We bred all animals in captivity under a 12:12-hour light-dark cycle in the laboratory. Each mantis consumed one appropriately sized feeder cockroach every two days. We used only adult females from all three species because they are larger, longer-lived, and more reliably maintained in captivity, providing higher-quality electrophysiological signals. The laboratory was maintained at a constant 23 °C.

We used a spectrometer (QE Pro, Ocean Optics) to measure the spectral reflectance of adult female mantis wings over a 350–650 nm wavelength range. Prior to measurements, we calibrated the spectrometer using polytetrafluoroethylene (PTFE) as a white standard. We positioned a full-spectrum light source (LS-1, Ocean Optics, Inc., Dunedin, FL, USA) at 45° toward the wing, with the spectrometer detecting probe 1 cm away, perpendicular to the wing surface at the point of measurement. Each measurement had an exposure time of 2 s, and reflectance curves represented the average normalized values from 3 randomly chosen points on each wing.

We measured the mantis body region primarily covered by the wings to quantify their morphology. Specifically, we measured three anatomical dimensions: (1) mesothoraco-abdominal length, defined as the distance from the anterior edge of the mesoscutum to the posterior tip of the abdomen, (2) maximum body width, measured as the widest lateral extent of the mesothoraco-abdominal region, and (3) maximum body height, determined as the greatest dorsoventral distance within this region. To minimize variation due to abdominal swelling in gravid individuals, all measurements were performed 6–10 days after adult emergence, before the significant development of ovaries. We then calculated and plotted body shape ratios of height-to-width and mesothoraco-abdominal length-to-width.

We adopted our electrophysiological recording method from Frank et al. (2012), with the equipment apparatus replicated from Qian and Frank (2024). ERGs measure the collective electrical response of photoreceptor cells to light (Autrum 1948), providing a direct estimate of spectral sensitivity. Unlike spectrophotometric techniques that measure isolated photopigments or rhabdoms, ERGs incorporate the natural pre-retinal filtering effects of screening pigments and other ocular tissues, thus reflecting the spectrum actually received by the receptors. Mantises were dark-adapted for at least three days prior to recording, and all animal preparations were conducted under dim red light to avoid unintended visual adaptation.

We secured animals inside a Faraday cage covered with a light-proof sheet in a dark room. With a micromanipulator and a dissecting microscope (Olympus), we inserted a glass-insulated tungsten microelectrode (Frederick Haer) into the dorsal region of the eye, resting its tip directly above the photoreceptor layer. We placed a silver chloride reference electrode on top of the untested eye to eliminate electrical noise from the body. After placing the electrode, we allowed mantises to dark adapt for 1–2 h.

We delivered stimuli using a 150 W quartz halogen lamp source coupled to a monochromator (CM110, Spectral Products) and guided via a bifurcated light guide cable (EXFO). We positioned the end of the light guide approximately 1 cm from the eye surface, oriented perpendicularly to the insertion site, which illuminated roughly 10% of the eye’s visual field. We amplified recording signals with an X Cell-3 Microelectrode Amplifier (FHC, Inc.) with a high impedance probe. We set the amplification level to 2000X and applied filters between 1 and 1000 Hz.

At the beginning of the recording, we flashed a 490 nm wavelength at an initially dim irradiance of 108 photons cm− 2 s− 1, then gradually increased the irradiance until the flash-elicited response stabilized at 50µV. Measurements began once the response to the adjusted test flash remained constant for one hour.

In the dark-adapted experiment, we stimulated the eye with monochromatic light flashes, adjusting flash irradiance to elicit a criterion response of 100 µV (or higher if the response was indistinguishable from background noise). Each flash lasted 0.1 s, followed by an interval of at least one minute before the next flash. Wavelengths ranged from 350 to 650 nm in 10 nm increments, presented in random order. Standard test flashes followed each wavelength presentation to assess the dark-adapted state of the eye. If the response to the standard test flash showed any reduction in amplitude (indicating slight light adaptation), we paused until the test flash response returned to baseline within a tolerance of ± 2 µV, ensuring full recovery to the dark-adapted state.

To investigate the possibility of multiple photoreceptor classes or varying expressions of visual pigments, we conducted chromatic adaptation experiments subsequent to the dark-adapted test. For these, we directed an adapting light to the eye via the secondary path of a bifurcated light guide. The source of the adapting light was a white halogen lamp (LS-1, Ocean Optics, Inc., Dunedin, FL, USA), filtered at wavelengths corresponding to the spectral sensitivity peaks previously identified (530 nm with 10 nm FWHM, 415 nm with 30 nm FWHM, and 380 nm with 15 nm FWHM; bandpass filters). We focused on detecting changes in the shape of the spectral sensitivity curves or the emergence of new peaks, which would indicate the presence of additional photoreceptor types. We then calculated the inverse of the irradiance required to evoke the criterion response at each wavelength, subsequently normalizing these values to that of the wavelength of maximum sensitivity.

We modeled spectral sensitivities by fitting visual pigment templates to the normalized ERG data, following Stavenga et al. (1993) and subsequent applications in arthropod spectral sensitivity studies (Jessop et al. 2020). For each adaptation condition, we fitted the normalized spectral sensitivity curves using a least-squares minimization in Microsoft Excel Solver. We assumed a photoreceptor length of 344 μm and set the specific absorbance at 0.00007 μm⁻¹. For each species, the peak wavelengths (λmax) of the underlying pigments were held constant, and only the relative contributions of the pigments were allowed to vary across adaptation conditions (dark, 380 nm, 415 nm, 530 nm). We tested both two- and three-pigment models and retained the simpler model unless adding a third pigment yielded a clear reduction in residuals. We obtained the relative contributions of pigments by normalizing fitted amplitudes to sum to 1.