Ultraviolet filters in stomatopod crustaceans: diversity, ecology and evolution

Stomatopod crustaceans, commonly referred to as mantis shrimp, are well known for charismatic, aggressive behavior (Caldwell and Dingle, 1975; Dingle and Caldwell, 1969) and their spectacularly elaborate visual systems (Cronin and Marshall, 1989a). Many species possess at least 16 spectral classes of photoreceptors in their retinas. Included in these are eight color-specialist photoreceptors sensitive to narrow ranges of ‘human-visible’ light (Cronin and Marshall, 1989a,b; Marshall, 1988), three polarization receptors sensitive to linearly or circularly polarized light (Chiou et al., 2008; Cronin et al., 1994b; Kleinlogel and Marshall, 2006; Marshall, 1988; Marshall et al., 1991a) and at least five receptors sensitive to various spectral ranges of ultraviolet (UV) light (Kleinlogel and Marshall, 2009; Marshall and Oberwinkler, 1999). Underlying these diverse visual sensitivities is an array of optical and retinal structural modifications (Horridge, 1978; Marshall et al., 1991a; Schiff et al., 1986), the expression of a great number of opsins resulting in the most visual pigments yet described in a single eye (Cronin and Marshall, 1989b; Cronin et al., 1993; Porter et al., 2009, 2013), and the tuning of spectral sensitivity via serial filtering effects due to more distal visual pigments as well as photostable colored pigments (Cronin and Marshall, 1989a; Cronin et al., 1994a,b, 2014; Marshall, 1988; Marshall et al., 1991b; Porter et al., 2010).

Stomatopod apposition compound eyes are subdivided into three regions. Most of the eye consists of the dorsal and ventral peripheral regions, which are bisected horizontally by a multi-row midband (Harling, 2000; Manning et al., 1984). The peripheral regions are composed of a homogeneous array of ommatidia presumably responsible for spatial vision. These ommatidia are organized similarly to those of other malacostracan crustaceans, with corneal and crystalline cone optical elements focusing light onto the retinal photoreceptors. Each ommatidium contains a proximal main rhabdom photoreceptor, produced by a cylindrical arrangement of retinular cells one to seven (R1–7), and a distal photoreceptor produced by retinular cell eight (R8). In stomatopods, R1–7 receptors in peripheral ommatidia are maximally sensitive to blue–green light, while the R8 is maximally sensitive to UV light. The notable diversification and specialization of stomatopod photoreceptors is found in the six midband rows of stomatopods from the superfamilies Gonodactyloidea, Lysiosquilloidea, Pseudosquilloidea and Hemisquilloidea (superfamilies as suggested in Porter et al., 2010). Here, the relatively simple ommatidial organization of the peripheral regions is modified into the most elaborate photoreceptor array in nature.

The R8 photoreceptors in the midband of gonodactyloids are responsible for the remarkable spectral diversity of the stomatopod UV visual system, with four distinct spectral classes (in addition to the single R8 spectral class in the periphery) having been identified by electrophysiology (Kleinlogel and Marshall, 2009; Marshall and Oberwinkler, 1999). In Neogonodactylus oerstedii, these R8s only contain two visual pigments, and the full spectral diversity is achieved by optical filtering effects from four distinct UV-specific filters located in the crystalline cones (Bok et al., 2014). The surprising localization of these filters in the crystalline cones was revealed by the fluorescent properties of three of the filter types found in rows 3–6, which emit blue, cyan or green light when stimulated with strong UV illumination. These filters are composed of mycosporine-like amino acids (MAAs), compounds commonly associated with photoprotection in marine organisms (Carreto and Carignan, 2011; Shick and Dunlap, 2002). In eukaryotes, MAAs are typically acquired through their diet (Hylander and Jephson, 2010; Newman et al., 2000) and subsequently modified for use as photoprotectants or, in the case of stomatopods, as optical filters.

Optical filters function by selectively absorbing and transmitting certain wavelengths of light, shaping the spectrum that reaches a photoreceptor, and thus altering its spectral sensitivity. Optical filtering is common in nature, with sets of multiple colored filters having been described in reptiles (Walls, 1942), birds (Goldsmith et al., 1984), lungfish (Bailes et al., 2006), butterflies (Arikawa and Stavenga, 1997; Arikawa et al., 1999a,b) and the main rhabdom photoreceptors of stomatopods (Cronin et al., 1994a; Marshall, 1988; Marshall et al., 1991b), among others (see Douglas and Marshall, 1999, for a review of optical filters in animals).

In N. oerstedii, five spectral classes of UV-specialist R8s are produced by the pairwise combination of only two visual pigments (wavelength of maximum absorption, λ_max, at 334 and 383 nm, respectively), with four UV-specific optical filters in the crystalline cones (Bok et al., 2014). One type of UV filter, found in the crystalline cones of midband rows 1 and 2, is composed of the MAA porphyra-334, with a λ_max of 334 nm. Midband rows 3 and 4 express similar but distinct filter pigments, with λ_max at 372 and 375 nm, respectively. These filters, which are responsible for the distinct fluorescent emission, do not match any currently characterized MAAs, and likely represent novel molecular structures. Finally, midband rows 5 and 6 contain identical notch filters composed of two pigments, mycosporine-glycine with a λ_max of 308 nm and a second weaker optical density (OD) pigment with a λ_max of 388 nm. This second pigment likely produces the green fluorescence visible in rows 5 and 6, and is also currently uncharacterized. The 334 nm visual pigment expressed in the R8s of midband rows 2 through to 6 is tuned to longer wavelengths by the pigments in the crystalline cones of rows 2, 5 and 6 (long-pass filtering), and to shorter wavelengths by the pigments of rows 3 and 4 (short-pass filtering).

Some stomatopod species can also have between two and four intrarhabdomal long-wavelength colored filters that act on the main rhabdom R1–7 photoreceptors in midband rows 2 and 3 (Cronin et al., 1994a; Marshall, 1988; Marshall et al., 1991b; Porter et al., 2010). Some species even tune the absorption properties of these filters by varying the lengths of the filters or altering the composition of the pigments they contain in order to best suit their spectral environment (Cheroske et al., 2006, 2003; Cronin and Caldwell, 2002; Cronin et al., 1994b, 2000, 2001). Considering the high degree of variability in the absorbance characteristics of these long-wavelength filters, we were curious to learn how the UV filters vary across species. Are these filters pervasive throughout the order, and can their physiological functions offer any clues to their evolution and ecological function? To these ends we set out to characterize the UV filters in a broad sampling of stomatopod taxa. We initially looked for UV-stimulated autofluorescence in the midband that belies the presence of certain UV filters. Of the 28 species we observed, only six lacked fluorophores in the midband, while the remainder displayed a beautiful diversity of fluorescent colors and patterns (Fig. 1). The presence and variability of these fluorophores from species to species suggested that UV filters in stomatopods are physiologically complex and ecologically important.

Here, we present a spectral survey of the UV filters in 21 stomatopod species. We characterized the number, optical properties and variability of these filters, both within and between species. We used the most current phylogeny of the Stomatopoda to reconstruct the evolution of the UV filters, and we analyzed their ecological significance in species with differing habitats and lifestyles. Our examination of these UV filters suggests the presence of an even greater level of spectral diversification in stomatopod visual systems than earlier described, making the most spectrally elaborate eye in nature even more complex than previously thought.

Using microspectrophotometry (MSP) and imaging spectrometry, we examined the crystalline cone absorbance spectra from 21 species of stomatopods representing seven families from all four superfamilies that have elaborated six-row midbands: Gonodactyloidea, Lysiosquilloidea, Pseudosquilloidea and Hemisquilloidea (Table 1, Fig. 2A). Also, the presence or absence of the fluorescent UV filters in seven additional species was determined by observation of UV-stimulated autofluorescence (supplementary material Table S1). Our survey indicated the presence of four primary filter UV types found in the midbands of stomatopods, each associated with particular rows of the midband. We refer to these filters as UVF1 (rows 1 and 2), UVF2 (row 3), UVF3 (row 4) and UVF4 (rows 5 and 6) (Fig. 2A, top left panel). Of the 21 species examined by MSP, 19 had at least one significant UV-absorbing pigment in the midband crystalline cones. Of these, eight had a single filter type, one had two filter types, two had three filter types, and eight had all four. The UVF3 filter is present in all 19 species that express any UV filters in the crystalline cones.

The absorbance spectrum of each primary UV filter type was compared across all species. We found that there was significant variation among species (Fig. 2B), but within a species each filter type was fairly constrained spectrally, with the λ_max rarely exceeding a standard deviation of ±2 nm between individuals (Fig. 2C). The λ_max of UVF1 varied by 10 nm among different species, UVF2 by 18 nm, UVF3 by 26 nm and UVF4 by at least 8 nm. For UVF1, filters of all but two species were very similar in shape, resembling the MAA porphyra-334 previously identified in these rows in N. oerstedii (Bok et al., 2014). A similar case was found with UVF4, where most examples of this filter appear to be composed of mycosporine-glycine, with filters in only four species having a shorter wavelength absorbance spectrum. In UVF2 and UVF3 there was no single dominant spectral absorbance curve. These pigments instead have a fairly stereotypical shape that is shifted throughout the spectral range measured for each filter.

The character state of the four primary UV filter types in stomatopods was reconstructed using the most recently published molecular phylogeny of the order (Porter et al., 2010, 2013), but including several additional species procured for this survey (supplementary material Table S2). This new tree agreed with the familial and superfamilial relationships reported in Porter et al. (2010). The number of UV filters found in each species based on the presence or absence of the four primary UV filter classes, UVF1–4, was used to reconstruct the ancestral number of filters across the phylogeny (Fig. 3). The character state reconstruction indicated that at least a single UV filter – probably akin to UVF3 – was present in the common ancestor of all included species. Based on the reconstruction of the number of filters in each species (Fig. 3A) there have since been at least three independent events of gains of additional filter types, and three of loss of complexity in the UV filter set. UV filters have apparently been lost altogether in lysiosquilloids, but are present in the three other included superfamilies. The UVF2 and UVF4 filters are only found in gonodactyloids, suggesting an elaboration of the ancestral filter set in this superfamily (Fig. 3B).

We examined the relationship between the habitat depth of stomatopod species and spectral properties of their filters. By plotting the λ_max of each filter versus that species’ maximum habitat depth (Fig. 4A), we found the λ_max of UVF2 and UVF3 short-pass filters tended to shift towards longer wavelengths as the species’ habitat depth increased. This logarithmic correlation was significant even when corrected for similarities in values due to species’ phylogenetic relatedness using phylogenetic generalized least squares (PGLS) analysis (UVF2 adjusted R²=0.45, P=0.0281; UVF3 adjusted R²=0.33, P=0.0059). In order to determine whether filter absorption varied with habitat depth among individuals of a species, we examined the UV filters of Haptosquilla trispinosa captured either from the intertidal zone or from a depth of 20 m. We found that the absorbance spectrum and OD of UVF3 (the only primary UV filter expressed in this species) in the two populations was nearly identical (Fig. 4B). We also carried out an experiment where H. trispinosa individuals were captured from intertidal reef flats and either kept in natural light or complete darkness for 14 days without food. We found that the filters possessed the original OD in both groups, and were spectrally identical (Fig. 4C).

Our survey uncovered additional filter types in crystalline cones beyond the four primary filters already discussed in the midband (Fig. 5). In the protosquillids (H. trispinosa, Haptosquilla glyptocercus, Chorisquilla tweediei and Chorisquilla hystrix) and the distantly related pseudosquilloid Pseudosquilla ciliata, midband rows 2 and 3 dimly fluoresce cyan–green, similar in emission color to UVF3 (Fig. 5A and Fig. 1I,J). There is a weakly absorbing pigment in these rows with a λ_max between 378 and 391 nm (Fig. 5A, Table 1). In a more dramatic case, in the first 10 rows of the ventral periphery adjacent to the midband, Odontodactylus havanensis expresses a very dense filter akin to UVF1 (Fig. 5B, Table 1). Finally, several gonodactylids (N. oerstedii, Neogonodactylus wennerae, Gonodactylus smithii and Gonodactylus chiragra) were found to have blue-emitting fluorophores in the first few rows of crystalline cones in the dorsal periphery adjacent to the midband (Fig. 5C, large arrowhead). The pigments in these cones were difficult to measure, but they had a λ_max near 360 nm (Fig. 5C). In G. smithii, this blue-fluorescing pigment was found not only in the crystalline cones of the dorsal periphery adjacent to the midband but also in the crystalline cones in the dorsal-most region of the eye (Fig. 5C, small arrowhead).

In order to look at the developmental origin of these pigments, we examined an individual Haptosquilla postlarva, which is one of the developmental stages where the adult retina is actively being built alongside the degenerating larval retina (Cronin et al., 1995; Feller et al., 2015; Jinks et al., 2002). Under 375 nm illumination, the postlarva exhibited UV-stimulated autofluorescence in the adult eye midband consistent with the row 4 UVF3 fluorophore typical of adult Haptosquilla (Fig. 6). The dim secondary filter fluorophore in rows 2 and 3 was also present. These fluorescent pigments appeared to generally aggregate over the degrading larval retina at the advancing edge of the developing adult retina, before being sequestered into specific crystalline cones of the adult midband.

As suggested by the variety of UV-excited fluorescent emission spectra observed in stomatopod eyes (Fig. 1), a great diversity of strong, optically effective UV filters exists in the crystalline cones of stomatopods (Fig. 2, Table 1). Based on these observations, we classified the UV filters into four ‘primary’ filter types. In certain species, we also observed ‘secondary’ UV filter pigments in the midband and periphery, which will be discussed in detail below. The four primary UV filters are referred to here as UVF1–4.

UVF1, found in midband rows 1 and 2, usually has an absorption spectrum similar to porphyra-334, with a λ_max near 334 nm, and does not fluoresce at human-visible wavelengths. UVF1 was found in P. ciliata (where it atypically appears only in row 1), Odontodactylus scyllarus and most gonodactylids excepting G. chiragra and N. wennerae. UVF1 is often relatively weakly absorbing compared with the other primary filters, with maximum densities through 200 μm crystalline cone sections typically under OD 1.0, ranging from 0.35 to 1.53. In stomatopods this filter likely acts as a long-pass filter in both midband rows (see Materials and methods for an explanation of assumptions regarding UV visual pigment absorbances in stomatopods).

UVF2 occurs in midband row 3 and has λ_max values between 358 and 376 nm. UVF2 fluoresces blue when stimulated with 375 nm UV light (Fig. 1A–F). In most species, this filter is likely produced by an uncharacterized MAA similar to pigments that have been observed in the lens of teleost fish (Thorpe et al., 1993). In Gonodactylus platysoma and Gonodactylellus viridis, however, UVF2's absorbance spectrum is very similar to that of the MAA palythene (λ_max=360 nm). UVF2 is found exclusively in gonodactyloids, and primarily in the family Gonodactylidae, where all species examined by MSP had this pigment. However, Gonodactylellus annularis, which was only observed by fluorescent emission, lacks a fluorophore in any midband row suggesting that UVF2 is absent in that species (supplementary material Table S2). In most species, UVF2 occurs at maximum densities exceeding 1.0 with values as high as OD 2.89 (through 200 μm crystalline cone sections), making for prodigious short-pass filtering effects. Odontodactylus scyllarus lacks a typical UVF2, and instead has a deep-UV-absorbing filter in row 3, more akin to UVF4.

UVF3 is found in midband row 4 and has a λ_max between 361 and 387 nm. UVF3 fluoresces cyan or light green when illuminated with 375 nm UV light (Fig. 1A–J). Like UVF2, this filter is probably based on an uncharacterized MAA pigment. UVF3 is the most pervasive filter type in stomatopods, and is present in every species that has any UV filters in the crystalline cones (Fig. 2A). Like UVF2, it also appears to function as a strong short-pass filter, and has ODs as high as 2.80 in some species. Interestingly, though the UVF2 and UVF3 absorbance spectra often overlap among different species, the emission spectra of the two filters remain separate (supplementary material Fig. S1), suggesting that two distinct pigment types produce these filters.

UVF4 exists in rows 5 and 6 of the midband and has a λ_max between 301 and 309 nm. In most species, UV4F seems to be primarily composed of mycosporine-glycine (λ_max=310 nm), but Gonodactylellus affinis, Gonodactylaceus ternatensis, Gonodactylaceus falcatus and O. scyllarus all have shorter-wavelength-absorbing UVF4 pigments. The primary UVF4 pigment does not fluoresce at visible wavelengths under 375 nm illumination, but midband rows five and six typically have a fluorescent secondary pigment absorbing much more weakly at λ_max values between 376 and 389 nm and fluorescing light blue or green (Fig. 1A–E,K). In our survey, the green fluorescence from the secondary pigment was diagnostic for the presence of the primary UVF4 filter in these rows. However, in N. wennerae (Fig. 1C) and O. scyllarus, the primary UVF4 pigment is present even when the secondary fluorophore is not (Fig. 2A). We encountered UVF4 only in the Gonodactyloidea, specifically in all gonodactylids and also O. scyllarus. UVF4 acts as a long-pass filter in conjunction with the secondary short-pass filter (when present), creating a notch filter.

The widespread occurrence of these four UV filter types suggests a degree of consistency in the UV tuning system of stomatopods. Of the 21 species examined by MSP, only two (Lysiosquillina maculata and Coronis scolopendra) had no detectable UV-absorbing pigments in the crystalline cones. The remaining 19 species all had at least UVF3 in midband row 4. Also, with the exception of the unusual UVF2 pigment in O. scyllarus row 3, all primary filters are essentially similar pigments with similar probable filtering functionality. This suggests that the filters first appeared during the early radiation of stomatopod superfamilies, and raises questions about the evolution of the diversity of filters observed in extant species.

Our character state reconstruction suggests that the ancestral stomatopod eye likely had at least one UV filter (Fig. 3A). Ancestral state reconstructions of the individual primary filters indicate that this ancestral UV filter was most likely to be UVF3, located in midband row 4 (Fig. 3B). UVF3 is also found in Hemisquilla californiensis, which is thought to represent a basal lineage amongst modern stomatopods (Ahyong and Jarman, 2009; Porter et al., 2010). The four UV filter classes found in the gonodactylids are apparently an elaboration unique to that family. The unusual nature of the UV filters in O. scyllarus implies that this species may have come to its elaborate UV filter set independently of the diversification in the gonodactylids. Our reconstruction also suggests that the absence of UV filters in lysiosquilloids represents a loss of ocular complexity in that superfamily, as it is nested among the hemisquilloids, pseudosquilloids and the gonodactyloids, all of which have UV filters. This loss of optical filtering complexity in the lysiosquilloids is consistent with the results of Porter et al.’s (2010) study of the intrarhabdomal long-wavelength filters, which also found a reduction of filter types in this group. The only other loss of UV filter diversity indicated by our reconstruction involves the loss of UVF1 in G. chiragra and N. wennerae from a gonodactylid ancestor that likely possessed all four UV filters.

Also of note is the UVF1 filter in P. ciliata. If only UVF3 were present in the common ancestor of the pseudosquilloids and gonodactyloids, it is surprising that both lineages independently stumbled upon a porphyra-334-like MAA for use in a UVF1 filter. However, P. ciliata is the only species we encountered in which UVF1 is expressed only in midband row 1. Perhaps, as porphyra-334 is potentially a precursor to most of the UV filters in stomatopods (Bok et al., 2014), it has been incorporated independently on multiple occasions. Additional taxon sampling in the pseudosquilloids, odontodactylids and protosquillids could better resolve these unusual cases.

As we have shown here, the four primary filter classes are quite rigid in their location within the midband, general filtering effect and spectral shape within single species (Fig. 2). Between species, however, the spectral properties of the filter classes do exhibit considerable variability (Fig. 2B). UVF1 is fairly consistent among species, appearing to conform mostly to the absorbance of porphyra-334 (λ_max=334 nm). However, in G. platysoma and G. viridis this filter has a broader red-shifted spectrum, perhaps owing to the inclusion of additional pigments (or elevated prominence of scattering effects caused by low pigment density in the case of G. viridis). UVF2 has a consistently shaped absorbance spectrum that nevertheless varies in λ_max from 358 nm in G. platysoma to 376 nm in G. ternatensis (excepting the strange midband row 3 pigment in O. scyllarus). Similarly, UVF3 also varies among species, from λ_max=361 nm in G. platysoma to 387 in H. californiensis. Finally, in most species with UVF4, the filter seems to be based on mycosporine-glycine (λ_max=310). However, Gonodactylaceus falcatus, G. ternatensis, O. scyllarus and G. affinis all have UVF4s with shorter-wavelength absorbance spectra. Aside from shared evolutionary history, what factors could be the driving force behind this variability as well as the differences in primary filter number mentioned earlier?

To answer this question, we examined the absorbance spectra from UVF3 in various species. We noticed several trends that seemed to be related to the habitat depth range of the stomatopod species. The shorter-wavelength-absorbing variants of this filter seemed to be found primarily in species that lived exclusively in shallow habitats (G. platysoma, G. chiragra and G. viridis), while species living in deeper habitats (H. californiensis, O. scyllarus and C. hystrix) tended to have the reddest-shifted UVF3 pigment variants. Additionally, the deeper-living H. californiensis had the lowest recorded UVF3 optical density. Finally, we noticed that species that lived at greater depths tended to have fewer filter types. Taken together, these observations imply that species’ habitat depth range influences the tuning, density, and number of their UV filters. We hypothesize that as short-wavelength UV light is attenuated rapidly in the aquatic environment at increasing depths (Jerlov, 1976), stomatopod UV optical pigments are tuned or lost in order to filter deep UV light less aggressively, thus maintaining a polychromatic UV visual system, without driving photoreceptor sensitivity below the threshold for function in the available environmental UV light.

In order to learn whether UV filter tuning correlates with species habitat depth, we plotted the λ_max of UVF2 and UVF3, the two primary short-pass filters, from each species versus that species’ mean habitat depth (Fig. 4A) (Ahyong, 2001; Manning, 1969; Schram and Müller, 2004). The results are consistent with our hypothesis that habitat depth plays a role in the evolutionary spectral tuning of UV filters; longer λ_max filters are found more regularly in deeper-dwelling species, independent of phylogenetic relationships. Interestingly, this distribution is best fitted with logarithmic trend lines, echoing the logarithmic attenuation of light in water. UVF2 is also less commonly present as mean habitat depth increases. The primary outliers in this analysis are G. smithii and G. affinis, both of which exhibit shorter λ_max filters than other species with their mean depth. Both of these species are found over a great range of depths, from intertidal and littoral reefs to depths of over 60 m, but all the individuals from these species included in this analysis were caught at depths of less than 2 m. Finally, it is interesting that UVF3 is preserved in primarily deep-dwelling species, albeit in a red-shifted state. This suggests an important ecological function for spectrally tuning UV receptors to short wavelengths in the UV.

We also tested whether species that live at a great range of depths spectrally tuned their UV filters in response to their residential depth, or whether instead the spectral characteristics of the filters are fixed within a species. We collected H. trispinosa from 20 m and from shallow intertidal reef flats at Lizard Island in Australia. We found that the absorbance spectra of UV3F in these two H. trispinosa populations were essentially identical (Fig. 4B), but the λ_max (mean±s.e.) of the two populations varied between 378±0.50 nm in shallow-caught individuals and 380±0.76 nm in deep-caught individuals. This degree of tuning is functionally irrelevant and is not consistent with the much broader spectral tuning in this filter seen among different species. We also carried out an experiment in which H. trispinosa were collected in intertidal habitats and placed in laboratory aquaria. Half of the animals were left in normal daylight, while the other half were placed in complete darkness for 2 weeks. The animals were not fed during this time. After 14 days, their UVF3 filters were measured (Fig. 4C). For both populations the absorption spectra remained identical to those of freshly shallow-caught animals, at λ_max=378 nm. This result suggests that if the filter is tuned in low light environments, any changes require longer than 2 weeks, and that the pigment that contributes to UVF3 is fairly photostable and persistent, as its density did not decrease noticeably after 2 weeks in either population without dietary input.

In addition to the four primary UV filter types discussed previously, we also identified several ‘secondary’ UV filters in the crystalline cones of various species (Fig. 5). Here, we discuss the spectral properties of these filters and examine their potential uses, evolutionary trends and ecological significance.

In all four protosquillid species examined here, as well as P. ciliata, we noticed a dim green fluorophore in the crystalline cones of midband rows 2 and 3 (Fig. 1F,G, Fig. 5A). There are no primary filters in these rows; protosquillids only have UVF3 in row 4, and P. ciliata has UVF3 in row 4 and UVF1 in row 1 only. The absorbance spectrum of this weak pigment was measured in all five of these species (Fig. 5A), and it was found to have a λ_max between 378 and 391 nm, putting it in similar general wavelength range as the UVF3 pigment in these species. It could be that this pigment in rows 2 and 3 is the same as UVF3. It is not a strong absorber, with a maximum density not exceeding OD 0.28, so it likely does not have a strong effect on R8 spectral tuning in these rows. However, this pigment is interesting in that it occurs in very distantly related species, including protosquillids and the pseudosquilloid P. ciliata, suggesting some well-conserved function.

UV filters were also found in the crystalline cones of the peripheral retinas of some species. In O. havanensis, the first 9 or 10 rows of the ventral periphery (adjacent to the midband) contain a strong pigment that is spectrally related to porphyra-334 UVF1 pigments (Fig. 5B). This pigment has a λ_max at 331 nm and an OD_max of 2.08. Interestingly, the peripheral crystalline cones that express this pigment nicely cover the pseudopupil in that region when the eye is oriented directly at the observer (Fig. 5B). Therefore, as the ommatidia from both peripheral regions and the midband image the same area of space in this orientation, this pigment could create a UV color vision system between the peripheral regions without input from the midband.

UV filters are also found in the dorsal periphery of certain gonodactylids. When examining crystalline cone fluorescence patterns in eye cross-sections, we noticed that in some species, crystalline cones in the first few rows of the dorsal periphery adjacent to the midband exhibited blue fluorescence when stimulated with 375 nm UV light, akin to the UVF2 pigment in midband row 3 (Fig. 5C, large arrowhead). This has been observed in N. oerstedii, N. wennerae, G. smithii and G. chiragra. It proved to be very difficult to record absorbance spectra from these crystalline cones as they are so narrow and distort heavily when sectioned, but we were able to record a few scans from G. chiragra and N. oerstedii (Fig. 5C). The λ_max of this pigment was near 360 nm, with a weak OD_max of 0.25. Interestingly, the crystalline cones expressing this pigment actually point at a slight downward angle, across the midband's field of view (Marshall and Land, 1993), so they do not image the same point in space as the midband and ventral periphery (as was the case for the ommatidia containing pigment in the ventral periphery of O. havanensis). This could imply a different function for this pigment, perhaps acting as an ‘eye-shade’ to keep off-axis UV light from entering the midband crystalline cones from above, as is seen in the UV-screening dorsal corneal pigments of some teleosts (Douglas, 1989; Muntz, 1982). This is reasonable as the heavily tuned R8 UV receptors of the midband are likely very insensitive to UV light, so it would benefit the eye to minimize scattered UV light from laterally entering the midband R8 rhabdoms. Furthermore, in G. smithii only, there are additional crystalline cones with this blue fluorescent pigment in the dorsal region of the dorsal periphery (Fig. 5C, small arrowhead). It could be that these pigments are being used as photo-protectants against down-welling UV light. It is even possible that the primary UV filters in the midband originated as ocular photoprotectants – the role most commonly associated with MAAs in nature – and they were subsequently exploited for UV spectral tuning in stomatopods.

The discovery of UV filter pigments in the peripheral hemispheres of stomatopod eyes supports the notion that these regions are not completely homogeneous arrays of ommatidia used solely for spatial vision. Together with the finding that the R8s in the dorsal and ventral peripheries express two different opsin transcripts (Bok et al., 2014), these specializations suggest additional roles for the peripheral hemisphere ommatidia, including contributing to UV color vision. It is interesting that the peripheral UV-absorbing filter pigments are found directly adjacent to the midband. It is as if the extreme retinal specialization typical to the midband is ‘leaking’ into the adjacent ommatidia. However, this observation could be a by-product of an incomplete survey of the peripheral crystalline cones that were not directly adjacent to the primary area of investigation. It is possible that many additional, non-fluorescent UV-absorbing pigments in the periphery simply evaded our notice.

We found that stomatopods possess UV optical filters at an early stage in their post-larval development. In stomatopods, the adult eye actually develops de novo alongside the degenerating larval eye during late larval developmental stages (Cronin et al., 1995; Feller et al., 2015; Jinks et al., 2002), and the newly developing retina already includes the long-wavelength filter classes found in the main rhabdoms. Unsettled pelagic Haptosquilla postlarvae were observed to already have UVF3 as indicated by fluorescent emission in row 4 of the midband, as well as the row 2 and 3 secondary filters (Fig. 6). This suggests that MAA filters are incorporated into the crystalline cones early in development, perhaps as soon as the adult eye develops during late larval stages. Additional observations are required from larval developmental stages, as well as longer-term experiments where postlarvae are reared in variable ambient UV light environments similar to previous studies on the long-wavelength colored filters in the main rhabdoms (Cheroske et al., 2003; Cronin et al., 2001).

Stomatopods are recognized as having complex and variable visual systems, well suited for an active social and predatory lifestyle. Here, we have shown that this complexity extends to the unique UV filters recently identified in the crystalline cones. Stomatopods express a variety of UV-absorbing pigments in various complements to form these UV filters. Furthermore, the UV filters were probably present in early stomatopods, and they have since been elaborated, and occasionally lost, over evolutionary history. The UV filters also seem to be tuned such that shallow-water species have more aggressively tuned UV photoreceptors in an environment with abundant UV light, while deep-living species have fewer, red-shifted filter pigments that maintain receptor sensitivity when less UV light is available. Finally, in some species, UV filter complexity extends beyond the midband, into the peripheral hemispheres, implicating these regions in more complex UV visual tasks than previously thought. These results further emphasize the importance of a highly tuned UV visual system for stomatopods. However, we still do not know how, and for what ecological tasks, stomatopods actually use this visual capacity. Future behavioral and ecological investigations will aim to better understand the role of UV vision in stomatopod predation, mate choice, species identification and aggression.

Twenty-eight stomatopod taxa were examined in this analysis. This group represented the four superfamilies that have the most complex eye designs, with six specialized midband rows: Gonodactyloidea, Lysiosquilloidea, Pseudosquilloidea and Hemisquilloidea. The majority of the studied stomatopod species were captured at Lizard Island Research Station (QLD, Australia) by various collection techniques. Additional species were collected at Catalina Island (CA, USA), Moorea (French Polynesia) and the Florida Keys (USA), as well as purchased from tropical marine life distributers. Species were identified based on anatomical characteristics. The UV filters from animals collected at Lizard Island were either analyzed in the field using imaging spectroscopy or returned to the laboratory and analyzed using MSP. Both techniques are described in detail below, and there were no discernible differences in absorbance spectra for species that were examined using the two methods. All other animals were examined in the laboratory using MSP.

Dissected whole eyes were affixed to a goniometer using superglue and the eyes were oriented to permit focusing directly on the midband with a dissecting microscope. The microscope was focused such that the depth of field extended from the surface of the eye, through the crystalline cones and into the distal tip of the rhabdoms, producing dark pseudopupils in the midband and both peripheral hemispheres. When possible, the eye was rotated to visualize the enlarged pseudopupils in the acute zone of the eye. The eye was illuminated with two identical UV LEDs (no. LED370E, Thor Labs, NJ, USA) positioned on either side of the eye at roughly 45 deg relative to the surface of the eye. Exposures were taken with a Canon T2i DSLR using a microscope adaptor lens. Raw photos were globally manipulated for exposure, contrast and noise reduction, and backgrounds were subtracted. Eye cross-sections (200 μm) were prepared by vibratome sectioning as described in Bok et al. (2014) and fluorescence was imaged using a compound fluorescence microscope producing 365 nm epi-illumination. Photographs were captured and processed as above. Six species were examined only by fluorescence photography, and the presence of UV filters in rows 3 and 4 as well as the long-wavelength secondary component of rows 5 and 6 was inferred from the presence or absence of the typical fluorescent emission in each row (supplementary material Table S1).

To examine the absorption spectra of the crystalline cone UV pigments, 200 μm sections through the crystalline cone layer of the eye were prepared by vibratome sectioning, and the sections were mounted in between two coverslips in filtered seawater. MSP was carried out according to methods described previously (Cronin, 1985; Cronin et al., 1994c), with modifications to the system to maximize the spectral range of UV wavelengths directed through the crystalline cones for measurement as described in Bok et al. (2014). We were able to obtain smooth and consistent absorbance spectra between 300 and 550 nm using this setup. All absorbance data presented in this analysis were collected using this MSP, except for those from L. maculata, C. hystrix and the experimental H. trispinosa discussed in the section on habitat depth, which were recorded using a field MSP at Lizard Island. This instrument was based on the portable MSP described in Loew (1982). A xenon arc light source was used, and the monochromator replaced with a housing placing a UG5 filter and an IR blocker in the optical path. An aperture was created with a pinhole in aluminium foil prior to the filters. A quartz prism was used to direct the beam through a sample between quartz glycerol immersion lenses (a pair of Zeiss 32X Ultrafluors). Absorbance was measured using Spectra Suite software and a QE6500 spectrometer (Ocean Optics, FL, USA) with a 200 μm fiber placed in the imaging plane above the microscope. This MSP was only functional down to about 330 nm, so it was not useful in characterizing the primary filter in rows 5 and 6. Whenever possible, all UV filter data presented here were compared between the main MSP and the portable MSP with fresh-caught animals, and no discrepancies were observed. This indicates that transportation of the animals to the laboratory did not seriously impact the spectral properties of the filters.

Absorbance spectral data were corrected for scattering through the thick sections by fitting and subtracting the baseline from the slope between 450 and 550 nm, where no pigments were ever observed in the crystalline cones. From each individual the cleanest scans (without severe short-wavelength scattering or post mortem pigment degeneration) for each filter type were selected, compared and averaged with data for other individuals of the same species. The λ_max for each filter in each species is reported in Table 1, along with the number of scans that were averaged, the number of individuals they were pooled from, and the maximum OD of the filter pigment observed in an end-on scan through a crystalline cone in transverse section. This approximation of OD is an unavoidable shortfall of measuring the absorbance spectra of these filters, which begin to leach out of the crystalline cones and degenerate when the cones are ruptured during sectioning, and the actual densities of these filters in life are likely higher than reported here (see Bok, 2013 and Bok et al., 2014 for additional details). Therefore, our analysis primarily concerns the presence or absence and the spectral absorbance of the pigments.

In discussing the role of the UV filters in spectral tuning in various species of stomatopods we assume that the UV visual pigment expression pattern reported in Bok et al. (2014) for N. oerstedii is mostly conserved in other stomatopods with six midband rows. Analyses of main rhabdom photoreceptors have shown that their visual pigment absorbance spectra are fairly invariant in comparison to the colored filters in those ommatidia (Cronin and Caldwell, 2002; Cronin et al., 2002). Electrophysiological and MSP studies of the R8s have also suggested conservation in visual pigment absorbance spectra among species (Bok, 2013; Cronin et al., 1994c; Kleinlogel, 2004; Kleinlogel and Marshall, 2009; Thoen et al., 2014).

DNA extraction, PCR and sequencing procedures were performed as described in Porter et al. (2010). New sequences were generated for species where data were not already available (supplementary material Table S2) and added to the data from Porter et al. (2010) for a dataset consisting of 47 stomatopod species. Representatives of three additional Eumalcostracan lineages were used to root the phylogeny (supplementary material Table S2). Each gene dataset was aligned separately using MAFFT v7 (Katoh and Standley, 2013; Katoh and Toh, 2008; Katoh et al., 2005, 2002), concatenated, and highly divergent and/or ambiguous regions of the entire alignment were removed using GBlocks 0.91 (Castresana, 2000). The resulting alignment was used to estimate phylogenetic relationships and node confidence as bootstrap values using RAxML v7 (Stamatakis, 2006; Stamatakis et al., 2008) as implemented in the CIPRES portal (Miller et al., 2010). The resulting phylogeny was drawn using FigTree v1.4.0 (http://tree.bio.ed.ac.uk/software/figtree).

Ancestral character states for the number of distinct types of UV filter pigments found in each species and the presence/absence of each type of filter pigment were reconstructed in Mesquite v2.75 (Maddison and Maddison, 2011), using the RAxML tree with species lacking filter pigment data pruned from the tree. All characters were assigned as standard categorical characters. Ancestral states for each character were inferred using Maximum Likelihood models and reconstructions were mapped onto the phylogeny.

The relationship between the λ_max of UVF2 and UVF3 and the maximum depth of each species was determined using a PGLS method, which accounts for the shared evolutionary history of genes in calculating the correlation between characters. PGLS regressions were run in the caper package v0.5.2 (Orme et al., 2013) of the software program R v3.1.1 (R Core Team, 2014) as implemented in RStudio v.0.98.994 (2012). For PGLS analyses, the habitat depth was taken as the averaged maximum recorded depth for each species reported in Ahyong (2001), Manning (1969) and Schram and Müller (2004), and from personal communications with Roy L. Caldwell, and normalized by a log₁₀ transformation.

Fifteen H. trispinosa individuals were captured in water less than 1 m deep at Lizard Island Research Station. They were split into three groups that were matched for sex and size. Five of the animals were killed immediately and their UV filters were examined by imaging spectrometry. Of the 10 remaining animals, five were placed in aquaria exposed to normal daylight, and the other five were placed in constant darkness. Both groups were not fed for this period, and after 14 days all animals were killed and their UV filters were measured using the field MSP. The average absorbance spectra from the three groups were then compared with one another as well as with H. trispinosa freshly caught at a depth of 20 m.

We are grateful to Sheila Patek and Roy Caldwell for collecting and sharing some animals used in this study (Patek funding: National Science Foundation no. 0641716 and National Geographic Society Committee for Research and Exploration). We thank Ellis Loew who graciously lent us his portable MSP for field measurements. We thank those who assisted with animal collection at Lizard Island (Kate Feller, Roy Caldwell, Justin Marshall and Hanne Thoen) and at Catalina Island (Brian Dalton). We thank the directors and staff at Lizard Island Research Station for facilitating our research. Thank you to Erin Sternhagen and Lee Wessel for help with generating additional sequence data for phylogenetic analyses.