Retinal MSP measurements.—Thirty-nine species of fishes were collected during May and June 1999 from locations surrounding the Hawaii Institute of Marine Biology (HIMB), Kaneohe Bay, Oahu, Hawaii. The species, sex, and age of fishes examined were dictated by availability and our desire to obtain a broad taxonomic sample. Fishes were transported to HIMB and maintained on a chopped squid diet for seven days or less in aquaria with running seawater on the natural light/dark cycle. Such holding times should have no effect on absorption properties of the eye. Prior to visual pigment measurement, we measured standard length (SL) and then dark-adapted the fish for a minimum of four hours. Fish were anesthetized with an overdose of MS222 (Sigma) and pithed. Under infrared (IR) illumination, each eye was enucleated and the retina removed in a standard Sorensen's buffer (pH 7.2) supplemented with 6% sucrose. Small patches of retina, sampled from throughout the retina, were placed on a 22 × 30 mm cover slip, chopped into small fragments, and teased with needles to release cells from the retinal mass. The preparation was sealed with an 18 × 18 mm cover slip edged with grease.

The MSP was a single-beam, computer-controlled instrument fitted with quartz and flourite optics (for details see Loew, 1994). A 100-W quartz iodine lamp provided sufficient UV radiation for absorption measurements to 350 nm. The retinal preparation was placed on the microscope stage of the MSP and examined under IR illumination. A baseline scan from 750 to 350 or 360 nm and back was taken through a clear area of the preparation. The measuring beam was then passed through an outer segment (OS) of an individual photoreceptor and the spectrum scanned and stored in the computer. Subtraction of the baseline provided a record of each cell's absorption spectrum.

Three criteria were used to establish that an absorption spectrum was from a visual pigment and not some other colored contaminant. The first criterion is based on the fact that visual pigments are photosensitive and thus are destroyed by light exposure—a process called bleaching. The second criterion rests on the fact that the fixed orientation of visual pigment molecules in the membranes of the outer segment renders them dichroic. That is, light of one plane of polarization is absorbed more strongly than orthogonally polarized light. This means that absorption spectra from the same cell measured under the two polarization conditions should have different peak amplitudes but the same shape. The third criterion is based on the fact that when visual pigment absorption spectra are plotted as normalized optical density versus normalized frequency they have a characteristic shape that is essentially the same regardless of the spectral position of the absorption maximum. This is the basis of the template-fitting routine used not only to establish the absorbing substance as a visual pigment but also to locate accurately the absorption maximum or λ_max (for details, see Loew, 1995; McFarland and Loew, 1994). In most instances, the best fitting template to the right-hand limb of a spectrum (i.e., the long-wavelength limb) was used to estimate the λ_max. When three or more similar identifiable cells (e.g., rods, single cones, etc.) were scanned the mean estimated λ_max-value was calculated ± 1 SD.

Ocular media transmission.—Fishes were captured by various means and either iced down on the collection boat or sacrificed within two days in the laboratory. Thorpe et al., 1993, used lenses frozen for up to 2 months with minimal effects. Our tests showed negligible effects of being iced down for up to 21 h, longer than any fish was iced down in our study. Eyes were removed with care to prevent separation of the dermal cornea that was sometimes loosely attached to the deeper scleral cornea. Our goal was to measure the spectral transmission through each of the components of the preretinal ocular filters. To measure transmission through the entire suite of filters, a slit or window was cut through the sclera, near the back of the eye and opposite the pupil, just through the retina. Following measurement of light transmission through the entire eye, the lens was dissected from the eye, rinsed in fresh water, and measured. When the dermal cornea was only loosely attached, measurement through the intact cornea was followed by separate examination of the dermal cornea. When absorption by the lens and cornea could not account for the absorption measured for the whole eye, the humors were also separately examined.

Spectral transmission was measured with an Ocean Optics S2000 spectrometer optimized for UV sensitivity with a 100 um slit. The broad-bandwidth light source was an Ocean Optics DT-1000 with deuterium and tungsten bulbs. All components were designed for UV-visible light analysis. Whole eyes were mounted with the pupil facing down over a hole in an aluminum slide. A modified microscope stand with a micromanipulator stand was fitted with a 400-micron fiber optic probe held in a syringe needle. The needle was introduced into the window or slit in the eye until just submerged in the humor. Light from the DT-1000 was fed through the fiber optic probe, through the eye, into a lens below the microscope stage, and through a 400-micron fiber to the S-2000.

Because we did not use an integrating sphere to capture all light passing through the eye, it was necessary to manipulate the eye carefully under the probe to obtain transmission curves with a more or less flat long wavelength slope at about 100–150% transmission. A flat curve at maximum transmission was an important indication that the effects of any chromatic aberration had been minimized. At least two samples were taken from different locations for each preparation and percent spectral transmission for each component was standardized to 100% at the largest value obtained below 601nm. Losey et al. (2000) found that this method produced an average difference between normalized samples at each wavelength of only 2.6%. A few eyes were also analyzed by the SubSpec instrument that had been validated against an integration sphere (Siebeck and Marshall, 2000, 2001) and no critical differences were detected between the methods.

Larger lenses were placed over a hole in an aluminum slide that was just smaller than the lens, and smaller lenses, corneas, and dermal corneas were placed on top of a UV-transmitting (UVT) plastic slide. UVT transmits nearly all light down to 320 nm. Lenses were manipulated for measurement of transmission similar to the entire eye. Humor samples were placed on a UVT slide with 0.25-mm thick cover slip spacers at each end. A second UVT slide was placed over the top to flatten the humor sample to 0.25-mm thickness (see Zamzow and Losey, 2002). For a few species with extremely thick, jelly-like humors, a “thick humor” sample was obtained from humor samples up to about 2 mm thick prior to flattening as above.

The wavelength at which the ocular media blocked 50% of the incoming radiation (T50) was used to indicate the short-wavelength cutoff of the eye. This is the wavelength at which half of the quanta of that wavelength are blocked. In general, eyes sharply increase their blocking characteristics at the shorter wavelengths, usually below blue and often in the range of UVA radiation (320—400 nm). All of the eyes we examined blocked essentially all of the UVB radiation (< 320 nm).

Each family was characterized by the average and range of T50 values found for that family. To interpret the widely divergent ranges found, a randomization technique was used to establish the random expected range. In each of 1000 iterations of this test, our entire dataset of T50 values was sampled the appropriate number of times as determined by the number of species sampled in each family (2–19). The 975th smallest and largest value for each family sample size estimated the random expected range of T50 values.

Visual pigment categories.—For the 38 species examined, the rod λmax ranged from 477 nm in the damselfish, Chromis hanui, to 502 nm for the squirrelfish, Neoniphon sammara (Fig. 1, Table 1). Nearly half of the fishes examined possessed cone cells that would provide enhanced photosensitivity in the violet or UV regions of the spectrum. Kuhlia sandvicensis possessed unusually large UV-sensitive cone cells (OS diameter of 4–5 μm) and low noise UV absorption spectra (Fig. 2).

We group cones with λ_max-values below 435 nm as short wavelength cones. We divide this group further into UV- (λ_max ≤ 376 nm) and violet-sensitive (λ_max 398 to 435 nm) cones. Although these designations are arbitrary, such groupings serve to emphasize the peculiar properties of the violet and UV portions of the spectrum underwater, where the scattering of light is increased and the visual range of fishes is reduced (Loew and McFarland, 1990; Losey et al., 1999). Of the 38 species, eight possessed UV-sensitive cones with λ_max from 347 to 376 nm, and 14 species had violet-sensitive cones with λ_max from 398 to 435 nm. All of these short-wavelength visual pigments were present in single cones. Specific examples include the flagtail, Kuhlia sandvicensis, that possessed large, single UV-sensitive cones and the filefish, Pervagor aspricaudus, with a violet-sensitive cone λ_max at 404 nm (Fig. 2). Several species possessed both UV-and blue-sensitive (λ_max between 436 and 500 nm) cones, but no species had both UV- and violet-sensitive cones. In the 10 species having violet-sensitive cones, we detected no blue-sensitive cones. In most of the coral reef fishes with a violet absorbing visual pigment, the λ_max was close to 400 nm. Only in the trumpet fish, Aulostomus chinensis and surgeon fishes of the genus Naso was the violet pigment located closer to 420 nm.

Paired cones (double and twin) were found in all species examined. Double cones consist of a principle member (double cone; P) and a morphologically dissimilar accessory member (double cone; A). In twin cones the two members are morphologically identical. The members of double cones contain different visual pigments. Twin cone members, however, may contain the same or different visual pigments in which case they are called identical twins or nonidentical twins, respectively. In our sampling, double cones were the most common (31 of 38 species), and four species had both double and twin cones. Twenty-two species had green-blue-sensitive pairs (i.e., blue < 500 nm vs green > 500 nm), and nine species had unequal green-green-sensitive pairs (Table 2). Eleven species had identical twin cones, nine of which contained matched green visual pigments. Identical blue-blue-sensitive twin cones were found in the trumpet fish, Aulostomus chinensis, and the boxfish, Ostracion meleagris.

Spectral transmission categories.—The T50 for the entire eye ranged from 337 to 455 nm for the 185 species and 49 families for which the entire eye was sampled (Appendix 1). The T50 for the lens alone ranged over slightly shorter wavelengths (322 to 441 nm) since the blocking effects of the cornea, and in some cases, the humors, were lacking. Douglas and McGuigan (1989) suggested three categories for ocular lens T50 cutoffs: type I (< 355 nm), type II or “colorless” (355 < T50 < 405 nm) and type III or “yellow” (> 405 nm). They suggested that, considering a typical UV-sensitive visual pigment with maximum absorption at 360 nm, only fish with type I lenses were likely to perceive UV radiation. Two findings suggest modification of their scheme. First, we have indicated UV-sensitive visual pigments at longer wavelengths in Hawaiian fishes. Second, Thorpe et al. (1993) indicated that the goldfish, Carassius auratus, known to possess UV color vision, could have a T50 as high as 391 nm. In recognition that fishes with UV vision may have longer wavelength T50 values, we organized our results into four categories of ocular transmission that expand those of Douglas and McGuigan (1989) as type I has T50 <= 355 nm, type IIa has 355 < T50 <= 380 nm, type IIb has 380 < T50 <= 405 nm, and type III has T50 > 405 nm.

If we take the lens transmission measurement as a substitute for the 10 species for which we lack a T50 estimate for the entire eye, 195 species can be classed as to ocular transmission category. Thirty-two species were in class I suggested by Douglas and McGuigan (1989) as likely to have UV vision. Of the remaining species, 80 were classified as transmitting short-wavelength light to the retina with 30 as class IIa and 50 in class IIb. The 83 remaining species were in class III.

Taxonomic distribution of T50 values.—The range of T50 values within a family was generally lower than expected (Appendix 1, Fig. 3). Only five of the 25 families with more than one species sampled fell clearly within the 95% confidence limits (randomization test): Carcharhinidae, Scombridae, Synodontidae, Serranidae, and Labridae. Other than sharing a predatory lifestyle, there is little in common among these five families.

A few families had both an adequate sampling of species (≥ 8) and genera and an unexpectedly low range of T50 values: Chaetodontidae, Acanthuridae, Pomacentridae, Holocentridae, Lutjanidae, and Carangidae. Their family average T50 values were diverse and ranged from 350 to 410 nm. In terms of their ecology, the species have little in common and range from nocturnal to diurnal activity and from pursuit predators to herbivores.

The phylogenetic order of the families taken from Nelson (1994), with a few modifications as suggested by D. Greenfield (pers. comm.; (Table 3), was used to indicate their relative phylogenetic position. Lacking knowledge of the relative phylogenetic distance between families, they were simply assigned an order number from 1 to 49. Linear regression of the family average T50 value on taxonomic position (Fig. 4) indicates that more phylogenetically advanced families are more likely to have short wavelength ocular blockers and lack UV vision. The scatter is, however, extremely wide such that phylogenetic position alone cannot indicate the short-wavelength properties of their eyes. Much of the correlation depends on the 15 most phylogenetically advanced families past the Pomacentridae (Labridae–Ostraciidae, Table 3). If these families are removed, the correlation is lost (r² = 0.01, n = 34, P > 0.1).

The frequency distribution of the T50 values is clearly bimodal (Fig. 5). As an initial attempt to discover the source for the bimodality, we separated out all families in our sample that are diurnally active, mobile and closely associated with the coral reef as opposed to sand flats, etc. These were Acanthuridae, Balistidae, Chaetodontidae, Labridae, Monocanthidae, Pomacanthidae, Pomacentridae, Scaridae, and Tetraodontidae. These families had an arguably unimodal distribution of T50 values and the remaining families remained bimodal but shifted to shorter wavelengths.

Transmission blocker location categories.—The anatomical location of the radiation blockers varies between species (Fig. 6). Here we use the anatomical location of the transmission blocking effect that results in the longest-wavelength blocking. When the difference between the T50 for the lens and that for the entire eye was less than 6 nm, this was assumed to be caused by a measurement error, and the lens was assumed to be the critical or “limiting” blocking agent. When the difference was within 15 nm but the shapes of the curves were identical, the lens was again assumed to be the limiting element. The differences were likely caused by either a measurement error or the cumulative effect of adding the blockers from the cornea and humors that were similar to those in the lens. In other cases, the lens was not classed as the limiting blocking agent, and, when possible, the location of the limiting blockers was noted as either corneal or, in a few cases, humoral.

We characterized slope of the spectral transmission curve for the whole eye similar to Siebeck and Marshall (2000, 2001) as: “Steep” or “Class I”: less than 30 nm between the 20 and 80% cutoffs (T20 and T80); “Gradual” or “Class II”: less steep (> 30 nm difference between T20 and T80) with a gently curving slope but never having intermediate maxima; “Variable” or “Class III”: changes in slope result in intermediate maxima.

The lens was the limiting filter in 73% of the species, but this varied with the species' spectral transmission category. Steep lens transmission curves were most common in eyes where the lens was the limiting filter (Chi-square = 30.4, df = 1, P < 0.001) and in species that blocked most short wavelengths (Chi-square = 33.0, df = 1, P < 0.001). As expected, vision likelihood categories least likely to possess UV vision (see below) were most likely to use the lens as the limiting filter (Table 4). Twenty-four of the 49 families had at least one species in which the lens was not the limiting filter. Only two families, Priacanthidae and Scombridae, included the unusual combination that employed the humors as the limiting filter. There was no obvious correlation for any of these features with taxonomic position or ecological group.

Vision likelihood categories.—The diurnal habits of each species were assigned a category as (1) diurnally active and exposed to full ambient light, (2) nocturnal but exposed to full ambient light during the day, and (3) hidden from daylight. We hypothesized that fish exposed to full daylight that had the adaptation to block short wavelengths were least likely to be adapted for UV vision. Five “vision likelihood” categories were created as (1) highly likely to possess true UV color vision (transmission category I, daylight exposure category 1 or 2); (2) likely to possess true UV color vision (transmission category IIa, daylight exposure category 1 or 2); (3) likely to have violet visual sensitivity, but unlikely to have true UV color vision (transmission category IIb, daylight exposure category 1 or 2); (4) sensitive to short wavelength light, but not adapted for UV color vision (transmission category I, IIa, or IIb, daylight exposure category 3); and (5) unlikely to have short wavelength vision (transmission category III, any daylight exposure category).

Thirty-eight of the 195 species were found in categories 1–2 in which UV vision is likely. Of the 26 families that had at least two species in our samples, 11 had at least one representative in categories 1–2 that are strong candidates for UV vision. For those 24 species for which we have both T50 and MSP data for single cones, there appears to be a strong correlation between the shortest wavelength absorption maximum for their single cone visual pigment and both T50 (r² = 0.76, P < 0.01, excluding the four nocturnal species) and vision category (r² = 0.62, P < 0.01; Fig. 7).

Types of short-wavelength vision.—Visual systems, taken in the context of stimulation of the retina by short-wavelength radiation, can be of three general types. The simplest type lacks specialized UV receptors and merely fails to have UV blockers in the eye that function below 400 nm. At least some UV radiation would be absorbed by receptor cells that have their maximum sensitivity in the visible range but also have considerable short-wavelength sensitivity in a secondary absorption peak (beta peak) in the UV (Dartnall and Lythgoe, 1965). We will refer to this type of eye as “UV-sensitive.” Fish with UV-sensitive eyes likely lack true UV color vision or hue discrimination because the beta peaks for different visual pigments are similarly placed and have far less sensitivity than the primary absorption peak or λmax. Such eyes will act to gather as much light as possible but will result in a lowered color constancy (perception of a certain pigment color as the same regardless of changes in the incident illumination) that could be improved by filtering out the short-wavelength light (Dyer, 2001). These eyes may also indicate a lack of selection pressure to avoid the detrimental effects of UV exposure in habitats with little UV radiation (see below).

“UV-specialized” eyes must fail to block at least some of the UV spectrum and possess receptor cells that are maximally responsive to UV radiation. The visual pigments in these cells absorbed maximally at about 360 nm, similar to a variety of other vertebrates (reviews in Bowmaker, 1990; Losey et al., 1999; Locket, 1999). Such eyes might, but do not necessarily, have true color vision that recognizes a UV component as a different color. Alternatively, the UV-sensitive cone cells could serve separate perceptual functions such as prey detection. At least in the goldfish, UV-sensitive cone cells serve the function of true color vision and hue discrimination (Neumeyer, 1992).

“Violet-specialized” eyes must fail to block at least some radiation below 435 nm and possess receptor cells that are maximally sensitive to radiation from about 400—435 nm. Such eyes are similar to UV-sensitive eyes but should have increased sensitivity in deeper water where UV radiation is scarce compared with violet radiation and be subject to less damage because of harmful UV radiation. These eyes would also achieve some advantages of improved color constancy and focusing ability (Dyer, 2001; Muntz, 1976) and still retain some of the advantages of UV-specialized vision such as detecting short-wavelength opaque plankton against a bright scattering background.

Selection for and against short-wavelength vision.—Ultraviolet-sensitive visual pigments occur in primitive fish families and likely occurred in fishes ancestral to all present day species (Bowmaker, 1991). We must then ask why UV-specialized vision does not occur throughout the fishes. Selection against short-wavelength vision can result in both a lack of visual pigments with the primary absorption peak or λmax at short wavelengths and blocking of short wavelengths by the ocular media.

The bimodal distribution of ocular media blocking wavelengths (T50s, Fig. 5) suggests that shallow, diurnal and active coral reef fishes, in general, have been selected to block somewhat more short-wavelength light than the remainder of the Hawaiian species studied. These families also have a lower than expected range of T50 values within each family (Fig. 3) even though many are ecologically diverse. This group of fishes does, however, include species with UV-specialized vision even though, for some, much of the short-wavelength radiation is absorbed prior to reaching the retina. Siebeck and Marshall (2001) found a similar bimodal distribution of T50 values for Australian reef fishes. This might be expected because both studies were conducted on tropical coral reefs with similar environmental light conditions and presumably similar selection pressures. Hawai'i, however, has a comparatively depauperate fauna with 24.3% endemism and is marked by a lack of many tropical families and large shallow reef predators such as the groupers. Most of the families sampled in Hawai'i were included in Siebeck and Marshall (2001), but many families sampled in Australia are not found in Hawai'i. Similar selection pressures must be responsible for this agreement regardless of the phylogenetic differences.

There is, however, a clear phylogenetic trend toward increased blocking of short wavelength radiation (Fig. 4) as suggested by Douglas and McGuigan (1989) and Thorpe et al. (1993). The remaining variability, however, is high as noted by Dunlap et al. (1989) and must be variable because of selection pressures. Siebeck and Marshall (2001) found that UV-transmitting species from Australian reefs were restricted to the Acanthopterygii but noted that the Rajiformes are quite diverse, and UV transmission in the Rajidae had been reported by Thorpe and Douglas (1993). We can now extend the transmission of UV and probably possession of UV-sensitive vision to at least one species each of carcharhinid and sphyrnid sharks.

It is difficult, however, to find firm evidence for specific ecological effects in our data. Families with unusually low T50 values (Fig. 4) are of various ecological types: Ophidiidae and Caproidae include species that are sheltered within the structure of the reef. The Caracanthidae often sit on coral branches and are exposed to full sunlight. The Pomacentridae are active above the reef and are ecologically diverse, and members range from benthic herbivores to zooplanktivores. The Coryphaenidae, another distinct outlier with a high T50, is notable for their near-surface, UV-rich habitat.

Selection pressure on species must also vary within families. Of the five families in which at least four species were sampled by MSP (Holocentridae, Chaetodontidae, Pomacentridae, Acanthuridae), none had short-wavelength visual pigments present in all of the species.

More detailed understanding of the effects of phylogeny demands comparison at the species and age class level with detailed ecological information. For example, without considering details of their ecology, the relatively long wavelength visual pigments found in the pomacentrid planktivores, Chromis verater and Chromis ovalis, appear to be at odds with their placement in vision likelihood categories 1 and 2, respectively, that are expected to have UV-specialized vision (Fig. 7B). Other planktivorous pomacentrids (Chromis hanui and Dascyllus albisella) have shorter wavelength visual pigments as predicted by studies that show improved detection of zooplankton in the presence of UV light (Browman et al., 1994). Chromis verator and C. ovalis inhabit deeper reaches of coral reefs than do the other species (Gosline and Brock, 1965; Tinker, 1991). The reduction in available UV-light relative to the longer wavelengths with increased depth caused by filtering by the water would favor a visual pigment centered more toward the bluer or longer wavelength regions of the spectrum (McFarland, 1986; Loew and McFarland, 1990; Frank and Widder, 1996).

Can ocular media transmission categories predict visual pigment sensitivities?—One goal of this study is to determine whether we can place any faith in our ability to predict visual pigment sensitivity based on spectral transmission data and vision likelihood categories. Spectral sensitivity MSP studies are extremely demanding, whereas determination of T50 values can be accomplished with minimal instrumentation and in a short time period.

Our sample size for species with both ocular transmission and MSP data is small, but the results are encouraging. So long as we remove nocturnal species from consideration, the correlation with ocular media transmission is good (Fig. 7A). Our vision likelihood categories also have a good correlation with visual pigment sensitivity (Fig. 7B) that can be improved with admittedly post hoc consideration of additional ecological variables such as the habitat depth for Chromis spp.

For two other species that have visual pigments at unexpectedly short wavelengths, explanation escapes us. Aulostomus chinensis is a stalking predator that uses several guises to approach and remain close to prey and moves slowly about the reef. It feeds on fishes and shrimps and, although feeding during both day and night, appears to be most successful during twilight (Hobson, 1974). The presence of UV and/or violet absorbing visual pigments accompanied by blue-sensitive cones will increase the relative short-wavelength photosensitivity of each predator. Other predators, however, such as carangids lack short-wavelength sensitivity and it is doubtful that long-range detection of prey is of much value to any of these species. In clear water, predatory success is largely a function of detecting a defensive error by prey that are already at close range (Hobson, 1974; Landeau and Terborgh, 1986; Parrish, 1989). For Sufflamen bursa, a triggerfish that feeds on algae and various benthic invertebrates, we are similarly at a loss to explain their short wavelength visual pigment.

Visual pigments: Rods.—The frequency distribution of the visual pigments from the 38 species emphasize the differences in spectral photosensitivity among rods, single and double cones (Fig. 1, also see Marshall et al., 2003b).

The wavelength of maximum sensitivity (λ_max) of the rods, which are centered tightly around 490 nm, agree for individual species with the earlier findings of Munz and McFarland (1973) with an average difference of only 2.8 nm between the methods. Rods in this region of the spectrum maximize photosensitivity during twilight and at night when the near-surface underwater spectrum is shifted further toward the blue-wavebands (for review, see Loew 1995). Inter- and/or intraspecific differences in rod pigment λ_max may be adaptations to behavioral pressures (e.g., predation) but are also constrained by the photic conditions (Munz and McFarland, 1973; Lythgoe, 1984; Loew, 1995). Differences of 5 to 10 nm, when related to the photic regime, can enhance photosensitivity and, therefore, visibility.

Do the λ_max-values actually represent ecologically meaningful differences in the spectral locations of the rod pigments for each species, or are they merely an expression of variance in the MSP technique? Two independent results suggest the former. First, dispersion of the rod λ_max-values are relatively small (mean SD = 3.1, n = 33) for all species except Dascyllus albisella and Sufflamen bursa. For these two species, the number of records was small (n = 3), but this was also true for many of the other species. Even though the dispersion was typical of what we often obtain with MSP for most vertebrate visual pigments (i.e., < 3–4 nm), the higher dispersion for the rod values for D. albisella and S. bursa and for the cone visual pigments of several species (see Table 1) cloud the issue. Second, the test of the MSP technique is to make multiple recordings from the same cell along its length. This is done as a control for all instrumental variables. In cases where there is no evidence of multiple pigments in the same cell, such controls show variation of no more than ± 1 nm.

Visual pigments: Cones.—Single cones that are typically involved in color vision were limited to short and blue wavelengths less than 500 nm. This agrees with prior findings that visual pigments in tropical marine fishes are of shorter wavelength than many of those found in freshwater species and agree with the available environmental light (Munz and McFarland, 1977; McFarland, 1986). Given our methods and sample size, both typical for MSP studies, if other visual pigments do exist they are in a very small number of cells, found only in other developmental stages or restricted to a very small retinal area that we, by chance, failed to sample. One probable exception is our failure to find any type of single cone cell in Parupeneus multifasciatus, that is as likely a sampling problem as a real absence of a short-wavelength cone cell. Our results for mullids generally agree with Shand (1993) except that she found no UV-sensitive cone cells.

Of the five families in which at least three species were sampled, none had short-wavelength visual pigments present in all of the species. UV and violet single cones were never found together but have been found to coexist in shiners (W. N. McFarland, unpubl. data) and in the atherinid, Menidia menidia (Novales-Flamarique and Harosi, 1999).

All 38 species possessed double and/or twin cones with the following three pairings of visual pigments: blue-green doubles (540–548 and 457–496 nm), identical twin cones (green 501–532 and blue 473 nm), and nonidentical green doubles (506–545 nm). It is doubtful that any of the cells identified as single cones could have been disassociated from a double cone. Our blue single cones looked different than the accessory members of the double cones and were certainly different than twin cone members.

Blue-green doubles were found in nine of the 16 families examined, and 20 of the 26 species within these families also had an UV-, violet- or blue-sensitive single cone, or combinations of a blue- and UV-sensitive single cones. Blue-green-sensitive doubles, in combination with short-wavelength-sensitive single cones, will enhance photosensitivity to the photic habitat around reefs (McFarland and Munz, 1975b; Marshall et al., 2003b).

Six of the 12 species that had identical twin cones lacked nonidentical blue-green doubles. Each species had either a blue-sensitive single cone, as in the squirrelfish, Neoniphon sammara, barracuda, Sphyraena barracuda, moorish idol, Zanclus cornutus, and surgeonfish, Acanthurus triostegus, or a violet-sensitive single cone as in the triggerfish, Sufflamen bursa. The significance of these visual pigment combinations is not clear, but as is the case for all species containing multiple cone spectral classes, the answer may lay in the relationship between visual pigment spectral location and hue discrimination (Marshall et al., 2003b).

Species with nonidentical green-sensitive double cones, especially the surgeonfishes, will have enhanced photosensitivity in the green rather than the short-wave regions of the spectrum. Differences between the two green absorbing pigments averaged about 14 nm. Except for Bodianus bilunulatus and Asterropteryx semipunctatus, blue singles also were recorded from these species. The surgeonfishes we sampled are herbivores (Randall, 2002; Tinker, 1961; Jones, 1968), although some species are planktivorous (Hobson, 1974). These fishes could possess an increase in hue discrimination in the green region of the spectrum or merely a broader sensitivity to green light.

Anatomical location of short-wavelength blockers.—When determining the short wavelength absorption or T50 for a species, it is important to consider all of the components of the eye (Fig. 6) and not just the lens (Douglas and Thorpe, 1992) or even lens plus cornea (Douglas and McGuigan, 1989) as was common practice in earlier studies. The “limiting filter” (Siebeck and Marshall, 2000, 2001) of the eye is that component with the longest wavelength cutoff. “Lens-Limited” eyes are the simplest and most common type found and the lens determines practically all of their spectral transmission properties. Our finding that 73% of the samples have lens-limited eyes agrees well with Siebeck and Marshall's (2001) finding of 80% in Australian coral reef fishes. Lens-limited eyes also tend to have the steepest transmission curve cutoff slopes and longer wavelength T50 values.

“Corneal-limited” eyes have a lens with a cutoff wavelength far shorter than the cutoff of the entire eye and additional filtration is provided in the cornea (review in Douglas and Marshall, 1999). Corneal limitation provides greater flexibility in the variation of T50 values on a seasonal or even daily basis (Kondrashev and Khodtsev, 1984; Siebeck and Marshall, 2000, 2001) without necessitating resorption of blockers from the lens. Corneal filters appear to be of two types. (1) Yellow and other colors, probably formed by carotenoid pigments, may be distributed unevenly around the cornea and may change on a daily cycle are especially common in the wrasses (Siebeck and Marshall, 2000) and block nearly all UV radiation. (2) Short-wavelength UV filters in an otherwise clear cornea may combine a shallow-sloped corneal transmission curve with a shallow-sloped lens transmission curve (Dascyllus albisella; Fig. 6) and reduce the amount of short-wavelength radiation that strikes the retina. These are also found throughout the species sampled and are common in fishes with UV-specialized vision such as some of the damselfish (Losey et al., 2000).

“Humor limited” eyes are rare. Douglas and Marshall (1999) attributed the only two earlier reports of filtration by the aqueous or vitreous humor to post mortem artifacts. Siebeck and Marshall (2001) found two instances in which neither the lens nor the cornea could explain the high T50 and suggested that the humors could be responsible. Humoral limitation has now been found in repeated samples of two fishes, Acanthocybium solandri (Scombridae), the Wahoo or Ono (Nelson et al. 2001), and the priacanthid fish, Heteropriacanthus cruentatus. Both their lens and cornea are relatively transparent to longer-wavelength UV radiation, and the humors have a strong absorption maximum at approximately 395–410 nm and 375–380 nm, respectively, and a steep T50 at 400–418 nm and 373–396 nm, respectively (Appendix 1). Without the absorption of UV by the humors, much shorter-wavelength radiation would reach the retina. The chemical components of the short-wavelength filters of the eye pass through the aqueous humor of the eye and are taken up by the lens and then diffuse into the vitreous humor (Posner, 1998). It is unknown why the lenses of these two species fail to incorporate some of these UV-blocking components found in the humors.

General conclusions and future studies.—Examples of species with short-wavelength sensitivity are common, but not the majority, in tropical reef fish families. The spectral transmission properties of the eye are largely conserved within most families, but exceptions that currently fail to match predictions based on ecological or behavioral characteristics are not unusual. Students of fish behavior and ecology should take caution in interpreting the results of research that involves visual perception systems of species in families such as the Pomacentridae that are highly likely to have UV vision. Presentation of a “model bottle” (e.g., Myrberg and Thresher, 1974) or use of aquarium dividers made of glass or Plexiglas of most types will alter the UV component of coloration and must be avoided in species that have UV vision. At the very least, one must report the short-wavelength transmission properties of model bottles, aquarium dividers, etc., and the incident illumination of the aquarium room. Species whose eyes can be shown to be UV-sensitive as opposed to UV-specialized still pose a problem. For example, if portions of the coloration of a species differ in their UV reflectance (Marshall et al., 2003a), these differences in brightness or luminance will be lost behind the UV-opaque barrier.