Evolution - Animal Visual Systems and the Evolution of Color Patterns: Sensory Processing Illuminates Signal Evolution lyrics

Abstract. Animal color pattern phenotypes evolve rapidly. What influences their evolution? Because color patterns are used in communication, selection for signal efficacy, relative to the intended receiver's visual system, may explain and predict the direction of evolution. We investigated this in bowerbirds, whose color patterns consist of plumage, bower structure, and ornaments and whose visual displays are presented under predictable visual conditions. We used data on avian vision, environmental conditions, color pattern properties, and an estimate of the bowerbird phylogeny to test hypotheses about evolutionary effects of visual processing. Different components of the color pattern evolve differently. Plumage s**ual dimorphism increased and then decreased, while overall (plumage plus bower) visual contrast increased. The use of bowers allows relative crypsis of the bird but increased efficacy of the signal as a whole. Ornaments do not elaborate existing plumage features but instead are innovations (new color schemes) that increase signal efficacy. Isolation between species could be facilitated by plumage but not ornaments, because we observed character displacement only in plumage. Bowerbird color pattern evolution is at least partially predictable from the function of the visual system and from knowledge of different functions of different components of the color patterns. This provides clues to how more constrained visual signaling systems may evolve. Mate choices often depend upon coloration and strongly affect both population divergence and speciation (Andersson 1994; Coyne and Orr 2004; Gavrilets 2004). In the rush to describe the genetic basis and evolutionary history of reproductive traits (e.g., Coyne and Orr 2004) and their possible fitness-affecting signal content (Andersson 1994; Hill 2002), insufficient attention has been paid to the evolutionary implications of how s**ual signals are processed by the sensory system (Endler 1992, 1993a,b, 2000; Boughman 2002). This could seriously undermine our ability to understand evolution because what we sense and measure may have little to do with animal neural processing and because the genotype-phenotype relationship is weaker. Ordinary traits consist of a genotype (G) responsible for the trait's construction and a phenotype (P) that interacts with the environment. Color patterns have more components; they include the signaler's color pattern genotype (G1), the signaler's phenotype that is the mosaic of differentially reflective surfaces (PI), and a second phenotype that is the corresponding set of stimuli in the sensory system of the receiver (P2). A second genotype (G2) in the receiver is responsible for constructing the receiver's signal processing and decision-making phenotype (P3). Color pattern evolution is driven by the P2-P3 interaction, which indirectly drives the evolution of G1 and G2. The commonly measured PI is not very relevant to evolution unless it serves other functions such as thermoregulation. The multiple-link causal chain between G1 and P2 may explain their loose evolutionary relationships. Color patterns (measured as G1 or P1) are very labile evolutionarily, often evolving more rapidly than and with different geometry than traits normally used to construct phylogenies, and show frequent reversals (examples in Prum 1997; Höglund and Shorey 2004). This lability is so high that the same colors in different species may be a result of phenotypic convergence among different genes (examples in Allen and Omland 2003; Hoekstra and Nachman 2003; Nachman et al. 2003; Mundy et al. 2004; Ugalde et al. 2004), and a given color may be controlled by two independent genetic systems within the same species (examples in Grether et al. 2001; Ugalde et al. 2004). This suggests that genetic details are not very important, natural selection can work quickly on coloration (as in Endler 1980, 1986), and colors can evolve rapidly as species shift between habitats or respond to environmental change. This proposal is supported by strong correlations between coloration and habitat types (examples in Lythgoe 1979; McNaught and Owens 2002; Heindl and Wickler 2003; Nachman et al. 2003; Gomez and Thery 2004) and coloration convergence in equivalent habitats even when their underlying genetics are different (as in Hoekstra and Nachman 2003). We need to understand the function of P2 and P3 to understand and predict their evolution and to make the link with reconstructed evolution in G1 and G2. Because it is GI and mostly P1 rather than P2 that have been studied, and our own perception of P1 is different than most animals (Lythgoe 1979; Endler 1990; Bennett et al. 1994), the evolution of color may appear less chaotic if we examine P2 rather than P1 or G1. In this paper we will examine visual processing effects on color pattern function (P2 and P3) and explore ways in which this can influence signal evolution. We will address three basic questions: Does signaling coloration show significant within-pattern contrast and significant contrast with the visual background (Endler 1978, 1990, 1993)? How can color patterns change while being sensed by the visual system? Are changes in coloration elaborations or innovations? FIG. 1. Distinction between elaboration and three kinds of color pattern innovation. The axes are those of any color space (such as Figs. 5, 6). The open circles are a set of colors of the starting relationship among the colors and they fall on a line (first principal component) that defines the relationship. In human terms the relationship may be regarded as a color scheme. Elaboration: if new color pattern components fall on those line as the old ones (same relationship among the components); innovation: if new components fall off the line; random innovation: new components fall off the line in random directions; specific innovation: new components fall off the line in only one direction (or a few); efficiency innovation: new components fall off the line in a way that avoids the visual background, increasing the signal/noise ratio or signal-background contrast. This conceptualization could apply to other sensory modes. Elaboration and innovation may occur during evolution. These terms have been used loosely and inconsistently. Here we will restrict the meanings and make them quantitative (Fig. 1). A color pattern consists of a mosaic of n colored patches (Endler 1978) resulting in n stimuli on the viewer's retina. We can estimate the quantitative relationships among the patch stimuli; different kinds of color patterns will show different relationships. In the vernacular, a set of patches with a particular stimulus relationship is a color scheme. We define "'elaboration" as a change in a color pattern that is not a**ociated with a change in the relationships among the patch stimuli; for example, a light blue and pink color pattern could change into a blue and red pattern, or the sizes could change but the proportions remain constant. We define "innovation" as a change in a color pattern that is a**ociated with a change in the relationships among the stimuli; for example, changes in hue or hue proportions. Innovations can be random, specific, and may result in noise reduction (Fig. 1). These definitions allow explicit statistical tests for elaboration and innovation in any system. We examine sensed color pattern evolution using Australian bowerbirds (Pizzey and Knight 1997). Bowerbirds (Ptilinorhynchidae) provide an unusual opportunity to investigate how signals can evolve because their colors include both plumage and a constructed bower (illustrations and descriptions in Frith and Frith 2004), freeing them from many of the genetic and evolutionary constraints of relying solely on plumage. Their colors function in s**ual selection and may also serve in species recognition, given that some but not all species live in sympatry (Frith and Frith 2004). Males make bowers to attract females (Marshall 1954; Gilliard 1969; Borgia 1997; Frith and Frith 2004). Bowers consist of ornaments (colored objects) placed upon a construction of sticks or a cleared area. The structure and function of bowers in mating success is well established (Borgia 1985, 1995, 1997; Borgia et al. 1987; Uy and Borgia 2000; Madden 2002, 2003a,b; Madden and Tanner 2003; Coleman et al. 2004). Marshall (1954) suggested that ornaments are elaborations of male or female plumage (the elaboration hypothesis). His suggestion and objections to it (Borgia et al. 1987) are based upon human visual comparisons, but human vision is very different from that of birds (Bennett et al. 1994; Hart 2001a,b), so it has not yet been adequately tested. Gilliard (1956, 1969) suggested that there has been a transfer of s**ual selection function from plumage to bower structure (the transfer hypothesis). He suggested that once the attention of females was centered on bowers, bright plumage might be a disadvantage under predation, eventually leading to brighter bowers and duller males. Phylogenetic work supports this in one branch of the bowerbird lade (avenue builders), but not the other (Kusmierski et al. 1997). However, the quantitative relationships among the visual signal components in plumage and bower have not been examined, nor have they been examined in terms of bird vision. We will consider these questions: What is the advantage of relying on an extended phenotype (the bower) for part of the signal? Do ornaments increase signal efficacy? Does separation of part of the signal from the body affect individual crypsis? Are ornaments an elaboration of the plumage? Does bower building result in a change in the relationships among the visual signal components on and off the plumage (an innovation)? Are these changes related to the visual backgrounds? Which parts of the pattern could be involved in species recognition? How do these functions evolve? The high evolutionary lability of bower components (Kusmierski et al. 1997; Uy and Borgia 2000) and current knowledge of vision (Kelber et al. 2003) makes addressing these questions practical and may provide clues as to how divergence occurs in species whose signaling systems are more constrained. MATERIALS AND METHODS Species Studied and Their Phylogenetic Context We collected data from plumage, bowers, and visual back-grounds of great (Chlamydera nuchalis), spotted (C. macu-lata), fawn-breasted (C. cerviniventris), satin (Ptilonorhyn-chus violaceous), regent (Sericulus chrysocephalus), toothbill (Scenopoeetes dentirostris) and golden (Prionodura newtoniana) bowerbirds, and from the non-bower-building spotted catbird (Ailuroedus melanotis; called black-eared catbird by Frith and Frith 2004) in eastern Queensland, Australia. We will refer to these species by two-letter codes for brevity: GR, SP, FB, ST, RG, TB, GO, and SC, respectively. We sampled GR and SP in Eucalyptus woodland, and the others in rainforest; typical habitats for each species (Pizzey and Knight 1997; Frith and Frith 2004). Localities are Townsville (GR), Taunton National Park-Scientific (SP), Iron Range National Park (FB), Atherton Tablelands (Wooroonoran National Park, Mt. Edith, Mt. Baldy; each with ST, TB, GO, SC), Paluma Range National Park (ST, TB, GO, SC), Lamington National Park (ST, RG), and Bunya Mountains (ST). FIG. 2. Phylogeny of most the bowerbirds using cytochrome b sequences extracted from GenBank (data from Kusmierski, Cristides, and Cracraft), aligned with CLUSTALW, neighbor-joining and bootstrap an*lysis using MEGA2. Numbers at the nodes are bootstrap values. Vertical bars indicate the loss (-) or acquisition (+) of s**ual dimorphism in plumage (plumage data from Frith and Frith 2004). Species codes and names (top to bottom): FB (fawn-breasted, Chlamydera cerviniventris), GR (great, C. nuchalis), la (Lauterbach's, C. lauterbachii), SP (spotted, C. maculata), ST (satin, Ptilonorhynchus violaceus, two samples), RG (regent, sericulus chrysocephalus), TB (toothbilled, Scenopoeetes dentirostris), sr (streaked, Amblyornis subalaris), GO (golden, Prionodura newtoniana), mg (MacGregor' s, Amblyornis macgregoriae), ar (Arch-bold's, Archboldia papuensis), vf (Vogelkop, Amblyornis inornatus from Fak Fak), va (Vogelkop from Arfak), gc (green catbird, Ailuroedus cra**irostris), SC (spotted catbird, A. melanotis), ly (lyrebird, Menura novaehollandiae, two samples). Species codes in lowercase are not in this study. The location of TB changes (dashed lines) with the a**umptions used in the phylogenetic an*lysis and its linkage to other species shows low bootstrap values for all model parameter combinations. We will just a**ume that TB branches out basally with respect to either the avenue or maypole clades. We use the current bowerbird phylogeny. Kusmierski et al. (1997) published a cytochrome b phylogeny, but some additional data by Cristides and Cracraft are available in GenBank. We used this and Kusmierski's data in GenBank to construct phylogenies using CLUSTALW and MEGA2, using various a**umptions about mitochondrial DNA (mtDNA) evolution. Figure 2 shows a typical result. Species codes (top to bottom in Fig. 2) and GenBank accession numbers are: FB (U76504), GR (U10372), la (U76506), SP (U10364), ST (X74256 [Cracraft] and U 10367, respectively), RG (U10365), TB (U101 13), sr (U76505), GO (U10370), mg (X60940), ar (U76503), vf (U76508), va (U76507), gc (U10371), SC (X74257 [Cracraft]), ly (U76509 and U5850 [Cristides], respectively). The data are Kusmierski's, except where noted. Lowercase species codes are for species we did not measure. Both molecular (Kusmierski et al. 1997; Fig. 2) and morphology-based phylogenies indicate two main clades, the avenue-builders and the maypole-builders, named for bower shapes (figures in Frith and Frith 2004). Figure 2 and all other phylogenies we constructed with different evolutionary a**umptions are very similar to that of Kusmierski et al. (1997) except for the location of TB. Kusmierski et al. (1997) had TB branching out basal to the maypole rather than the avenue clade. However, the location of TB changes (Fig. 2, dashed lines) with the a**umptions used: TB clusters sometimes with the avenue builders, sometimes with the maypole builders (as in Kusmierski et al. 1997), and sometimes as part of a trichotomy, with bootstrap values of 40-60%. The evolutionary pattern of plumage s**ual monomorphism and dimorphism (data from Frith and Frith 2004) mapped on the mtDNA phylogeny (Fig. 2) suggests that TB could be part of a trichotomy. We simply regard it as more basal to the other species. The low TB bootstrap values contrast with the high ones in the rest of the phylogeny, which tends to be stable under almost all combinations of a**umptions. We sampled all Australian species except the western bowerbird and the green catbird (gc), the remaining unsampled species are in New Guinea. We define the degree of derivation of a species by the number of nodes (speciation events) between that species and the base node (common ancestor to the Ptilonorhynchidae). Less derived species are fewer nodes from the base than more derived species. We divided the species into four groups by degree of derivation and the switches to and from s**ual dimorphism (Fig. 2): the non-bower-building catbird (SC); the least derived bower-builder TB; the intermediately derived GO, RG, and ST; and the more derived FB, GR and ST. Since this is a crude division, it should be qualitatively unaffected by the species missing from the phylogeny. Species missing from both the phylogeny and our study are Chlamydera guttata (should cluster with SP; formerly a SP subspecies), two Sericulus species (S. aureus and S. ardens, should cluster with RG), Amblyornis flavifrons (on morphological and behavioral grounds in Frith and Frith [2004] it should cluster with sr, GO, or mg), and Ailuroedus buccoides (should cluster with gc or SC). General Remarks on Data Collection and an*lysis In bowerbirds the visual signal is a mosaic that includes colored patches in the plumage, the bower structure, and the bower ornaments. Catbirds (SC, gc) do not build bowers, so their color patterns consist only of the body colors. For brevity we will use the term "plumage" in the sense of feathers, legs, bill, gape, and iris rather than just feathers. During displays, some species actively display their nuchal area, bill, gape (lining of inside of mouth), iris, and legs; we will refer to these as the display or D components. The visual background is also a mosaic consisting of elements in the leaf litter; bowers are placed on the ground. We sampled both signal and background patches. The perception of color depends upon differences in photon (light) capture rates among photoreceptor types that differ in the wavelengths that they can capture (Lythgoe and Partridge 1989; Kelber et al. 2003; Endler and Mielke 2005). In birds, a given patch in a visual signal or visual background results in a set of four photon capture rates by four kinds of cones. This depends upon the ambient light (irradiance) spectrum illuminating the color pattern under the viewing conditions, the reflectance spectrum of the patch, the transmission spectrum of the air between the patch and the viewer during viewing conditions, the viewer's ocular media transmission spectrum (including oil droplets in birds), and the absorption spectrum of the photoreceptors (reviews in Endler 1990; Endler and Mielke 2005). We gathered data on all of these components. We calculated photon capture stimuli for each cone cla** for each patch and corrected for light adaptation by natural light. We converted these to relative stimuli; color perception is derived from relative rather than absolute stimuli. We used a new nonparametric version of compositional an*lysis allowing comparison of entire color patterns. We calculated receptor-noise ellipsoids and mapped them on the distributions as estimates of the scale of how different colors must be in order to be discriminated. All calculations were done by programming on a PC using PASCAL (eye models), FORTRAN (LSED-MRPP statistical tests), and MATLAB 7 (other an*lysis). For detailed review and justification of our methods see Endler (1990, 1993a) and Endler and Mielke (2005). Critical details and a summary of the an*lysis are provided in the remainder of this section. Data Collection Irradiance measures (Endler 1990) were taken on the bower courts where displays and ornaments occur under both clear and cloudy conditions (as in Endler 1993a). Plumage reflectance data (Endler 1990; Endler and Thèry 1996) were collected from live birds trapped or netted (and quickly released) in the study sites: Two to 15 birds per species depending upon s**ual monomorphism or dimorphism. We also took data from skins in the CSIRO Australian National Wildlife Collection; except for spotted bowerbirds, we found no significant difference between skins and live birds (as in Endler and Thèry 1996). Reflectances were taken at 15 different locations on each bird, obtaining a representative sample of all of their differently colored patches. Reflectance data were collected from 15-29 bowers of all species except fawnbreasted and regent bowerbirds (five and four, respectively). Reflectances of all bower ornaments were taken except for very abundant ornaments (e.g., lichen for the golden, snail shells/bones for great and spotted), which were scanned in multiple transects across the avenue and court to estimate their relative areas. Reflectances of bower structure (avenue, platform, court, or stick tower) and visual backgrounds within 1 m of the bower were taken as repeated 1 -m transects with samples every 2 cm (bower twig structure) or whenever color changed along the transect (others). Random transects of visual backgrounds were also taken more than 10 m from the closest bower at random locations in the same habitat and microhabitats. We refer to these as "near" and "far" backgrounds, respectively. Several hundred scans were made per bower, and several thousand scans were made on visual backgrounds. Sample sizes are shown in Table 1. Reflectance and irradiance spectra were collected in the 300 to 700-nm (bird-visible) wavelength range with an Ocean Optics (Dunedin, FL) S2000 portable spectrometer with either (irradiance) a Li-Cor (Lincoln, NE) LI-1800-11 cosine-corrected sensor, or (reflectance) a specially made shielded radiance attachment and an Ocean Optics PX- 1 synchronized Xe flash lamp. Optic coupling between the sensor, spectrometer and light source was through quartz fiber optics to ensure efficient sensing of UV as well as visible light. For reflectance the incident and sensor angles were 45° to prevent glare (spec-ular reflection) from the surface and shielded from incident light (see Endler 1990). The irradiance sensor was calibrated each sampling session with a Li-Cor LI-1800-02 optical radiation calibrator, and the reflectance sensor was calibrated for every bower or bird with a Spectralon (Ocean Optics) reflectance standard (Endler 1990). The position and rotation of the fiber optics was set with reference marks on the couplings to make the configuration identical to that during calibration when the fibers were detached between calibration and measurement. Bowerbirds communicate visually over less than 100 m and s**ual displays occur within a 1 m, so we set the air transmission spectrum to unity for all wave-lengths. Photon Capture We calculated the relative photon (light) capture for each patch of each color pattern by each photoreceptor cla**. This method has been used to make successful and detailed tests of predictions about spectral sensitivity and the results of color-based choices in a variety of organisms, including birds. Because the methods we used are described in detail elsewhere (Endler and Mielke 2005), we will give only a brief description and the parameters we used. Reviews, dis-cussions, and justification can be found in Kelber et al. (2003), Rowe et al. (2004), and Endler and Mielke (2005). Because bowerbirds are similar in size, we used two avail-able ST eye parameters: pupil area 8.6 X 10-6 m2 and eye posterior nodal distance 8.5 X 10-3 m (GO is slightly smaller). We used average pa**erine data for cone cross-sectional area 8 X 10-13 m2 (similar for all photoreceptors), specific absorbance of cone outer segments 0.015 µm-1, and cone outer-segment length 13-15 µm (Nathan Hart and Julian Partridge, pers. comm.). These parameters are not critical because they drop out in the light adaptation calculations (Endler and Mielke 2005). We used average parameters of woodland birds for the ocular media transmission spectrum (Endler and Mielke 2005). The important parameters are λmax the wavelength of maximal absorption of each cone cla**, and λ50, the wavelength of 50% absorbance by the corresponding oil droplets. Published data show that bird species fall into two groups; those with V-type and those with U-type eyes (reviewed in Endler and Mielke 2005; Hart and Vorobyev 2005). For each eye type, we used the mean for each of the four cone cla**es within eye types given in Endler and Mielke (2005). The four cone cla**es are named VS (or UVS), SWS, MWS, and LWS, respectively; these names indicate violet- (or ultraviolet-), short-, middle-, and long-wavelength sensitivity, respectively. To date, the U-type eye has only been found in the rhea (but not in the ostrich), gulls, parrots, and higher pa**erines (Ödeen and Håstad 2003; phylogeny in Barker et al. 2004). Given their less-derived position in the pa**erine phylogeny (Barker et al. 2004), bowerbirds should have V-type eyes, the probable ancestral state for birds. Opsin DNA sequence work on great and satin bowerbird by Belinda Chang (pers. comm. 2004) and on all bowerbirds by Rob Fleischer, Paul Zwiers, and Gerry Borgia (pers. comm. 2005) also suggests that bowerbirds have V-type eyes. Note that the possession of a V-type eye does not imply lack of UV sensitivity, it just means less shorter-wavelength UV sensitivity than in a U-type eye (see fig. 1 in Endler and Mielke 2005). Both types of bird eyes are much more sensitive to UV than humans, and ST are known to use UV in bower a**essment (Doucet and Montgomerie 2003). For further discussion, see Endler and Mielke (2005). We found two light environments to be common on bowers, woodland shade and open/cloudy (terminology of Endler 1993a). We calculated the effects of light adaptation using a mixture of these two habitat spectra because bowerbirds experience both at the bower; this is essentially white light. The effects of using other light environments will be discussed in another paper and have small effects compared to those discussed here. We used the log-linear version of the Vorobyev and Osorio's (1998) receptor noise model (Vorobyev et al. 1998, 2001) this means that we a**ume that the eye is accommodated to the light at the bower. For details, see Endler and Mielke (2005). After the calculations, each patch in a color pattern is a**ociated with four values {S} = {Svs, Ssws, SMWS, SLWS} (SUVS in the U-type eye). A color pattern with n patches can be described by an n X 4 matrix S, with the {S} as rows. S is the raw stimulus matrix a**ociated with a color pattern captured by the eye under the specified conditions. We calculated S for plumage, bower structure, bower ornaments, and visual backgrounds for each species, using both the V- and U-type eye parameters. Calculations using published species-specific parameter sets (Endler and Mielke 2005; Hart and Vorobyev 2005) yielded results quantitatively similar to either the V-or U-type eye parameters and will not be reported here. Compositional Data and Color Space Because color is processed by the visual system independently of brightness and color perception is based upon rel-ative photoreceptor stimulation, S data are appropriate for compositional an*lysis (Endler and Mielke 2005). Compositional an*lysis is designed for data for which the relative values are of greater interest than the absolute values (Aitchison 2003) as for cone capture rates in color vision (Kelber et al. 2003). We converted S to the compositional matrix Sc by dividing each row of S by its row total; each row of Sc sums to one. We removed the nonindependence of Sc columns (Aitchison 2003) by transforming the points (Sc rows) into positions in a tetrahedron with height = 1; the position of a patch's point in tetrahedral space {ST} = {x, y, z} is set so that the distances between that point and each tetrahedron face is given by its row in Sc (Aitchison 2003, ch. 1). The resulting n X 3 data matrix ST describes the color pattern as a cloud of points in tetrahedral space (for discussion and the transformation, see Endler and Mielke 2005). Relationships among the patches (ST rows) can be described by multivariate statistics (Aitchison 2003). ST has been used to plot cone captures (Burkhardt 1989; Goldsmith 1990; Vorobyev et al. 1998; Thery and Casas 2002; Kelber et al. 2003), but those studies did not an*lyze the data. Here we use ST to examine differences among color patterns and relationships among their components. Statistical an*lysis Standard multivariate statistics a**ume multivariate normality, sphericity (equal variances in each dimension), h*mogeneity of variances among groups, and error variables having independent normal distributions. Tests of S and ST (Aitchison 2003) indicate that all of these a**umptions are invalid. We therefore used a nonparametric multivariate statistical method, LSED-MRPP, which is independent of these a**umptions. We developed the method to do the equivalent of a nested (hierarchical) an*lysis of variance (Endler and Mielke 2005). Our method avoids another problem: standard techniques test for differences among means, but this is inappropriate for color patterns because the eye and brain do not work on mean colors. For example, a color pattern having equal total areas of red and green and another with blue and yellow would both have means of gray; standard methods cannot distinguish them, but LSED-MRPP can do so easily. For details, justification, and tests of its efficiency in detecting differences, see Endler and Mielke (2005). We report the results of the among-group LSED-MRPP tests and the a**ociated among-group effect size or disparity. The effect size K is the within-group agreement measure and its magnitude increases with the divergence between groups (Mielke and Berry 2001, p. 28). It is very different from conventional effect size measurements: K increases with differences in variance, skewness, and other aspects of distributional shape as well as differences in medians, whereas conventional effect sizes are sensitive only to differences in means (Rosenthal et al. 2000). Consequently, K values are not comparable with conventional effect size values. Color patterns are generally significantly different when K > 0.01, significant and distinctly different when K > 0.05, and significant and well separated when K > 0.2 (details in Endler and Mielke 2005). To avoid confusion between K and conventional effect size measures we will refer to K as the disparity between color patterns. SENSORY PROCESSING AND COLOR PATTERN EVOLUTION Photoreceptor Noise Ellipses It is important to know whether colors are distinguishable by birds. Mutation and genetic drift could cause changes in color in different parts of the plumage, but if not distinguishable, they would not be selected. The inherent noise in photoreceptors can prevent discrimination between colors that are too similar (Vorobyev and Osorio 1998) or points in the tetrahedron less than a critical distance apart. To estimate this error we constructed error ellipsoids in tetrahedron space. For diurnal (normal display) conditions, the receptor noise of photoreceptor type r is approximately er = w/√gr, where w is the Weber fraction and gr is the relative abundance of photoreceptor r (Vorobyev and Osorio 1998; Vorobyev et al. 1998). A typical value of w is 0.05 for vertebrates (Vorobyev, pers. comm.). The gr generally increase with λmax, we used typical values of 1 (UVS/VS), 2 (SWS), 2 (MWS), and 4 (LWS) (Hart 2001b). Typical values have been used to calculate er and to predict successfully the discrimination among colors and spectral sensitivity in many species (see Endler and Mielke 2005). Previous studies calculated a critical difference (color space distance) needed to discriminate two colors (Kelber et al. 2003); here we calculate the error around a given point. To do this we take {S} as the origin of a four-dimensional ellipsoid (4D ellipse) with radii er. Points inside this ellipsoid are indistinguishable from {S} due to receptor noise. Next we take points on the surface of this 4D ellipsoid (extreme error values) and transform each of them into tetrahedral coordinates. This creates a 3D ellipsoid showing the scale of confusion between points in tetrahedral space. Examples are illustrated in Endler and Mielke (2005). In terms of Vorobyev and Osorio's (1998) model, two points are just distinguishable by color if their ellipsoids barely touch and are likely to be confused if their ellipsoids overlap. Generally, points within the scale of the ellipsoid will be indistinguishable and points further apart will be distinguishable. Assumptions and Limitations of our an*lysis The receptor noise model was designed to explain and predict near-threshold detection of color when visual fields are large enough for color detection and discrimination between pairs of colors at or near the threshold of discriminability (Vorobyev and Osorio 1998; Vorobyev et al. 2001; Kelber et al. 2003). Both the model and real visual systems work best at higher light intensities and for larger targets; color delectability declines with declining irradiance and target size (Vorobyev and Osorio 1998; Wyszecki and Stiles 2000). Irradiance in the bowers were 50-500 µmol m-2 sec-1 and much higher in sun flecks (direct sunlight in the open is about 1800 µmol m-2 sec-1; Endler 1993a), ample for color vision. During displays, females view 0.5 to 2-cm ornaments in the avenue and 1 to 4-cm ornaments on the court (6-15 cm in TB), at distances of 10-20 cm. Allowing for loose cone packing with centers at 9 X 10-13 m and the other eye parameters given above, the average bower ornament subtends target angles of 2-20° or 104-106 cones of a single type. These should be large enough for color detection receptive fields. These target angles are very much larger than pigeons (with similarly sized eyes) can discriminate under often darker conditions in the laboratory (Osorio et al. 2001). Males manipulate ornaments at shorter viewing distances and experience still larger target angles. Consequently it is reasonable to use these methods to model visual performance based on color. The receptor noise model a**umes discrimination near the threshold, but we are using it to compare color patterns whose components vary from below through well above the threshold. We a**ume that perceived color difference is positively related to the number of threshold steps (discussion in Wyszecki and Stiles 2000); more distant points (different spectra) are perceived as more different than close ones. This a**umes no categorical perception or other nonlinearities. For example, humans categorize a large number of stimuli as green, and a step between greens is not the same as a step of equal size between green and yellow. However, we can also distinguish many different kinds of greens and yellows. Studies of a variety of human populations show that the ability to discriminate colors is independent of categorical boundaries, and this pattern is consistent across a variety of different ethnic groups who might be expected to use different categories (Heider and Olivier 1972). The ability to distinguish colors is separate from decisions to put them into categories. In our an*lysis we are basically asking: No matter how birds categorize the colors, does a color scheme differ from another color scheme? Although our method ignores nonlinearities in perception, it successfully predicts mate choice (Endler and Houde 1995), and other color-guided behavior in a variety of taxa (Endler and Mielke 2005). This linear a**umption is widely used in both human and animal vision research (e.g., Chittka 1996; Wyszecki and Stiles 2000; Kelber et al. 2003; Rowe et al. 2004). For these reasons, we feel it is a useful way to investigate the evolution of color patterns. We will consider the possibility of categorical per-ception of ornaments in another paper. Our an*lysis does not a**ess overall conspicuousness because it ignores luminance (brightness), pattern shape, and motion. These aspects are processed independently of color in bees, birds, and humans, often by different kinds of photoreceptors (Osorio et al. 2001). A full a**essment of these factors would be necessary to predict, for example, mating success as a function of color patterns. In spite of this, our method successfully predicts mate choice in guppies (Endler and Houde 1995), whose patches subtend much smaller viewing angles and have larger luminance ranges than bowerbird patches. Here we are concerned only with the color components; a future study will a**ess the joint effects of color and luminance. RESULTS We tested for differences between all components of the color patterns and between the color patterns and the visual backgrounds of each species. Table 2 shows the results of these tests for both the V-type (bowerbird) and U-type (e.g., higher pa**erine) bird eye types, for comparison (sample sizes in Table 1). Note the qualitative similarity of the two sets of results. Sequencing of bowerbird opsin genes by Belinda Chang (pers. comm.) indicates that bowerbirds have a V-type eye. Given this, and the qualitative similarity in the results of the two sets of eye parameters (Table 2), we will report only the V-type results. Table 3 gives the disparities for the V-based tests. TABLE 2. LSED-MRPP tests: Probability that two color patterns are drawn from the same distribution using both kinds of bird eyes as the basis of the calculations. Table values are -log(P) for the between group LSED-MRPP tests; for example, 2 indicates P = 0.01; subtract 2 for tablewise Bonferonni correction. A conservative value indicating differences between the color pattern components is 2-4. For each test: row I for V-type eye (bowerbirds) and row 2 for U-type eye (e.g., higher pa**erines) for comparison. Species codes and sample sizes (identical for both sets of tests) are shown in Table 1. Results for components versus near backgrounds are similar or stronger than for all backgrounds; an example is shown for ornaments. Sexual Dimorphism as a Test of the Methods If LSED-MRPP works properly, then s**ually dimorphic species should have high between-s** disparity values and monomorphic species should have very small values. See Pizzey and Knight (1997) and Frith and Frith (2004) for illustrations of the birds. Tables 2 and 3 confirm this-there are no significant differences between s**es for GR, SP, TB, and SC, and highly significant differences for ST, RG, and GO; disparities are either around 0.001 (monomorphic) or above 0.2 (dimorphic). TABLE 3. Disparities (K of LSED-MRPP tests) between color pattern components. Color patterns are different for disparities larger than 0.01, easily distinguished at 0.05, and completely different for disparities above 0.2 (Endler and Mielke 2005). Negative disparities indicate differences small enough to be caused by measurement error. Disparities are not necessarily comparable for different kinds of distributional differences as found among components (rows); see Endler and Mielke (2005) for details. Species codes are defined in Table 1 Contrast with the Visual Background For a signal to be effective in intraspecific communication, it should be different from the visual background (Endler 1978, 1980, 1993b; Lythgoe 1979; Endler and Thèry 1996; McNaught and Owens 2002; Gamble et al. 2003; Gomez and Thèry 2004; Rowe et al. 2004). The results are shown in Tables 2 and 3. There are significant differences between all components of the color patterns and the visual backgrounds except for the plumage of FB (Table 2), suggesting that FB is the most cryptic of the species. Although differences are significant, disparities are small for TB plumage and FB, RG, and TB bower structure, indicating that these components are not as conspicuous against their visual backgrounds as the other components and species (Table 3). Females of the dimorphic species (ST, RG, GO) are less conspicuous (lower disparity, Table 3) than males. The differences in plumage-background relationships among species are easier to see in a diagram (Fig. 3). FIG. 3. Differences between plumage and visual backgrounds for each species (species codes in Table 1). Each point is a single measurement in tetrahedral coordinates, projected on the S-M-L face of the tetrahedron (looking down along the z- or V-axis). The orientation and scale of the tetrahedral base is shown in the scale panel. The other panels are magnified and do not show the tetrahedron edges. The vertical and horizontal lines go through the gray point (equal stimulation of the cones). The visual background data are so numerous (Table 1) that they are shown as density contours with black the highest density. Circles, male samples; triangles, female samples; closed symbols, nondisplay parts of plumage; open symbols, display parts of body Figure 3 shows the way plumage and visual backgrounds differ for each species looking vertically downward on tetrahedral space. This view of the data, which ignores the V cone output, is useful because the leaf litter visual background components have little UV reflectance, and strong UV reflectance is only found in ST plumage; it underestimates differences but illustrates the patterns. (The corresponding statistical tests in Tables 2 and 3 use the full tetrahedral space.) Very similar patterns are found using the U-type eye, plotting the data as relative cone difference (opponency) functions such as (MWS - VS)/(MWS + VS) versus (LWS - SWS)/(LWS + SWS), or plotting the data in the segment cla**ification space of Endler (1990). A color pattern will be more difficult to detect if it is entirely within the range of the visual background and will be more conspicuous the further outside the background range it is found (Endler and Thery 1996). GR and SP are dry woodland species, hence the visual background contours range from gray to reddish brown (left to right). The other species are rainforest species (where sampled for FB and ST) and the upper lobe of the background distribution consists of live (green) leaves and stems that are largely absent in woodland backgrounds. The more concentrated the plumage points are in the darker shaded portions of the background contours the less conspicuous will be the color patterns. The species vary greatly in plumage-background differences (Table 3) and in how they are different (Fig. 3). FIG. 4. Differences between ornaments and visual backgrounds for each species. As Figure 2 except that plumage is not shown, circles indicate natural ornaments, and closed triangles indicate artificial (man-made) ornaments. Like the plumage, the bower structure and ornaments are highly significantly different from the visual backgrounds (Table 2), but with larger disparities (Table 3; except for RG, which may be a small sample effect-they use few ornaments). Disparities are larger when the comparison is made between the ornaments and visual backgrounds adjacent to the bowers (near backgrounds) compared to comparisons to random background measurements made in the same habitat more than 10 m away from the bowers (far backgrounds, Table 3), suggesting microhabitat choice for bower location. Figure 4 shows the distributions of ornaments and backgrounds using the same projection as in Figure 3; however, this is a stronger underestimate of the differences because UV is common in many of the ornaments, particularly in ST (all LSED-MRPP tests include all three dimensions). The differences between ornaments and backgrounds is much greater than that of plumage and backgrounds; background disparities are 0.001-0.074 for plumage and 0.068-0.187 for ornaments (Table 3). Effects of Adding Bower Structure and Ornaments to the Ancestral Visual Signal In addition to contrast with the visual background, a good signal should have high within-pattern contrast (Endler and Houde 1995; Endler and Thery 1996). We can measure within-pattern contrast in two ways: volume of color space occupied by the color pattern ST and the mean distance between color pattern component patches. Volume is a measure of the range of colors in the pattern; a larger volume means that more different colors are present. Colors can vary (in human terms) in hue, chroma, or both but a larger volume means a greater range of both hue and chroma. The mean distance between points is a measure of how different the colored patches are within the color pattern-larger mean distances indicate relatively more even spacing of colors and hence greater contrast between different parts of the pattern. Both contrast measures are needed because a color pattern could have a low volume but a large mean distance or vice versa. For example, a color pattern with many very different colors but smooth intergradations among them would have a large volume and a smaller mean distance. Color patterns with the greatest within-pattern contrast should be larger by both measures. TABLE 4. Changes in within-pattern contrast accumulating the effects of bower structure, natural ornaments and artificial ornaments. Volume: volume occupied by all of the points (patches) in the color pattern. Distance: mean distance between all pairs of points calculated over all points in the color pattern. Color pattern codes: b, bird colors only; bs, birds + bower structures; bsn, bs + natural ornaments; bsna, bsn + artificial ornaments. Significance levels are the results of bootstrap tests and indicate the fraction of bootstrapped values equal to or greater than the observed value. Symbols: +P = 0.051; *P < 0.03; ***P < 0.001 all after the Bonferonni correction. Significance levels apply to the difference between a given cell and the cell above it (incremental difference); for example 10.9*** (GR column) indicates that 10.9 (bsn) was significantly greater than 4.4 (bs) or natural ornaments significantly increase contrast. No symbol indicates no evidence for a difference; 4.4 (GR column) indicates that bower structure (4.4) does not significantly increase to the bird alone (1.4). The last number in a column for volume or distance indicates that species' current state. Species codes are defined in Table 1. During the course of bowerbird evolution, the birds added bower structure and natural ornaments, and after humans arrived, artificial (man-made) ornaments. We therefore calculated the two measures of within-pattern contrast for plumage alone (p), plumage + bower structure (ps), plumage + structure + natural ornaments (psn), and plumage + structure + natural + artificial ornaments (psna; not applicable to FB, GO, and TB, which do not use artificial ornaments). It is not clear whether structure preceded or followed ornaments; the TB position in the cladogram is problematic (see methods) as is what we could call "structure." Both TB (Frith and Frith 2004) and manakins (Uy and Endler 2004) clear courts for display and these, as well as the bower structures of the other bowerbirds, affect the visual contrast of the bird. More-over, juvenile GR and SP cover practice display areas with twigs and later add ornaments, and eventually go on to build avenue bowers when older. By "structure," we mean a substrate cleared (TB) or constructed (others) against which ornaments are displayed; in that sense structure came first. Similar results were obtained when reversing the order of structure and natural ornaments. We tested the effects on the two measures of contrast by taking the observed ratio (ps/p, psn/ ps, or psna/psn) as the observed statistic, bootstrapping the pooled data (e.g., p and ps for the ps/p test) with the same data structure (sample sizes) calculating the bootstrapped ratio for each bootstrapped calculation, then examining what proportion of the 1000 bootstraps fell above (increase) or below (decrease) the observed value. The data and test results are shown in Table 4. Bower structure has little effect on within-pattern contrast while ornaments, particularly man-made ornaments, have a large positive effect. Elaboration and Innovation If the bower components (structure and ornaments) are an elaboration of the plumage (Marshall 1954), then, by our definition, the relationships among the bower colors should be the same as the relationships among the colors of the plumage; ST of plumage and bower should have similar structure (Fig. 1). If the relationships among the bower components differ from the plumage, then the bower color patterns are innovations (Figure 1). Figure 5 shows the relationships among the plumage components as stereo-pair views of the tetrahedral color space. It is apparent that the plumage of each bower-building bowerbird falls on a line (Fig. 5), and the lines have varying orientations. The non-bower-building catbird (SC) is both less linear than the bower building species and has an orientation almost perpendicular to the bower builders. The pattern in Figure 5 (Figure is large - SEE THE PDF) is confirmed by a principal component an*lysis (PCA) on each bird separately, on the pooled data, and remains no matter what bird eye parameters are used. Table 5 gives the percentage of the total variance among color pattern components explained by each PCA for each species for both the V- and U-type eyes. As expected from the linear distribution of the data, the first principal component (PC1) explains about 90% of the variance, except in SC. Means of the individual bird percentage explained are slightly higher (not shown) because each bird's line does not have exactly the same centroid or 3D slope (eigenvectors). Note how PCI explains less in the U-type than the V-type eye (Table 5) as expected from the opsin sequences and position in the phylogeny. The PCA are presented in Table 5 for three sets of data, male plumage (whether or not s**ually dimorphic) excluding the parts of the body used in visual displays (D components), males and females pooled excluding D, and males and females including D. Note the small decline in the percentage explained by PC1 as we add females and display components. This is expected because the D (open circles in Figs. 3, 5) are distributed away from the nondisplay components, particularly in the most derived species (GR, SP, FB). We obtained the same pattern using different kinds of an*lysis, including PCA of the simple cone outputs (S, excluding PC1, which is luminance in S), and separate sets of eye parameters from each species for which there are published data. The explanation is simple. Except for the catbird (SC), bowerbird plumage color pattern components appear to consist mostly of a mixture of two colors (pigments and/or structural colors) in various proportions. This forms a linear pattern in any linear transformation of the data. PC2, which explains 3- 10% of the variation, is largely due to the D components that involve a third base color, and therefore yields a narrow planar plumage distribution. In contrast, the catbird is broader, with PC2 explaining about 20% of the variation. TABLE 5. Percentage of the bird color pattern variance explained by the three principal components. D indicates the parts of birds actively used in displays. Species codes are defined in Table 1. The three rows are for the three PCs, respectively, for example, 93.1, 3.6, and 3.3. Unexpectedly, males and females of all but ST fell on the same species-specific PC1 line (Fig. 5); this includes two of the three s**ually dimorphic species (RG, GO). The angles between male and female PCI lines are GR, 16.0°; SP, 10.7°; ST, 21.0°; RG, 10.9°; GO, 4.7°; TB, 7.6°; and SC, 17.3° (no data for FB). Tests of differences between these s**ual dimorphism angles using 10,000 bootstraps yielded P ranging between 0.1 and 0.6, except for ST (P < 0.001). Although significant, the angular difference between male and female ST is small (21°). There was significant s**ual dimorphism in the angle between the PC1-PC2 planes and the distances between the centroids for ST, RG, GO (P < 0.02 to < 0.001). The plane angle differences disappear for RG and GO if the D components are removed before an*lysis. This indicates that, except for ST, the significant s**ual dimorphism (Table 2) is along the PC1 line and in the parts of the bird used by males in display. Differences in Plumage among Species The non-bower-building SC has a distribution completely different from that of the bower building species, with PC1 being weaker and almost orthogonal to that of the bower-builders. The differences between the bower-building species vary. We tested for possible function in species recognition and isolation by comparing sympatric and nonsympatric (allopatric + parapatric) species with respect to the between-species angles between PC1, angles between the PC1-PC2 planes, the distance between the species centroids (mean colors), and the differences between their volumes in the tetrahedron using the Kruskal-Wallis test. ST, GO, TB, and SC are sympatric in the north; ST, RG, and gc are sympatric in the south; and the other species are allopatric or parapatric (Frith and Frith 2004). We excluded SC from the an*lysis because it is always sympatric with other species and is very different from them and would be expected to bias the results. We found that sympatric bower-building species had significantly larger angles between their PC1 lines and greater centroid distances than between allopatric species (PC1 angle 33.9 ± 17.8 versus 15.1 ± 11.8, P = 0.045; centroid distance 0.24 ± 0.08 versus 0.11 ± 0.07, P = 0.016). Although the other measures went in the same direction, test differences were not significant (P = 0.057 and P = 0.61 respectively), suggesting that the location and orientation of the plumage line (which goes through the centroid) may have been involved in character displacement and may be used in species recognition. TABLE 6. Plumage line length ratio and measures of color pattern component divergence from the plumage line (PC1 axis) relative to receptor error around the line. Line ratio: ratio of plumage line length to error ellipsoid width along the PC1 axis. A pattern is recognizable as a particular color scheme if the line ratio is greater than 2. To be conservative length was calculated as the range of nondisplay plumage components; ratios are larger when display components are included. For each color pattern component, the first row is the percentage of patches distinguishable from the plumage color scheme, and the second row is a measure of how divergent the component is from the plumage. First row: percentage of color pattern components outside of the error ellipsoid around the plumage line axis (using the PC2-PC3 plane); second row: mean of the ratio of deviation from the PC1 axis to the mean ellipsoid radius for patches outside the ellipsoid. Points are distinguishable from the PC1 axis if this deviation ratio is greater than 2. Species codes are defined in Table 1. Discrimination of the Plumage Line and Plumage Display Components Because the plumage components fall along a line (PC1, see Fig. 5) we can ask how much of the variation along the line is distinguishable and whether the display components (D) are distinguishable from the rest of the plumage on the line. Color pattern components (points) closer than the size of the ellipsoid centered on PC1 are not likely to be distinguished from the line. Variation along the line is detectable if the line is longer than at least twice the width of the centroid centered on that (PC1) axis. Colors (such as D) are distinguishable from the plumage line if they are outside a rodshaped solid created by calculating noise ellipsoids at all points along the line within the plumage range. These discrimination thresholds are most easily seen by plotting them and the data with respect to the three PCs (Fig. 6, green ellipses and blue symbols). For each species the range of points along PC1 is greater than twice the error ellipse width (Table 6; Fig. 6, left panels), indicating that the plumage color scheme is recognizable as a varying mixture of two colors; the plumage line is detectable. The deviations from the plumage line (Fig. 6, right panels) are interesting. For all species, the standard deviation of the nondisplay plumage components (Fig. 6, solid blue symbols) from the plumage axis (origin of axes in right panels of Fig. 6) is very similar to the radius of the noise ellipse (compare the black and green ellipses in the right panels of Fig. 6). The percent of nondisplay plumage that is outside the error ellipses range from 0% to 31% with the highest values shown in the intermediately derived species and s**ually dimorphic species (GO, RG, ST; Table 6). The relative size of the error ellipse suggests that the lines may be distinguishable in the one species that showed significant s**ual dimorphism in the lines (ST), but not as distinguishable as variation along the length of the line. These results suggest that most of the nondisplay plumage variation off the plumage axis is not detectable; minor color variation (dirt and wear) would not affect the perception of the plumage line. The thickness of the plumage line is therefore about the right size to be perceived strongly but be buffered against random developmental and environmental factors. The pattern for display plumage components is very different (open blue symbols in Fig. 6); 40-100% of them are outside the error ellipsoid, they are generally further from the line than the nondisplay components and far enough away to be distinguishable from the nondisplay components (Table 6). We conclude that the linearity of the birds' color patterns is de-tectable for each species, and the plumage display components (D) are detectably different from the rest of the plumage. Tests for Elaboration and Innovation We can use the linear relationship among the colors of the birds (the PCl line, Fig. 5) to test for elaboration and innovation. If bower structure and ornaments are elaborations of plumage, then the structure and ornament patch stimuli should fall on the same line as the plumage (Fig. 1). If the bower is an innovation, then the color components of the bower structure and ornaments should fall off the line (Fig. 1). Figure 6 shows the distribution of structure and ornaments plotted in the PC planes and Table 6 summarizes the geometric relationships. It is clear that the ornaments fall off the plumage line, but the bower structure may only be partially distinct for ST and GO (Table 6). The ornaments fall many receptor error units away from the line (Fig. 6), indicating that they are easy to distinguish from the plumage, with man-made ornaments more distinct than natural ones (Table 6). The differences are not an artifact of color availability because the distributions overlap in 3D. There is a tendency for the maximum distance away from the line to increase with phylogenetic position-it is least in the least derived SC; intermediate in the intermediately derived GO, RG and ST; and greatest in the most derived GR, SP, and FB. In addition, RG is less derived than ST and its ornaments also show less deviation from the line than ST. The large differences in distributions of plumage, structure, and ornaments is also shown by the overall LSED-MRPP tests (Tables 2, 3), but innovation can be shown directly by testing for differences in deviation off the line. To test for innovation we performed LSED-MRPP tests on the distribution of colors in the PC2-PC3 plane (right panels of Fig. 6); all variation in this plane represents deviations from the line, which is at the origin of this plane. As in the original tests (Tables 2, 3) we examined whether the distribution of bower structure or ornaments is different from that of plumage. If the tests are nonsignificant, then the components are an elaboration of plumage; if significant, then they are innovations from plumage. For natural, artificial, and all ornaments all LSED-MRPP tests yield P < 10-8 with disparities of 0.03 to 0.5. For bower structure, all tests yielded P < 10-8 except FB (P = 0.002) and RG (P = 0.0002), with disparities of 0.02 to 0.39. The results are virtually identical using the U-type eye parameters. For all species, both bower structure and ornaments are innovations from plumage and not elaborations. FIG. 6 (Figure is large - See PDF paper). Distribution of plumage and bower components in tetrahedral color space (Fig. 5) projected onto the plumage principal component (PC) planes. The left panels show the data on the PC1-PC2 plane and the right panels show the data on the PC2-PC3 plane. A plot of PC1 versus PC2 is a planar slice through Figure 5 along the long axis (PC1) of the plumage variation with all points projected on that plane. A plot of PC2 versus PC3 is a slice through Figure 5 perpendicular to the long axis of the plumage with all points projected on that plane. Closed blue circles, male plumage components not actively used in displays; open blue circles, display components; blue triangles; female plumage; black +, bower structure; gray triangles, natural ornaments; inverted gray triangles, artificial ornaments. Black ellipses enclose two standard deviations of the plumage data. Green ellipses are zones of confusion calculated from the receptor noise model, centered on the data centroid (see text for details). These confusion ellipses indicate how different (far apart) points have to be in order to be distinguishable. Think of them as a scale or sliding window on the plane; all points in the window are indistinguishable. As plotted they indicate what patches (points) should be distinguishable from the plumage line (if outside the green ellipse). For example, most patches not used in displays (solid blue symbols) will not be distinguishable from the line but most display components and ornaments will be (Table 6). The ellipses have different shapes for each species because the species lines have different orientations, and therefore cut through the ellipsoids (3D ellipses) at different angles. The original version of Marshall's (1954) elaboration hypothesis was verbal, made with reference to human vision, and restricted to hue; for example, because the satin bowerbird (ST) is largely blue and its ornaments are blue, Marshall suggested that it chooses ornaments to elaborate and enhance its plumage; ornaments and plumage have the same hues. That observation was based upon human vision, which ignores UV (300-400 nm), although birds can see it. We can still test this restricted version of the hypothesis by defining a bird hue as the location of a line from the origin (gray point) to one of the faces of the tetrahedron; all points on that line have the same hue and elaboration is a change of distibution of points along that line. Changes in position along a hue line are changes in chroma. A species line would only correspond to a hue line if the species line went through the origin. A species would contain two complementary hues if its plumage points were on both sides of the origin on a line through the origin. Bowerbird species lines do not run through the origin because they contain two or more nongray base colors. We can test for elaboration of hue by mapping the hue lines for each patch as points on a sphere that touches each vertex of the tetrahedron; both the tetrahedron and this sphere are centered on the gray point. This is done for each point by converting its Cartesian coordinates (position in the tetrahedron) into 3D polar coordinates. This is equivalent to finding the latitude and longitude of each point's projection on the surface of the sphere. Each latitude-longitude pair (LL) defines a hue. Points at a given LL have the same hue but vary in chroma (distance from the gray point). Ornaments are an elaboration (in Marshall's sense) if they have the same distribution of LL as the plumage and an innovation if the distributions differ. To test this we first transformed the LL data into a 2D map using the Robinson transformation, using the Mapping toolbox in MATLAB (MathWorks, Natick, MA). Of all the geographic transformations, the Robinson is best because it minimizes distortion of both area and distance jointly. Figure 7 shows the plumage and ornaments in the Robinson projection, as well as the world for geometric reference. As before, the ornament distribution is very different from that of the plumage. This is confirmed by doing LSED-MRPP tests on the mapped data: all P-values were less than 10-11 except for GR (< 10-5) and RG (< 10-4), with dis-parities of 0.01 to 0.06. We therefore reject the original ver-sion of the elaboration hypothesis, as well as the more general one. We have clear evidence for efficiency innovation (Fig. 1) because the ornaments are significantly different from that of the visual background (Fig. 4; Tables 2, 3). Aside from the avoidance of visual backgrounds, are the ornaments chosen at random or in a particular direction? We could test this by subjecting the data in the PC2-PC3 plane (Fig. 6, right panels) to a Raleigh test, examining if there is a significant mean vector (Zar 1999) or if the vector differed from that of the plumage or backgrounds. Unfortunately, that test depends on an a**umption of unimodality, but our data are at least trimodal (Figs. 4, 6). We can, however, examine whether the ornaments actually used differ from what is available in the same habitat. To do this, we selected visual background data from objects that could be picked up by the bird and were within the size and weight range of objects actually used by the birds. We then used LSED-MRPP to test for differences between used and available objects. If the tests are nonsignificant, then ornaments are chosen at random and innovation is fortuitous, but if the tests are significant, then ornaments are a nonrandom sample of available objects. The LSED-MRPP tests all yield P < 10-6 and disparities of 0.03 to 0. 1; the birds are not choosing ornaments at random. At least some of this nonrandom choice is due to the avoidance of background colors. Rather than being a narrow preference for some colors, the birds do not take objects that have colors similar to that of the backgrounds. DISCUSSION We discuss the results in the context of the rough a**ignment into four degrees of derivation from the bowerbird ancestor (Fig. 2): no bower (SC), least derived (TB), intermediately derived (GO, RG, ST), and most derived (FB, GR, SP). For the entire family, s**ual monomorphism is ancestral (although plumage dimorphism is found in the lyrebird outgroup), appears in the intermediately derived species, and then disappears in the most derived species, except for degree of development of the nuchal crest in GR and SP. If we a**ume a molecular clock, then s**ual dimorphism was gained and later lost in both the avenue and the maypole clades at roughly the same times (Fig. 2). This suggests a parallel gain and loss of s**ual dimorphism in the two clades. This rise and fall in s**ual dimorphism in the Australian species is paralleled by the degree of conspicuousness of their plumage and bower structure against the visual backgrounds. At the same time the ornaments steadily increase in conspicuousness from least to most derived. It would be interesting to know if this were also true for the maypole lade in New Guinea. Species in the genus Chlamydera (FB, GR, SP) are distributed allopatrically or parapatrically relative to other species, but the others are found together in rainforest in the north (ST, GO, TB, and SC) and the south (ST, RG, and gc; see Pizzey and Knight 1997; Frith and Frith 2004). Note how sympatric bower-builder male components range outside the visual background in different directions (Fig. 3); sympatric species have very different color distributions, whereas allopatric species are similar (the latter is confounded by phylogeny-Chlamydera are all found by themselves). Although ST is found in both northern and southern localities, it is sympatric either with GO (north) or RG (south); note also how the distributions of GO and RG extend outside the background in the same direction, indicating the use of similar colors in allopatry and similar divergence from ST in sympatry. This suggests that the plumage may be used as a species recognition character and has diverged in sympatry. In contrast to plumage, there is no obvious divergence of the ornaments of sympatric species (ST, RG, GO); all three sympatric species show similar excursions into nonbackground color regions of color space (Fig. 4). This suggests that plumage is important in species recognition and not ornaments. The catbirds probably use the plumage both for species recognition and mate attraction (s**ual selection), but once bowers evolved, the mate attraction function could be restricted to the bower structure and ornaments. Restriction of mate attraction components to ornaments and removing them from the body is also likely to have reduced predation risk while the birds forage, although the reduced risk in foraging may have been counteracted by increased risk of displaying at a fixed site on the ground. FIG. 7 (Figure is large - see PDF of paper). Distributions of plumage and ornaments in hue space. Hue space is a projection of the tetrahedron onto the surface of the enclosing sphere, then projected onto a map as the Robinson projection. A map of the world in the Robinson projection is shown for geometric reference. Solid lines connect the edges and vertices of the tetrahedron. Black symbols, plumage; gray symbols, orna-ments; open circles, males; closed circles, females. SC has no ornaments and is shown for comparison to the bower-building species. The non-bower-building SC plumage is the most evenly distributed in color space of the species examined; it is less conspicuous than it would be if some colors extended beyond the rainforest floor colors, however it is more conspicuous than it would be if the shape of the plumage distribution were the same as that of the background-a color pattern is cryptic (inconspicuous) if it is a random sample of the background (Endler 1978). However, our measure of the SC disparity is against leaf litter, where other bowerbirds place their bowers. But catbirds spend little time on the ground, and most of the time they will be seen against green leaves; there the disparity will be very much less than against the unused leaf litter. For this reason, and because LSED-MRPP tests for any differences in distributions, it is not surprising that there was a significant difference between SC plumage and background (Table 2). The plumage of TB, the least derived of the bower-building species (Fig. 2), is well within the more common components of the background distribution, but does not correspond to the most frequent components, so again it is significantly different from the background (Table 2), but with a disparity only 24% of that in the SC (Table 3). The three intermediately derived species (GO, RG, ST, Fig. 2) are similar in that the females are inside the range of the backgrounds and the males are mostly outside the background range (Fig. 3), indicating that females are much less conspicuous to birds (including bird predators) than males. A similar pattern was found by Endler and Thèry (1996) for four s**ually dimorphic neotropical bird species. All plumage components of the intermediate species are significantly different from the background (Table 2), and disparities are much larger for males than either females or TB and SC (Table 3). Note how the axis of variation fits between the two lobes of the background distribution (Fig. 3); this allows females to be relatively cryptic on a large scale (within the background range) but still be distinct from the background (differently shaped distribution). The most derived species (FB, GR, SP) are s**ually monomorphic, and like the TB are located just off the more common background colors (Fig. 3). However, unlike all the other species, the plumage components used actively in displays (open symbols, Fig. 3) are well outside the background distribution, making them the only conspicuous part of the bird. There is no pattern with plumage display components in any of the less derived species. The evolutionary pattern is therefore distinct but semicryptic plumage in non-bower-builders, cryptic in the least derived bower-builder, cryptic females and conspicuous males in intermediate species, and cryptic plumage with only the plumage components used for display conspicuous in the most derived species. Although we have only data from one species in the maypole lade (GO), it is noteworthy that, like the avenue builders, the most derived maypole species (Vogelkop) is also s**ually monomorphic (Fig. 2va, vf). Plumage conspicuousness first increases then becomes restricted to plumage display locations and ornaments during the evolution of bowerbirds. In contrast to the plumage, the ornaments cause an increase in contrast during bowerbird evolution (Figs. 4, 6; Tables 3-6). The least derived species (SC and TB) extend only slightly beyond the visual backgrounds. Bowers of the intermediately derived species (RG, ST, GO) extend outside the background, particularly in RG and ST; one wonders if this could have been achieved by plumage alone in the presence of visual predation. Bowers of the most derived species (FB, GR, SP) show ma**ive excursions beyond the visual backgrounds with much larger disparities. It is clear that both plumage display (D) components and ornaments contrast with the visual background, indicating that they increase the efficiency of the visual signals, and that this contrast has increased during bowerbird evolution. Our conclusions are for conspicuousness to birds, and snake and mammal predators may see them differently. On the other hand, mammals and snakes have fewer cone types and most are nocturnal, so our comments about conspicuousness may be generalizable. In summary, the data and the phylogeny suggest the following sequence of events: (1) increase of contrast under s**ual selection; (2) onset of increased predation pressure with increased conspicuousness; (3) evolution of dimorphism to reduce predation on females; (4) transfer of conspicuous-ness to bower ornaments; and (5) reduction of conspicuous plumage components to parts of the males used in displays and consequent reduction of s**ual plumage dimorphism. Although the causes of each stage are speculative, the pattern is clear. The results appear to support Gilliard's (1956, 1969) transfer hypothesis, but things are a little more complex. First, the TB ornament pattern looks very similar to the plumage pattern of SC (Fig. 3). This supports the transfer hypothesis, but implies that the transfer of the s**ually attractive com-ponents from the plumage to ornaments occurred early in the evolution of bowerbirds (before TB; Fig. 2), rather than throughout the group, as implied by Gilliard. However, TB may be part of a trichotomy so the transfer may have occurred later in the two main clades. Second, the transfer hypothesis suggests that as the bowers increase in visual contrast the plumage contrast should decrease, but the range of plumage colors remains the same (Fig. 6, blue symbols and black ellipses). However, the conspicuous parts of the plumage evolve to become restricted to the display positions (Fig. 6, open blue symbols), which are a small fraction of body area and can be hidden. The net effect of this incomplete transfer is to increase the relative crypsis of the birds in the more advanced species. Increasing overall conspicuousness is a**ociated with increasing contribution to conspicuousness by the bower rather than the plumage. The contrast with the background, and hence signal efficacy, increased in other ways. Disparities are larger when the comparison is made between the ornaments and visual backgrounds adjacent to the bowers (near backgrounds) compared to far backgrounds in the same habitat (Table 3). This suggests some degree of microhabitat choice; bower sites may be chosen on the basis of backgrounds that give more visual contrast than if bowers were placed at random. We will explore microhabitat choice in another paper. Two of the intermediate species (RG, ST) also show significant divergence in the V axis (not shown in Figs. 4, 5); they use violet and UV-rich objects, which are rare in the background, in addition to other colors (UV is included in all the statistical tests). UV is also found in the nuchal crests of GR and SP, which is a highly efficient reflector in the bluish light of woodland shade, their typical habitat. It is clear that the visual signals of bowerbirds are very conspicuous and therefore should be easy to detect and discriminate (Endler 1993b). Efficacy of signals depends upon within-pattern contrast as well as background contrast (Endler 1993b; Endler and Thèry 1996). Bower structure effects are weak while ornaments are strong. Although adding structure (bs; Table 4) increases the volume contrast (range of hues and chromas), it only has a significant effect in GO, which has a very large bower structure (1-2 m high and wide). Except for TB, there is a consistent reduction of distance contrast (between patch contrast) by adding bower structure, it is significant only in RG. The relative uniform color of structure in combination with its larger size than the bird may account for some of this reduction. There is a consistent and highly significant increase of volume contrast as a result of adding natural ornaments (bsn; Table 4). Except for RG, there is also a highly significant increase of distance contrast of adding natural ornaments. For those species that use man-made ornaments, there is a significant positive effect of adding them (bsna; Table 4); the effect is marginal for RG after the Bonferroni correction. There is also a consistent increase in distance contrast, nonsignificant only for RG. The difference between RG and the other species may be due to small sample effects; regent bowerbirds have very few ornaments and we could not find very many bowers. We conclude that the effect of bower structure and especially bower ornaments is to increase visual contrast of the entire signal, particularly man-made ornaments. We could say that contrast is elaborated by ornaments, although the strong among-species differences in the details of which colors are used to increase contrast (Figs. 3, 4, 6) indicates innovation. Relating Table 4 to the phylogeny, the least derived species have the least within-pattern contrast and the more advanced species the most contrast. For both volume and distance contrast measures, SC (no bower) is about the same as TB, both about 2 or 9 resealedd for ease of comparison, see Table 4). Considering the species before the arrival by humans (bsn rather than bsna), the intermediately derived species (ST, RG, GO) increase markedly in volume (5-14) and to a lesser extent in distance contrast (10-13) compared to the least derived species. The most derived species (FB, GR, SP) are similar to the intermediately derived species (6-11 for volume and 9-15 for distance). With the advent of man-made ornaments the intermediately derived species increase further in volume (9-22) but little in distance contrast (14-15), but the most derived species that use them (GR, SP) increase dramatically in volume (37-39). We can conclude that ornaments significantly increase within-pattern contrast, and that there has a been a further increase in contrast when most species used man-made ornaments. Given that visual contrast is is augmented by man-made ornaments, the advent of Europeans in Australia might have accelerated bowerbird s**ual selection. This might even extend to color preferences; for example, red ornaments appear to be used extensively by GR only near cities or towns. Given that neither TB nor GO use man-made ornaments, and both are sympatric with the artificial-ornament-using ST, and we have seen ST bowers with artificial ornaments with a few meters of GO, and TB, it is surprising that neither GO nor TB use them. One possibility is that they are too specialized; the shape and size of the natural ornaments (large leaves with pale sides uppermost for TB, lichen and only a handful of fruiting structures and flower species for GO) may be too different from available man-made ornaments for any to be similar enough to be used. Another possibility is that their contrast works well enough, although Table 4 suggests that they could both increase contrast by adding artificial ornaments. The FB also uses no man-made ornaments, but its distribution is remote from most human habitation. However, an argument against the isolation explanation for FB is that they have a very large vocal repertoire of human and human-related sounds. The lack of use of artificial ornaments by FB, GO, and TB would be worth further study. Our results clearly reject Marshall's (1954) elaboration hypothesis (Figs. 6, 7). Considering bird rather than human vision, the birds are clearly not selecting ornaments that are elaborations (Fig. 1) of their own plumage. The original hypothesis was an artifact of human color vision. Instead, the bowerbirds are choosing colors and color combinations that are innovations relative to their plumage. Instead of an elaboration or a random innovation, bowerbirds are choosing color pattern components (ornaments) that increase the visual contrast with the background as well as increase the within-pattern contrast. There is a tendency for the degree of innovation (divergence from the line) to increase with phylogenetic position-it is least in the least derived SC; intermediate in the intermediately derived GO, RG and ST; and greatest in the most derived GR, SP, and FB. In addition, RG is less derived than ST and its ornaments also show less deviation from the line than ST. The effect of increased divergence from the line would be not only a stronger visual contrast, and hence greater or faster mating stimulation and discrimination, but also stronger species-specific signals, since the line and ornaments are further apart (Table 6). Both effects could result in a fitness advantage to the innovations. These ideas also need to be tested. A visual processing a**essment of bowerbird signals allows us to conclude that different components of visual signals evolve separately. Plumage first increases and then decreases in contrast, while ornaments and the entire color pattern increase in contrast. The transfer of s**ually selected parts of the visual signal onto the bowers and reduction to a few parts of the plumage allows relatively increased crypsis of males to predators as well as better intraspecific communication. In spite of the evolutionary shifts, high within-pattern contrast and high signal-background contrast is maintained and even increased. Ornaments do not act to elaborate existing plumage but instead are an innovation that increases signal efficacy. Species identification may be facilitated by plumage but not ornaments because only the former appear to have undergone character displacement in sympatry. an*logous phenomena may have occurred among different components of body coloration in other species but this has not been investigated because color patterns are normally treated as single traits evolving a balance between selective factors rather than suites of traits with semi-independent or independent functions and evolutionary histories. Such an approach may reduce some of the apparent chaos in color pat-tern evolution. The bower-building species' plumage fall on species-specific lines, and the non-bower-building SC plumage data are much less linear with a very different orientation. This supports Gilliard's transfer hypothesis if we a**ume that the plumage is retained for species recognition while most of the other plumage functions have been transferred to the bower. The species recognition function is supported by sympatric species having more divergent plumage and ornaments than allopatric species, the angles between PC1s of sympatric species being more divergent than among allopatric species, and hybridization only being known among species of Chlamydera that have very small PC1 angles between them. Considering neural processing, the presence of lines for possible species recognition is particularly interesting. PCs are linear combinations of relative cone outputs, which is the same processing done by the horizontal and amacrine cells in the retina, forming the basis of color perception (Jacobs 1985; example in Villa et al. 1991). This suggests that, if the relative input coefficients to a particular kind of horizontal or amacrine cell happened (or evolved) to be the same as the coefficients in the species plumage PC1, then a particular cla** of ganglion cell would only respond to a color pattern containing the appropriate set of colors. We will call this hypothetical ganglion cell the species-recognition ganglion (SRG). This is a computationally trivial mechanism for species recognition. It should be less costly than more complex systems because it requires fewer neurons; there will be lower maintenance costs and shorter time delays because there are fewer synapses. There is an additional advantage of linear species recognition colors. If males and females fall on the same line (as all do except for ST, and ST is not that far off the line) then the same circuit can be used by both s**es to recognize species, requiring fewer genes and fewer s**-senstive developmental switches during maturation. Evidence suggests that stickleback may be doing something similar (Rowe et al. 2004). There is extensive evidence from a variety of animals that apparently complex information about the world can be extracted by very simple neural processing (Prete 2005). Because maintenance of neural architecture and neural processing are energetically costly, there is no reason why this cannot happen in birds and other higher taxa as well. We hope that the SRG idea will eventually be tested directly by neurobiologists. The distribution of colors around the species line has about the same radius as that predicted by the receptor noise confusion ellipsoid. This has additional implications for species recognition. If the ellipsoid is an accurate measure of the scale of the range of colors that are not distinguished by bowerbirds, then all colors falling within that range centered on the line will stimulate the SRG in the same appropriate way. Individuals with slightly dirty plumage or having minor plumage development accidents would stimulate the SRG as strongly as normal individuals because this slight variation would still be within the critical radius around the line (largest ellipsoid radius). On the other hand, individuals that were too dirty, were subject to strong developmental accidents, or have drifted or actively evolved further away than the critical radius would yield a weaker SRG signal than individuals with colors entirely inside that range. This would result in stabilizing selection for species recognition components to remain within the rod-shaped error zone around the species line. This would constrain species-line evolution relative to the rest of the signal, and we predict more phylogenetic information in species recognition lines than other components of color patterns. The presence of a SRG does not preclude the use of plumage (and other) components for s**ual selection. However, to avoid reduction of the SRG signal strength and to minimize confusion between the two components during retinal processing, we predict that they should be in geometrically separate parts of the overall color pattern, as in bowerbird plumage and bowers, or geometrically separated on the body, as in the nuchal crest versus the rest of the plumage in Chlamydera. In bowerbirds the separation can be very strong, but in other species the body will have to do it all. A consideration of neural processing may make species recognition, mate choice, and the evolution of s**ual traits more predictable and less seemingly mysterious. Identifying the target of selection by considering how color patterns are processed yields additional testable predictions. Species-specific signals will work better if they are computationally simple to detect-as are the species-specific lines we found in bowerbird plumage. Therefore, we predict that species living with many sympatric species should have com-putationally simpler species-specific color pattern components than those living with few or none. Computationaly simple color patterns may also be easier to change than complex ones because fewer neuronal connections are needed, consequently we also predict that speciose genera should have computationally simpler color patterns than species-poor genera. We expect no such pattern for mate a**essment signal components for two reasons: cheating and species-independent properties. First, the tension between the best-quality males favoring easily distinguishable patterns and the more abundant average or low-quality males favoring less easily distinguishable patterns should on average select for mate a**essment patterns that are computationally more difficult to recognize and discriminate than species-recognition marks. This should cause marked divergences in phenotypes and gene phylogenies of s**ually selected and species recognition traits. Second, if certain colors always relate to condition in the same way, neural circuitry will converge (or not diverge) among species to detect them. For example, higher concentrations of carotenoids would usually result in relatively stronger VS and LWS than SWS and MWS cone outputs since they reflect in the UV and longer wavelengths. Consequently, circuitry that calculated something like (VS + LWS - SWS - MWS) would evolve in any species that used carotenoids in mate a**essment. We need to understand more about how signals are processed to understand the course of s**ual selection and speciation. The target of signal selection is perception rather than morphology or genotype. ACKNOWLEDGMENTS LITERATURE CITED Aitchison, J. 2003. The statistical an*lysis of compositional data. Rev. ed. Blackburn Press, Caldwell, NJ. Allen, E. S., and K. E. Omland. 2003. Novel intron phylogeny supports plumage convergence in orioles. Auk 120:961-969. Andersson, M. 1994. Sexual selection. Princeton Univ. Press, Princeton, NJ. Barker, F. K., A. Cibois, P. Schikler, J. Feinstein, and J. Cracraft. 2004. Phylogeny and diversification of the largest avian radia-tion. Proc. Natl. Acad. Sci. USA 101:11040-11045. Bennett, A. T. D., I. C. Cuthill, and K. J. Norris. 1994. Sexual selection and the mismeasure of color. Am. Nat. 144:848-860. Borgia, G. 1985. Bower quality, number of decorations and mating success of male satin bowerbirds (Ptilonorhynchus violaceous): an experimental an*lysis. Anim. Behav. 33:266-271. --. 1995. Complex male display and female choice in the spot-ted bowerbird: specialized functions for different bower deco-rations. Anim. Behav. 49:1291-1301. --. 1997. Comparative behavioral and biochemical studies of bowerbirds and the evolution of bower-building. Pp. 263-276 in M. L. Reaka-Kudla, D. E. Wilson, and E. 0. Wilson, eds. Biodiversity. Vol. II. Understanding and protecting our biolog-ical resource. Joseph Henry Press, Washington, DC. Borgia, G., I. Kaatz, and R. Condit. 1987. Flower choice and the decoration of the bower of the satin bowerbird Ptilinorhynchus violaceous: a test of hypotheses for the evolution of male display. Anim. Behav. 35:1129-1139. Boughman, J. W. 2002. How sensory drive can promote speciation. Trends Ecol. Evol. 17:571-577. Burkhardt, D. 1989. UV vision: a bird's eye view of feathers. J. Comp. Physiol. A 164:787-796. Chittka, L. 1996. Optimal sets of color receptors and color opponent systems for coding of natural objects in insect vision. J. Theor. Biol. 181:179-196. Coleman, S. W., G. L. Patricelli, and G. Borgia. 2004. Variable female preferences drive complex male displays. Nature 428: 742-745. Coyne, J. A., and H. A. Orr. 2004. Speciation. Sinauer Associates, Sunderland, MA. Doucet, S. M., and R. Montgomerie. 2003. Multiple s**ual ornaments in satin bowerbirds: ultraviolet plumage and bowers signal different aspects of male quality. Behav. Ecol. 14:503-509. Endler, J. A. 1978. A predator's view of animal color patterns. Evol. Biol. 11:319-364. --. 1980. Natural selection on color patterns in Poecilia reticulata. Evolution 34:76-91. --. 1986. Natural selection in the wild. Princeton Univ. Press, Princeton, NJ. --. 1990. On the measurement and cla**ification of color in studies of animal colour patterns. Biol. J. Linn. Soc. Lond. 41: 315-352. --. 1991. Variation in the appearance of guppy color patterns to guppies and their predators under different visual conditions. Vision Res. 31:587-608. --. 1992. Signals, signal conditions, and the direction of evo-lution. Am. Nat. 139(Suppl.):S125-S153. --. 1993a. The color of light in forests and its implications. Ecol. Monogr. 63:1-27. --. 1993b. Some general comments on the evolution and design of animal communication systems. Philos. Trans. R. Soc. Lond. B 340:215-225. --. 2000. Evolutionary implications of the interaction between animal signals and the environment. Pp. 11-46 in Y. Espmark, T. Amundsen, and G. Rosenqvist, eds. Adaptive significance of signalling and signal design in animal communication. Proceedings of the fifth International Kongsvoll Symposium, September 15-20, 1998. Tapir Publishers, Trondheim, Norway. Endler, J. A., and A. L. Basolo. 1998. Sensory ecology, receiver biases and s**ual selection. Trends Ecol. Evol. 13:415-420. Endler, J. A., and A. E. Houde. 1995. Geographic variation in female preferences for male traits in Poecilia reticulate. Evolution 49: 456-468. Endler, J. A., and P. W. Mielke Jr. 2005. Comparing color patterns as birds see them. Biol. J. Linn. Soc. Lond. 54: in press. Endler, J. A., and M. Thèry. 1996. Interacting effects of lek placement, display behavior, ambient light and color patterns in three neotropical forest-dwelling birds. Am. Nat. 148:421-452. Frith, C. B., and D. W. Frith. 2004. The bowerbirds. Oxford Univ. Press, New York. Gamble, S., A. Lindholm, J. A. Endler, and R. C. Brooks. 2003. Environmental variation and the maintenance of polymorphism: the effect of ambient light spectrum on mating behaviour and s**ual selection in guppies. Ecol. Lett. 6:463-472. Gavrilets, S. 2004. Fitness landscapes and the origin of species. Princeton Univ. Press, Princeton, NJ. Gilliard, E. T. 1956. Bower ornamentation versus plumage char-acters in bower-birds. Auk 73:450-45 1. --. 1969. Birds of paradise and bower birds. The World Naturalist, Weidenfield and Nicholson, London. Goldsmith, T. H. 1990. Optimization, constraint, and history in the evolution of eyes. Q. Rev. Biol. 65:281-322. Goldsmith, T., and B. K. Butler. 2003. The roles of receptor noise and cone oil droplets in the photopic spectral sensitivity of the budgerigar, Melopsittacus undulatus. J. Comp. Physiol. A 189: 135-142. Gomez, D., and M. Thèry. 2004. Influence of ambient light on the evolution of colour signals: comparative an*lysis of a Neotropical rainforest bird community. Ecol. Lett. 7:279-284. Grether, G. F., J. Hudon, and J. A. Endler. 2001. Carotenoid scarcity, synthetic pteridine pigments and the evolution of s**ual coloration in guppies (Poecilia reticulate). Proc. R. Soc. Lond. B 268:1245-1253. Hart, N. S. 2001 a. The visual ecology of avian photoreceptors. Prog. Retinal Eye Res. 20:675-703. --. 2001b. Variations in cone photoreceptor abundance and the visual ecology of birds. J. Comp. Physiol. A 187:685-697. Hart, N. S., and M. Vorobyev. 2005. Modeling oil droplet absorption spectra and spectral sensitivities of bird cone photoreceptors. J. Comp. Physiol. A 191:381-392. Heider, E. R., and D. C. Olivier. 1972. The structure of the color space in naming and memory for two languages. Cognit. Psychol. 3:337-354. Heindl, M., and H. Winkler. 2003. Vertical lek placement of forest-dwelling manakin species (Aves, Pipridae) is a**ociated with vertical gradients of ambient light. Biol. J. Linn. Soc. Lond. 80: 647-658. Hill, G. E. 2002. A red bird in a brown bag: the function and evolution of colorful plumage in the house finch. Oxford Univ. Press, New York. Hoekstra, H. E., and M. W. Nachman. 2003. Different genes underlie adaptive melanism in different populations of rock pocket mice. Mol. Ecol. 12:1185-1194. Höglund, J., and L. Shorey. 2004. Genetic divergence in the superspecies Manacus. Biol. J. Linn. Soc. Lond. 81:439-447. Jacobs, G. H. 1985. Cones and opponency. Vision Res. 26: 1533-1541. Kelber, A., M. Vorobyev, and D. Osorio. 2003. Animal colour vision: behavioural tests and physiological concepts. Biol. Rev. Camb. Philos. Soc. 78:81-118. Kusmierski, R., G. Borgia, A. Uy, and R. H. Crozier. 1997. Labile evolution of display traits in bowerbirds indicates reduced effects of phylogenetic constraint. Proc. R. Soc. Lond. B 264: 307-313. Lythgoe, J. N. 1979. The ecology of vision. Oxford Univ. Press, Oxford, U.K. Lythgoe, J. N., and J. C. Partridge. 1989. Visual pigments and the acquisition of visual information. J. Exp. Biol. 146:1-20. Madden, J. R. 2002. Bower decorations attract females but provoke other male spotted bowerbirds: bower owners resolve this trade-off. Proc. R. Soc. Lond. B 269:1347-1351. --. 2003a. Male spotted bowerbirds preferentially choose, arrange and prefer objects that are good predictors of mating success. Behav. Ecol. Sociobiol. 53:263-268. --. 2003b. Bower decorations are good predictors of mating success in the spotted bowerbird. Behav. Ecol. Sociobiol. 53: 269-277. Madden, J. R., and K. Tanner. 2003. Preferences for coloured bower decorations can be explained in a nons**ual context. Anim. Behav. 65:1077-1083. Marshall, A. J. 1954. Bower-birds: their displays and breeding cy-cles. Clarendon Press, Oxford, U.K. McNaught, M. K., and I. P. F. Owens. 2002. Interspecific variation in plumage colour among birds: Species recognition or light environment? J. Evol. Biol. 15:505-514. Mielke, P. W., Jr.,and K. J. Berry. 2001. Permutation methods: a distance function approach. Springer-Verlag, New York. Miklosi, A., Z. Gonda, D. Osorio, and A. Farzin. 2002. The effects of the visual environment on responses to colour by domestic chicks. J. Comp. Physiol. A 188:135-140. Mundy, N. I., N. S. Badco*k, T. Hart, K. Scribner, K. Janssen, and N. J. Nadeau. 2004. Conserved genetic basis of a quantitative plumage trait involved in mate choice. Science 303:1870-1873. Nachman, M. W., H. E. Hoekstra, and S. L. D'Agostino. 2003. The genetic basis of adaptive melanism in pocket mice. Proc. Natl. Acad. Sci. USA 100:5268-5273. Ödeen, A., and 0. Håstad. 2003. Complex distribution of avian color vision systems revealed by sequencing the SWS1 opsin from total DNA. Mol. Biol. Evol. 20:855-861. Osorio, D., M. Vorobyev, and C. D. Jones. 1999. Colour vision of domestic chicks. J. Exp. Biol. 202:2951-2959. Osorio, D., A. Miklósi, and Z. Gonda. 2001. Visual ecology and perception of coloration patterns by domestic chicks. Evol. Ecol. 13:673-689. Pizzey, G., and F. Knight. 1997. A field guide to the birds of Australia. Angus and Robertson, Sydney. Prete, F. R., ed. 2005. Complex worlds from simpler nervous systems. MIT Press/Bradford, Cambridge, MA. Prum, R. 0. 1997. Phylogenetic tests of alternative inters**ual se-lection mechanisms: trait macroevolution in a polygynous clade (Aves: Pipridae). Am. Nat. 149:668-692. Rosenthal, R., R. L. Rosnow, and D. B. Rubin. 2000. Contrasts and effect sizes in behavioural research. Cambridge Univ. Press, Cambridge, U.K. Rowe, M., C. L. Baube, E. Loew, and J. Phillips. 2004. Optimal mechanisms for finding and selecting mates: how threespine stickleback (Gasterosteus aculeatus) should encode male throat colors. J. Comp. Physiol. A 190:241-256. Spaethe, J., J. Tautz, and L. Chittka. 2001. Visual constraints in foraging bumblebees: flower size and color affect search time and flight behavior. Proc. Natl. Acad Sci. USA 98:3898-3903. Thèry, M., and J. Casas. 2002. Predator and prey views of spider camouflage. Nature 4 15:133. Ugalde, J. A., B. S. W. Chang, and M. V. Matz. 2004. Evolution of coral pigments recreated. Science 305:1433. Uy, J. A. C., and G. Borgia. 2000. Sexual selection drives rapid divergence in bowerbird display traits. Evolution 54:273-278. Uy, J. A. C., and J. A. Endler. 2004. Modification of the visual background increases the conspicuousness of golden-collared manakin displays. Behav. Ecol. 15:1003-1010. Villa, P., M. D. Bedmar, and M. Bar6n. 1991. Studies on rod horizontal cell s-potential in dependence of the dark/light adapted state: a comparative study in Cyprinus carpio and Scyliorhinus canicula retinas. Vision Res. 31:425-435. Vorobyev, M., and D. Osorio. 1998. Receptor noise as a determinant of colour thresholds. Proc. R. Soc. Lond. B 265:351-358. Vorobyev, M., D. Osorio, A. T. D. Bennett, N. J. Marshall, and I. C. Cuthill. 1998. Tetrachromacy, oil droplets and bird plumage colors. J. Comp. Physiol. A 183:621-633. Vorobyev, M., R. Brandt, D. Peitsch, S. B. Laughlin, and R. Menzel. 2001. Colour thresholds and receptor noise: behaviour and physiology compared. Vision Res. 41:639-653. Wyszecki, G., and W. S. Stiles. 2000. Color science, concepts and methods, quantitative data and formulae. 2d ed. John Wiley and Sons, New York. Zar, J. H. 1999. Biostatistical an*lysis. 4th ed. Prentice-Hall, Upper Saddle River, NJ. Corresponding Editor: P. Phillips