Spider collection
Thirty male and seven female Saitis barbipes were collected as subadults and adults from a submediterranean forest in Osp, Slovenia (45°34′47.7″N 13°51′20.0″E), in June 2019 and May 2021. This forest community is classified as Aristolochio luteae-Quercetum pubescentis (Poldini 2008), occurs on calcareous bedrock, and is composed of mixed stands of coppice and seed source trees, with a high tree level and well-developed shrub and herb layers (Marinček & Čarni 2002). Most spiders were encountered in partial shade at the forest edge, in the herb layer among leaf litter and large rocks. The spiders were spotted by eye and collected using a simplified aspirator. After species identification was verified in the laboratory of ZRC SAZU, Slovenia, thirty-one spiders were transferred to the University of Hamburg, Germany, for lens transmittance measurements (N = 3 males and 3 females) and multispectral imaging (N = 24 males and 1 female), and the remaining spiders were transferred to the University of Cincinnati, USA, for microspectrophotometry (N = 3 males and 3 females).
Microspectrophotometry
Microspectrophotometry (MSP) was used on cryosections of the principal eye retinas of S. barbipes to measure photoreceptor absorbance profiles. Spiders were dark-adapted overnight before cryosectioning. Sample preparation, cryosectioning, and MSP measurement occurred under dim red light to avoid bleaching of retinal tissues. The legs and opisthosoma of the spider were cut off and the cephalothorax was flash-frozen in Tissue Plus OCT Compound (Fisher Healthcare, Houston, Texas). The embedded cephalothoraxes were cryosectioned in the coronal plane using a thickness of 13 μm on a Leica CM1860 cryostat at − 20 °C. All sections that contained the principal eye retinal tissue were kept and inspected in the MSP. Prior to measuring, sections were placed between two glass cover slips (22 × 22 − 1 Fisherfinest, Fisher Scientific, Pittsburgh, Pennsylvania) and immersed in mineral oil (Fisher Scientific, Fair Lawn, New Jersey) surrounded by a ring of silicone grease (Dow Corning Corporation, Midland, Michigan).
The absorbance of individual photoreceptor cells was measured between 300 and 700 nm using a custom-built single beam, scanning MSP with a 32 × Ultrafluar objective and a 32 × Ultrafluar condenser (Carl Zeiss, Germany). The light source was a xenon arc lamp (XBO 75 W/2, Osram Sylvania, Wilmington, MA) and it was dispersed from 300 to 700 nm in steps of 1 nm using a monochromator (H10 UV, Jobin Yvon Instruments, SA, Edison, NJ). First, a reference scan was measured in a clear area away from the section but within the mineral oil and subsequently subtracted from measurements to account for light absorption by the preparation itself (e.g., mineral oil and coverslips). Second, measurements were performed in areas with photoreceptor cells following this procedure: a photoreceptor was measured, then photobleached for 30 s using white light, and then re-measured. The difference between the pre-bleach spectrum and the photobleached spectrum was used to confirm the presence of photopigments. The peak sensitivities of the photoreceptors were then estimated by fitting pre-bleach absorbance curves to visual templates (Govardovskii et al. 2000) representing a range of possible values for the wavelength of peak sensitivity (i.e., alpha-peak lambda max values; see Govardovskii et al. 2000), with the best fit template (and associated lambda max) identified using least-squares model comparisons. On rare occasions, noise in the short wavelength region of the absorbance spectra (i.e., where the MSP signal-to-noise ratio is lowest) required manual adjustment to reduce the impact of this noise on model estimations. After being measured in the MSP, each section was visually inspected for the presence of possible intraretinal filters (as have been found in H. pyrrithrix; see Zurek et al., 2015) under a Leica ICC50 HD microscope using a 40 × HI-Plan objective and bright light.
Lens transmittance
Spiders were killed by over-anesthesia with CO2. Each principal eye lens was excised and rinsed in spider Ringer’s solution (190 NaCl, 2 KCl, 4 MgCl2, 4 CaCl2, and 1 Na2HPO4 (units in mmol/l) (Schartau and Leidescher 1983)) and kept in this solution until measurement. The lens consisted of a rigid cornea contiguous with the exoskeleton, and a softer, internal lens bathed in vitreous fluid. In intact spiders, these two optical components were attached to one another at the point at which the corneal edge meets the exoskeleton. This attachment was preserved during lens excision such that the transmittance of both optical components was measured together. For each spider, the entire process, from the beginning of dissection to the measurement of both principal eye lenses, was completed within about 1 h. One male lens was accidentally punctured during excision and was not measured.
The illumination light path consisted of a pulsed-xenon light source (PX-2) directed through a 115 µm extreme solarization-resistant optical fiber (QP115-2-XSR) and collimating lens (74-UV). The spider lens was placed in the center of the light path on a custom-built lens holder, which consisted of a 0.5-mm-thick sheet of black plastic with a 100 µm hole drilled through it. The upper part of the hole was widened and cup-shaped to cradle the lens over the 100 µm aperture in such a way that only light passing through the lens as it would in a living spider made it through the lens and into the measurement path. A small drop of spider Ringer’s solution was placed in the hole such that the part of the lens that would normally be bathed in vitreous fluid was sitting in solution, whereas the air-facing portion was dry. The measurement path consisted of a collimating lens (74-UV) connected to a 1000 µm solarization-resistant optical fiber (QP1000-2-SR), which directed the light to a spectrometer (QEPRO) (Suppl. Figure 1). All optical fibers, collimating lenses, and the spectrometer were sourced from Ocean Optics (Ostfildern, Germany). The placement of the lens in its holder and its alignment with the beam path was checked through an obliquely mounted microscope. The spider lens was measured first, followed by a reference measurement with the lens and Ringer’s solution removed from the plastic lens mount. Transmittance was calculated by dividing the first measurement by the second.
Multispectral imaging and visual modeling
Each spider was killed by placing it in an Eppendorf tube in a − 80 °C freezer. Such flash freezing did not visibly affect the appearance of spider colors and has previously been found not to affect the chroma and brightness of the red hair and cuticle of the salticid Lyssomanes viridis (CT, unpublished data). Shortly before the spider was to be photographed, it was removed from the freezer and the ventral surface of its prosoma was glued to the head of a nail. The nail was stuck into a flat piece of styrofoam covered with undyed brown paper having a reflectance spectrum similar to that of leaf litter. The spider’s ornamented third pair of legs were posed in a display position using bent insect pins. For photography, the mounted spider was placed in front of a 20% reflective 2 inch fluorilon gray standard (Avian Technologies, New London, NH, USA), which reflects light evenly across the UV–VIS spectrum.
Images were taken in a dark room under a xenon light source (XE-140BF, Seric Ltd., Tokyo, Japan) closely mimicking the spectrum of natural daylight. The lamp’s irradiance spectrum was measured with a spectrophotometer (QE Pro) fitted with an extreme solarization-resistant (XSR) fiber optic cable and cosine corrector (CC-3-UV) that had been calibrated to absolute light intensities using a factory-calibrated deuterium and tungsten halogen light source (DH-3P-CAL), all sourced from Ocean Optics (Ostfildern, Germany) (Fig. 1). The lamp was tilted 60° from vertical and the spider was placed in the center of the cone of light emanating from it.
Images were taken using the multispectral camera and five of the bird-based optical filters described in Tedore and Nilsson (2019). Two new filters were added to the system, which enabled the use of the computational filter technique described in Tedore and Nilsson (2021) to closely mimic S. barbipes spectral sensitivities (Figs. 2 and 3). This technique takes a weighted sum of pre-existing camera filters to generate new spectral sensitivities. First, the spectral sensitivity of each real camera channel i was calculated as:
$${{F}_{i}(\lambda )=S}_{\mathrm{sensor}}(\lambda ){ T}_{\mathrm{lens}}(\lambda ) {T}_{\mathrm{IRblock}}(\lambda ) {T}_{\mathrm{filter},i}(\lambda )$$
(1)
where Ssensor(λ) is the spectral sensitivity of the camera sensor, Tlens(λ) is the transmittance spectrum of the camera lens, TIRblock(λ) is the transmittance spectrum of an infrared blocking filter mounted on the front of the lens, and Tfilter,i(λ) is the transmittance spectrum of camera filter i. Next, we used constrained linear least squares to solve for a set of seven nonnegative coefficients to multiply by each camera channel such that the sum of the seven channels would generate a spectral shape matching each of S. barbipes’ spectral sensitivity curves. Such computational filters were generated by solving for the set of seven coefficients that best satisfies the following equation, while constraining the solution to prevent negative coefficients:
$$R_j(\lambda)T_{\mathrm l}(\lambda)=aF_1(\lambda)+bF_2(\lambda)+{cF}_3(\lambda)+{dF}_4(\lambda)+{eF}_5(\lambda)+{fF}_6(\lambda)+{gF}_7(\lambda),$$
(2)
where Rj(λ) is the spectral sensitivity of photoreceptor j (generated by the template of (Govardovskii et al. 2000)), and Tl(λ) is the transmittance spectrum of the Saitis barbipes lens.
To get an impression of the spider color pattern as a whole, we took photos of entire spiders posed in a display stance (Fig. 4). To get the spider to fill the whole frame, extension tubes (Kenko Extension Tube Set DG, Kenko Tokina Co., Ltd., Tokyo, Japan) were inserted between the 60 mm lens and the filter wheel of the camera. Photos were taken at different focus depths and then combined in Adobe Photoshop (Adobe Inc., San Jose, CA, USA) using the auto-align and auto-blend functions. To obtain extreme close-up images of individual male body parts for color patch selection and analysis, more extension tubes were added such that individual segments filled the entire frame (Fig. 5). No focus stacking was used in these latter photos. Close-up images were taken of all segments of both the left and right third leg pair. If any part of the spider was found to have been damaged or to have moved slightly between photos taken through different filters, these body parts were excluded from further analysis. Over- and under-exposed pixels were also excluded from analysis.
The camera sensor has a linear response to light, so no non-linearity corrections were needed. Dark noise was obtained from several columns of pixels on the camera sensor that do not receive any light and was subtracted from all pixels that receive the image. Each pixel value of real and computational filter images represented the quantum catch by the simulated photoreceptor class at a single point in space. To adapt quantum catches to the spectral distribution of the illuminant (i.e., convert to relative quantum catches) (Vorobyev et al. 1998), each pixel value was normalized by the mean pixel value of a large selection of the gray standard located behind the spider.
The pixel locations corresponding to the gray standard and color patches of interest on the spider were selected interactively in MATLAB using the “roipoly” function. Nine distinct combinations of colors and structures were identified from avian false-color images and selected on the prosoma and third pair of legs, which are raised and waved during courtship. These included red hair and cuticle, black hair and cuticle, white hair and cuticle, iridescent UV cuticle, orange hair, and beige hair. We used avian false-color images to make selections, since the greatest number of human-discriminable colors could be seen in these photos. For categorization purposes, we use human color names to describe how colors appear in avian visible-light false-color images (i.e., RGB = LMS; see “Methods”: Visualizations and Fig. 5) but acknowledge that these color names reflect what humans see, not animals. As many as possible, up to a maximum of ten, selections of each combination of color and structure (indicated in Table 1) were selected. The position of the fourth segment had to be shifted slightly between shots in order to obtain images in which the UV iridescence was and was not visible.
Table 1 Combinations of colors and structures selected on the prosoma and each of the five segments of the third leg pair It should be noted that S. barbipes should only be able to resolve individual hairs at close viewing distances. Even if we assume that S. barbipes possesses the best spatial acuity ever reported in a salticid, i.e., the 0.04° interreceptor angle reported in Portia fimbriata (Williams and McIntyre 1980), the spider would need to be situated at a viewing distance of ~ 4 mm in order to see individual hairs in sharp focus. At greater distances, the colors of the hair and underlying cuticule would blend together into an intermediate color. In order to estimate how large of an effect the dark cuticular colors surrounding the red, orange, and beige hairs had on the degree of color contrast between these and adjacent black color patches when viewed at greater distances, we additionally selected large, standardized patches on the tibia, metatarsus, forehead, and dorsal prosoma. Color patch boundaries were defined using predetermined landmarks, such as cuticular contours and the edges of eyes. A total of 4761 small patches including only hair or cuticle were selected (Table 1) and a total of 131 large patches including both hair and cuticle were selected.
Pixel locations were saved and later used to calculate the median relative quantum catch P of each photoreceptor for each selected color patch. These medians were then converted to non-linear receptor excitation values following Naka and Rushton (1966),
Within-channel differences in receptor excitation E between color patches were then calculated for each of the spider photoreceptor classes i,
$$\Delta {E}_{i}={E}_{i, 1}-{E}_{i, 2}$$
(4)
Finally, receptor noise-limited (RNL) color contrasts ΔS (1) between color patches on the same leg and (2) between color patches on the prosoma and each leg were calculated by plugging ΔE values into the RNL color contrast equation (Vorobyev and Osorio 1998; Vorobyev et al. 1998). For trichromats, this was calculated as
$$\Delta S=\sqrt{\frac{\omega_1^2{(\Delta E_3-\Delta E_2)}^2+\omega_2^2{(\Delta E_3-\Delta E_1)}^2+\omega_3^2{(\Delta E_1-\Delta E_2)}^2}{{(\omega_1\omega_2)}^2+{(\omega_1\omega_3)}^2+{(\omega_2\omega_3)}^2}},$$
(5)
and for dichromats, as.
$$\Delta S=\sqrt{\frac{{(\Delta E_1-\Delta E_2)}^2}{\omega_1^2+\omega_2^2},}$$
(6)
where ωi is the standard deviation of the noise in photoreceptor channel i and is calculated as
$$\omega_i=\frac\upsilon{\sqrt{\eta_i}},$$
(7)
where υ is the noise in a single photoreceptor and ηi is the relative number of photoreceptors in photoreceptor class i. The noise in a single photoreceptor has only been measured in one invertebrate, the honeybee Apis mellifera, 0.074 (Vorobyev et al. 2001), which we use as an approximate estimate for salticids. Relative photoreceptor numbers in salticids are also unknown, so we used the relative numbers sampled during microspectrophotometry as an approximation, i.e., 1:7 for the dichromatic ultraviolet:green (U:M) model and 0.4:0.6:7 for the trichromatic ultraviolet:blue:green (U:S:M) model.
Visualizations
The grayscale images in Figs. 4 and 5 are visualizations of the relative excitation of each photoreceptor class on a pixel-by-pixel basis. To generate such images, we normalized receptor excitations Ei by the maximum receptor excitation value across all color channels of both visual systems (spider and bird). The same values were plugged into each of the R-, G-, and B-channels of the computer display in order to obtain a grayscale image. This gives receptor excitation images in which bright pixels correspond to high excitation and dark pixels to low excitation.
False-color images in Figs. 4 and 5 help one get an approximate sense of the color contrasts that may be visible to different visual systems. Contrasts in the UV most resemble those in the blue, so for tetrachromatic visual systems (i.e., birds), we alternately plugged the S- and U-cone excitation images into the B-channel of the computer display, while keeping the L- and M-cone excitation images plugged into the R- and G-channels of the computer display, respectively. To visualize a trichromatic visual system, each of the different photoreceptor excitation images was plugged into a unique channel of the RGB display. To visualize a dichromatic visual system, one photoreceptor excitation image was plugged into one channel, and the other into two channels, of the RGB display to avoid a strongly tinted image.