also a central front for progress [12–14], and studies among invertebrates (which offer benefits in terms of tractability) would be invaluable in complementing the excellent work accumulating on birds. Finally, signalling ecology should remain front-of-mind given the inherent spatio-temporal complexity of signals and visual systems [83–85]. This offers exciting opportunities for integrative studies of signal development, production and perception, which will fuel a richer view of this pervasive adornment of the natural world.

Data accessibility

All data and code are available via GitHub (https://github.com/EaSElab-18/ms_metacol) and are persistently archived through Zenodo (https://dx.doi.org/10.5281/zenodo.3718617).

Competing interests

I declare I have no competing interests.