Keeping an Eye on the Time The Cogan Lecture

The improver of natural science absolutely refuses to acknowledge authority, as such. For him, scepticism is the highest of duties: blind faith the one unpardonable sin. Thomas Henry Huxley, Addresses and Reviews, 1871. 

The title of this article, and the topic that will dominate this review, relates to the role of the eye in regulating mammalian circadian rhythms. This review, however, will not be confined to a consideration of the circadian system alone. My objective in the following pages is to summarize the development of a line of research that has taken place over approximately 10 years and that has forced us to think very differently about the eye. What started in my group as a fairly narrow consideration of how biological clocks are regulated by light, has now expanded into a much broader interest in how the eye might collect light to generate a measure of the overall amount of light in the environment. The course of these investigations has led to the discovery that varied aspects of physiology and behavior appear to be regulated by non-rod, non-cone ocular photoreceptors. I will outline the evidence that supports this radical conclusion, review what we know about this novel photoreceptor, and draw attention to some of the broader implications of these results. Because much of this work has its origins in the study of biological clocks, this is where I start the discussion. 

The primary role of the circadian system is to set the phases at which physiological and behavioral events occur with respect to the 24-hour environmental cycle or in relation to rhythmic events within the organism. Thus, the circadian system must remain synchronized, or entrained, to the solar day. When entrained, the circadian clock adopts a distinct phase relationship with the astronomical day, and each of the different expressed rhythms adopts its own phase relationships with the clock. The systematic daily change in the irradiance of light at dawn or dusk provides the most reliable indicator of the phase of the day. As a result, most organisms have evolved to use the twilight transition as the primary zeitgeber to adjust circadian phase (photoentrainment). ¹ But, of course, environmental irradiance does more than regulate biological time. Varied aspects of mammalian physiology, endocrinology, and behavior respond to gross changes in environmental light. For example, pineal melatonin production, pupil size, adrenal cortisol secretion, heart rate, and body temperature are all affected by irradiance. ^{2 3 4 5} Furthermore, in our own species, increasing environmental irradiance can result in marked improvements in alertness and performance ⁶ and changes in electroencephalic (EEG) and electro-oculogram (EOG) correlates. ⁷  

It is important to note that the circadian and physiological responses just mentioned are modified by gross changes in environmental light, rather than by specific patterns of light. In addition, the thresholds for eliciting these responses are higher, and the time over which the light stimuli are integrated is significantly longer, than for classic vision. ⁸ For example, hamsters can recognize optical gratings at a luminance level 200 times lower than the irradiance required to induce phase shifts in locomotor rhythms. ⁹ In addition, hamsters are relatively insensitive to an entraining light stimulus of less than 30 seconds’ duration. ^{10 11} Such observations have led to the appreciation that the eye performs two very different light-detecting tasks that can be likened to physical detectors that measure either radiance or irradiance (Fig. 1) . For the purposes of brevity, the primary photosensory task of irradiance detection is referred to as non-image-forming (NIF) photoreception, and classic visual responses are referred to as image-forming (IF) photoreception. 

An anatomic separation between NIF and IF pathways, for at least some irradiance-detection tasks, has been demonstrated by the observation that irradiance responses are regulated by specific brain nuclei that receive distinct retinal projections. For example, the primary brain nuclei of the circadian timing system are the suprachiasmatic nuclei (SCN) and intergeniculate nucleus (IGL). Both of these structures contain cells that exhibit light-dependent activity, suggesting that they function as “irradiance detectors.” ^{12 13 14} The SCN and IGL receive retinal axons via the retinohypothalamic tract (RHT) which is formed from a small number (approximately 0.1%) of morphologically homogeneous (type III) retinal ganglion cells (RGCs). The type III RGCs that project to the SCN and IGL have the largest dendritic arbor and, consequently, the largest receptive field of all ganglion cell types. ¹⁵ They are also distributed over the entire retinal expanse and send a retinotopically unmapped projection to the SCN. ^{16 17 18 19 20} These anatomic features would result in the spatial integration of light that would effectively “dilute” photons from bright light sources, such as the moon or stars, that could interfere with stable entrainment to solar irradiance. ²¹  

Significantly, type III ganglion cells have also been shown to innervate several other regions of the brain that respond to gross changes in environmental light. The olivary pretectal nucleus (OPN) is involved in pupilloconstriction ^{22 23} and the consensual pupillary light reflex ²⁴ and receives substantial input from type III RGCs. ²¹ Along with the SCN and IGL, the OPN is the only other dedicated irradiance detector described in any detail within the mammalian brain. ^{22 23} Another target for type III RGCs is the ventrolateral preoptic nucleus (VLPO), a region of the rodent brain that appears to be involved in sleep-wake regulation. ²⁵ Environmental irradiance regulates sleep through the circadian system, but also has effects that are quite independent of the clock. For example, sustained exposure to light increases non-rapid eye movement (NREM) sleep in rats, ^{26 27 28} and darkness delivered during the light phase of a 24-hour light-dark cycle can trigger rapid eye movement (REM) sleep. ^{28 29} It has been argued recently that these acute effects of light on sleep may be mediated by the type III RGC projection to the VLPO. ²⁵  

In contrast to the SCN, IGL, OPN, and VLPO, the IF visual system consists of subcortical pathways to the geniculate complex and superior colliculus. These projections are highly ordered and retinotopically mapped. Thus, the mammalian eye has parallel outputs, encoding predominantly NIF or IF information. Until recently, there was no reason to expect that these divergent projections from the eye would use distinct classes of photoreceptors. The very reasonable assumption was that retinal rods and cones mediate both NIF and IF light detection. 

A Novel Ocular Photoreceptor in Mammals

Unlike mammals, nonmammalian vertebrates possess several photoreceptor organs in addition to their ocular photoreceptors. These extraocular photoreceptors mediate NIF responses to light and have been classified as the pineal organ or pineal body (epiphysis cerebri), which is photoreceptive in all nonmammalian vertebrates; the parapineal organ, found in many teleost fish and well-developed in the lamprey; the extracranial “third eye,” called a frontal organ in frogs and the parietal eye/body in lizards (missing in birds); and deep brain photoreceptors, which are located in several sites of the brain and are found in all nonmammalian vertebrates. ³⁰ Note that these extraocular photoreceptors are more complex than one might imagine. For example, the pineal organ of fish and birds contains multiple photopigments, expressing rod- and conelike, as well as novel, photopigments within the same organ. ^{31 32 33 34 35} Of these extraocular photoreceptors, the deep brain and pineal photoreceptors have been shown to play a critical role in the regulation of circadian clocks (for a full review see Ref. ³⁰ ). Mammals are unique among vertebrates, in that they appear to have lost extraocular photoreceptors. ³⁶ Why this has occurred is unclear, but it may be due to the early evolutionary history of mammals and their passage through what has been called a “nocturnal bottleneck.” ³⁷ Modern mammals seem to have been derived from nocturnal insectivorous or omnivorous animals approximately 100 million years ago. Extraretinal photoreceptors may not have been sufficiently sensitive to discriminate twilight changes in primitive mammals spending their days hiding in holes or crevices and then emerging at dusk. ^{38 39}  

The existence of extraocular photoreceptors in nonmammals and their role in regulating circadian rhythms in these vertebrate classes reinforced the view that extraocular photoreceptors regulate NIF responses to light, whereas the eye is dedicated to the collection of IF information. By extension, it was assumed that because the mammals had lost their extraocular photoreceptors they were “forced” to use ocular rods and cones for both IF and NIF responses to light. Until recently, there was no reason to doubt this assumption; however, studies on the mammalian circadian system have provided the substrate to challenge this view. 

Photoentrainment of the Mammalian Circadian System

The overwhelming mass of experimental evidence tells us that the circadian system of mammals is entrained by photoreceptors within the eye. However, some mention should be made of a report in Science that suggests that bright light of approximately 13,000 lux applied to the popliteal region (skin behind the knee) can induce a shift in circadian rhythms of body temperature and melatonin. ⁴⁰ There has been much reluctance to accept these results by most circadian and vision biologists, not least because eye loss in humans ^{41 42} and other mammals (e.g., Refs. ^{43 44 45} ) always blocks photoentrainment. Nevertheless, the implications of these findings were so important that several groups have tried, and continue to try, to provide independent support for the possibility that dermal irradiation can modify the circadian physiology of humans and other mammals. To date, however, results in all such studies have failed to support this possibility. For example, popliteal illumination produced no effect on the suppression of nocturnal levels of pineal melatonin in human subjects, ⁴⁶ and exposing the chest and abdomen to 13,000-lux broad-spectrum light did not shift rhythms of melatonin, cortisol, and thyrotropin. ⁴⁷  

Although clearly ocular, ⁴⁸ disentangling which retinal cells mediate photoentrainment has presented a problem. The use of mice with naturally occurring genetic disorders of the eye, however, provided a way forward. The original intent was to correlate rod and cone photoreceptor loss with a loss of sensitivity in the circadian system to light. The first animals used for such studies were mice homozygous for retinal degeneration (rd/rd). Despite the massive (but not complete) loss of rods and cones in rd/rd mice ^{49 50} these animals showed circadian responses to light that were indistinguishable to nondegenerate congenic control mice (rd ⁺, rd ^+/+). The sensitivity of the circadian system to light did not parallel the loss of either rod or cone photoreceptors, ⁵¹ but removal of the eyes abolished all circadian responses to light in rd/rd mice. ⁴⁴ Our studies on the rd/rd mice contradict an earlier report that suggested that this mutation would attenuate circadian photosensitivity. ⁵² This earlier report showed that the threshold for entrainment in C57 wild-type mice was approximately 2 log units lower than the threshold for entrainment in C3H rd/rd mice. This difference, however, was later shown to be an artifact of comparing different strains of mice rather than an effect of the rd/rd mutation. When congenic C3H^+/+ and C3H rd/rd mice are examined under the same experimental conditions, no reduction in circadian photosensitivity is observed. ^{53 54} This example illustrates the importance of controlling for genetic background when examining the impact of a gene defect. ^{55 56 57}  

Our findings in rd/rd mice and parallel findings in mice homozygous for retinal degeneration slow (rds) ⁵⁸ suggest that at some level the processing of light information for IF and NIF photoreception must be profoundly different. Additional support for a distinction between these photic pathways has come from studies in humans. Two reports have demonstrated that a significant subset of individuals who have eyes but have lost conscious light perception due to retinal disease have retained the ability to suppress melatonin ⁴¹ and entrain circadian rhythms of behavior. ⁴² Although the findings in retinally degenerate mice and these observations in humans were surprising, they did not demonstrate the existence of novel ocular photoreceptors. Because small numbers of rods and/or cones remain in the retina of these retinally degenerate mammals, these experiments were unable to resolve whether the circadian axis is able to retain normal photosensitivity using only very small numbers of rods and/or cones. To investigate this, we generated two lines of transgenic mice in which both rod and cone photoreceptors were absent (Fig. 2) . This was achieved by introducing a synthetic transgene (cl) consisting of an attenuated diphtheria toxin coding sequence driven by a portion of the human red cone opsin promoter ^{59 60} into two lines of mice that had no rods. We used either rd/rd mice or mice that carried a separate diphtheria toxin-based transgene (rdta). ⁶¹ Despite the absence of both rod and cone photoreceptors, rd/rd cl and rdta cl mice retain superficially normal circadian responses to light, as demonstrated both by the phase-shifting effects of a single 15-minute pulse of light on free-running rhythms of circadian locomotor activity and by the acute inhibition of pineal melatonin production. ^{62 63} As with rd/rd mice, enucleation of the eyes in rd/rd cl or rdta cl mice abolishes these effects of light on the circadian system. These results demonstrate that the circadian system can use unidentified ocular photoreceptors. Furthermore, our most recent studies show that two other NIF light responses, pupillary constriction and acute alterations in locomotor behavior, also survive the loss of rod and cone photoreceptors. 

Pupillary Light Responses

In mammals, light-induced pupillary constriction is regulated by both rods and cones ^{64 65 66 67} but also survives the profound loss of these classic photoreceptors. For example, rats and mice with either inherited or environmentally induced photoreceptor degeneration show robust pupillary constriction in response to light stimuli. ^{67 68 69 70} As in the early circadian studies, the residual pupillary photosensitivity was thought to depend on the survival of a few rod and/or cone photoreceptors. ⁶⁷ These observations, together with our finding that type III RGCs innervate a brain nucleus involved in pupilloconstriction (OPN; see the introduction), prompted us to investigate pupillary light responses (PLRs) in rd/rd cl and wild-type mice. Pupilloconstriction was observed in rd/rd cl mice, showing that rods and cones are not required for this light response. However, there is a clear impact of rod and cone loss. Both genotypes showed maximal pupillary constriction after bright light exposure, but a comparison of the irradiance response curve for each of the genotypes showed that rd/rd cl mice are less sensitive to light when compared with wild-type mice (Fig. 3) . Indeed, irradiances that trigger the PLR in rd/rd cl mice come close to saturating the PLR of wild-type mice. ⁷¹ There have been some reports, although not in mice, that the mammalian iris may be directly light sensitive. ^{72 73} To determine whether central pathways are responsible for the PLR in rd/rd cl mice, a series of experiments were undertaken. ⁷¹ The strongest evidence in support of a central mechanism was the demonstration of a consensual response. Illumination of the left eye resulted in pupillary constriction in the right. ⁷¹  

On the basis of these data we have developed the following working hypothesis. We suggest that the normal PLR in mice involves a two-stage process. When a subject moves from dim to bright light, the rods and cones provide the initial input for pupillary constriction, whereas the NIF photoreceptors allow pupillary constriction to be maintained under relatively high and sustained environmental irradiances, a task poorly suited to the rapidly habituating rod and cone responses. If this hypothesis is correct, we predict that animals with defective NIF photoreceptors would be unable to maintain pupillary constriction under sustained, relatively bright light conditions. Should this be the case, then pupillary state may provide a powerful screen for identifying mice with defective NIF photoreceptors. 

The Regulation of Masking Behavior

Nocturnal rodents show a marked inhibition of locomotor activity when exposed to light during the night. ⁷⁴ This response is thought to complement photoentrainment by ensuring that activity is restricted to the hours of darkness. As a result, it has been commonly referred to as “circadian masking” or simply “masking.” ⁷⁵ Previous reports have indicated that masking survives significant photoreceptor loss. ^{76 77} However, the use of rd/rd cl mice allowed us to ask whether masking can occur in animals that have no rod and cone photoreceptors. ⁷⁸ Both wild-type and rd/rd cl mice exhibited a marked inhibition of wheel-running activity during acute exposure to a broad-spectrum green light presented 2 hours after normal lights off (Fig. 4A) . At high irradiances, activity was reduced to around 10% of normal wheel running in both genotypes. These effects on activity were irradiance dependent (Fig. 4B) , and there was no significant difference in sensitivity associated with rod and cone loss. ⁷⁸  

What Else Might be Regulated by NIF Photoreceptors?

Our initial studies using rd/rd cl mice have shown that circadian photoentrainment, the inhibition of pineal melatonin production, pupillary constriction, and masking, survive the complete loss of rod and cone photoreceptors. The survival of such diverse responses to light raises the question of how general the influence of these novel photoreceptors might be on mammalian physiology and behavior. As outlined in the introduction, different aspects of mammalian physiology and behavior are regulated by environmental irradiance. The finding that type III RGCs innervate the VLPO suggests that novel photoreceptors may also directly modulate sleep activity. ²⁵ We are exploring this possibility by studying light activation of VLPO neurons in rd/rd cl mice. In humans, bright light exposure elevates cortisol levels, ³ increases heart rate, ⁷⁹ and modulates alertness and even mood. ⁷ Could these responses all be influenced by non-rod, non-cone ocular photoreceptors? Again, studies on the rd/rd cl mouse will allow us to assess the likely contribution of novel photoreceptors to some of these physiological and behavioral responses in our own species. 

A Novel Ocular Photopigment in Mammals

Beyond the fact that novel photoreceptors exist and that they may regulate diverse aspects of our “vegetative” responses to light, what do we know about these receptors? Photoreceptor characterization has traditionally been based on a number of complementary approaches (see Critical Criteria, to follow), but perhaps the most useful single technique is to conduct an action spectrum to define the nature of the photopigment. A photopigment has a discrete absorbance spectrum, which describes the probability that photons are absorbed as a function of wavelength, and the magnitude of any light-dependent response depends on the number of photons absorbed by the photopigment. Thus, a description of the spectral sensitivity profile (action spectrum) of any light-dependent response must by necessity match the absorbance spectra of the photopigment mediating the response, provided that any confounding factors, such as screening pigments or absorption of the ocular media, are taken into account. ^{80 81}  

Although light can be used in a variety of ways (as an energy source, as a sensory stimulus, or as a regulatory signal), relatively few photopigments appear to have evolved. ⁸² This may relate to the demanding task of a photopigment molecule. It must be able to absorb a photon with high probability and, again with high probability (high quantum efficiency), pass on this information to a transduction mechanism. ⁸³ In nature, a wide variety of light responses are regulated by only a few types of photopigment, and these have highly conserved absorbance spectra. For example, both the action spectra for phototaxis in the algae Chlamydomonas and visual photosensitivity in humans are described by the same standard absorbance spectrum for an opsin-vitamin A photopigment. ⁸³ In the same way, flavoprotein-based photoresponses can be described by conserved absorbance spectra. ^{82 83} It is worth stressing that when an action spectrum is conducted correctly, it provides one of the few approaches that can be used to identify a class of protein and simultaneously link that protein to a particular cellular or behavioral function. 

Because action spectra reflect photopigment absorbance spectra and because individual photopigment families have characteristic absorbance spectra, the involvement of candidate pigments can be assessed using action spectrum techniques. ^{82 83} In an attempt to characterize the photopigment(s) mediating NIF responses to light, we have initiated a series of action spectra on rd/rd cl and wild-type mice. The first completed was an action spectrum for the PLR. ⁷¹ The PLR is particularly well suited for spectral analysis, in that pupillary constriction is relatively easy to monitor and shows only a small degree of interanimal variance. ⁷¹ This permitted us to generate a well-resolved action spectrum based on the irradiance required to induce a 50% pupil constriction at 10 monochromatic wavelengths between 421 and 625 nm. This action spectrum was then compared to the absorbance spectra of the known photopigments of the mouse (Fig. 5) . The results demonstrate that the PLR in rd/rd cl mice, over the range 420 to 625 nm, is driven by a single, previously uncharacterized, opsin-vitamin A-based photopigment with peak sensitivity at 479 nm (opsin photopigment [OP⁴⁷⁹]). These data represent the first functional characterization of a non-rod, non-cone photoreceptive system in the mammalian eye. ⁷¹ Whether OP⁴⁷⁹ mediates all non-rod, non-cone NIF responses to light remains to be determined. A preliminary action spectrum for phase-shifting circadian rhythms of locomotor behavior in rd/rd cl mice suggests the involvement of an opsin-vitamin A-based photopigment with a wavelength of maximum sensitivity significantly different from the mouse visual opsins, but very close to that of OP⁴⁷⁹ (Thompson C, Lucas R, Foster R, unpublished observations, 2002). A more complete action spectrum for phase shifting and additional action spectra for those light responses preserved in rd/rd cl mice (e.g., melatonin suppression and masking behavior), will determine the extent to which OP⁴⁷⁹ contributes to the light responses of the mouse. 

Additional evidence for the presence of novel ocular photopigments in mammals comes from three very recent studies in humans. Two of these studies independently determined an action spectrum for light-induced melatonin suppression in normal-sighted subjects. One study suggests the involvement of a novel opsin-based photopigment with peak sensitivity between 446 and 477 nm. ⁸⁴ The other also implicates a novel opsin-based photopigment and, in this case, with peak sensitivity between 457 and 462 nm. ⁸⁵ In addition, an action spectrum from our group has also identified a novel photopigment with a maximum sensitivity at 483 nm, which provides the irradiance signal that regulates the temporal properties of the cone pathway in the human retina. ⁸⁶ The similarity of all these three human studies to each other, and to mouse OP⁴⁷⁹, suggests that humans possess an ortholog of mouse OP⁴⁷⁹. 

So, which photoreceptor cells of the eye use OP⁴⁷⁹? In the absence of an outer retina in rd/rd cl mice and knowing that light information is transmitted to the SCN, IGL, and OPN through type III RGCs, it seems that these photoreceptors must reside within the inner retina. Two lines of evidence have suggested that a subclass of RGC may be directly light sensitive. Indirect and preliminary evidence comes from our studies in the aged rd/rd cl mouse. ⁸⁷ After 9 months of age rd/rd cl mice start to lose much of the inner retina, so that by 18 to 24 months of age, there is little left of the inner retina other than the RGC layer. Despite this loss, rd/rd cl mice show both circadian and pupillary responses to light. Furthermore, a subset of RGCs in these aged rd/rd cl mice show induction of the immediate early gene c-fos in response to light exposure. On the basis of these data, we proposed that a subset of ganglion cells house the novel ocular photoreceptors. ⁸⁷ More direct evidence for photosensory ganglion cells comes from an exciting study by Berson et al. ⁸⁸ at Brown University (Providence, RI). Rat RGCs were retrolabeled from the SCN with rhodamine beads, then recorded by whole-cell current clamp in isolated retinal flatmounts. Light was shown to tonically depolarize most SCN-projecting RGCs, and, significantly, these responses persisted when these cells were isolated from rod and cone inputs using both microsurgical isolation and a cocktail of agents that block synaptic transmission. ⁸⁸  

Inner Retinal Photoreception in Nonmammals

Thus far, the discussion of the existence of a novel opsin-based photoreceptor within the inner retina has been limited to the mammals. However, parallel experiments on the retina of teleost fish provide independent support of the existence of a non-rod, non-cone ocular photoreceptor in vertebrates. 

In an effort to characterize the extraocular photoreceptors of Atlantic salmon, we developed degenerate PCR primers to amplify opsin cDNAs from the salmon central nervous system (CNS). To optimize our PCR procedures, we first tested these degenerate primers using salmon ocular cDNA. An initial screen identified a cDNA whose conceptual translation shared only 37% to 42% identity with any of the known opsin families. This level of identity immediately isolated this opsin into its own, previously unrecognized opsin family ⁸⁹ (Table 1) . 

A phylogenetic analysis of this opsin indicated that it diverged from a common ancestor before any of the known opsin families, and on this basis we called the novel family the vertebrate ancient (VA) opsins. ⁸⁹ Although the deduced VA opsin sequence has all the features that would be necessary to form a functional photopigment, this can never be assumed (see later section, The Opsin Candidates and Critical Criteria). To this end, we expressed VA opsin within COS cells and regenerated VA photopigment, with 11-cis -retinal (vitamin A₁), analyzing the spectral characteristics of in vitro-generated VA pigment and its photoproducts by difference spectroscopy. These experiments demonstrated that VA opsin can form a photopigment with photoproducts similar to previously described visual pigments. ⁹⁰ The sites of VA opsin expression were determined using in situ hybridization. Significantly, VA opsin was never observed in the retinal rods or cones but was restricted to a subset of cells with a location and morphology characteristic of horizontal and amacrine cells (Fig. 6) . The number of horizontal cells expressing VA opsin varies across the retina, but is always greater in number than the VA opsin-expressing amacrine cells. ⁹⁰ Furthermore, VA opsin is also expressed in cells of the pineal and subhabenular, areas of the brain previously implicated in fish photoreception. ^{31 90 91} VA opsin was initially isolated from Atlantic salmon but has now been isolated from the carp, zebrafish, and lamprey. ^{33 91} In some species at least, VA opsin seems to be expressed in two variants or isoforms with a short (VAS) or long (VAL) carboxyl terminus. ^{33 91} The production of the short VA opsin isoform seems to be achieved by alternative splicing of intron 4 of the VA opsin gene. Currently, the role of VA opsin photopigments in the retina, pineal, or habenular is unknown. However, our electrophysiological studies are beginning to characterize the responses of VA photoreceptors. ⁹²  

It seems more than a little ironic to us that in fish, we found by accident an inner retinal photoreceptor, but have little idea what this photoreceptive system might do. Whereas in mammals, considerable resources are being invested by a number of laboratories in an attempt to identify a novel photoreceptor gene within the inner retina! 

Candidate Novel Photopigment Genes in Mammals

As discussed in the foregoing section regarding a novel ocular photopigment in mammals, behavioral studies and action spectrum results in rd/rd cl mice have provided overwhelming evidence for the existence of a novel opsin-based photopigment within the eye. Results from aged rd/rd cl mice and exciting studies emerging from David Berson’s laboratory suggest that some retinal ganglion cells may be directly light sensitive. These findings in mammals, together with the VA opsin results in teleosts reviewed in the section just prior, are all consistent with the hypothesis that the rods and cones are not the only photoreceptive cells within the vertebrate retina. But further analysis of this photoreceptor pathway is going to be heavily dependent on the identification of the photopigment gene(s) and protein product. In recent years, two classes of photopigment molecule have been proposed for these novel photoreceptors, the opsins and the cryptochromes. 

The Opsin Candidates

The isolation of VA opsin in fish immediately led several groups to look for orthologues of this gene in mammals. To date, this search has been unsuccessful. However, several other opsin genes have been identified and proposed as candidates. ⁹³ It is also clear from our own investigations that yet more opsin candidates will emerge from the human and mouse genome data bases. But how are we to assess the candidacy of these genes based on their sequence identity alone? What features should guide the search for a novel photopigment gene? This is more complex than might first be thought. All the known opsin-based vertebrate photopigment families share approximately 40% amino acid identity with each other, approximately 25% identity with the invertebrate photopigments, and approximately 17% with all other G-coupled receptors (Table 1) . Other than levels of overall identity, surprisingly few features of the photopigment opsins distinguish them from other members of the G-coupled receptor superfamily. Two features, however, have been considered critical: the possession of a lysine residue at position 296 for Schiff base attachment of the chromophore, and a glutamine at position 113 as a charge counterion for the chromophore. ⁹⁴ Unfortunately, these features cannot be considered diagnostic for an opsin photopigment. Not all photopigment opsins use a glutamine counterion. For example, some members of the short-wavelength visual pigment family use an aspartate at this site. ⁹⁵ Furthermore, the lysine residue at 296, which is essential for chromophore attachment, is also found in opsins that are not thought to have a photosensory function. For example, the retinal G protein-coupled receptor (RGR) differs from the photopigment opsins in that its preferred chromophore is all-trans-retinaldehyde rather than 11-cis-retinal. On exposure to light, RGR photoisomerizes the all-trans chromophore to the 11-cis configuration. This known function, coupled with the fact that RGR is expressed in the retinal pigment epithelium (RPE) and Müller cells ⁹⁶ and that RGR shares only 22% identity with the visual pigment opsins (Table 1) , suggests a role as a photoisomerase rather than a sensory photopigment. ⁹⁷ We are left therefore with few strong guidelines to direct the search for novel opsin photopigment genes. 

Of the various opsin candidates proposed recently, ⁹³ melanopsin has emerged as the most interesting. ^{98 99 100 101} It can be argued that melanopsin is not a strong candidate based on its level of identity; it shares only 27% identity with the known vertebrate photopigments (Table 1) . Furthermore, the genomic structure of this gene bears no resemblance to the known photopigment opsins of the vertebrates. ¹⁰¹ Both of these features contradict the argument that functionally related molecules are likely to share a close phylogenetic relationship based on both amino acid identity and genomic structure. However, melanopsin is interesting, because in many respects it resembles an invertebrate photopigment. Moreover, it is expressed in very few RGCs and in even fewer cells within the amacrine cell layer in the mouse retina. This distribution is strikingly similar to the distribution of murine RGCs that form the retinohypothalamic tract, and project to the OPN. ²¹ In a very recent study, rat RGCs that show intrinsic photosensitivity have been shown to express melanopsin. ¹⁰² Thus melanopsin, at the very least, is a good marker of a subset of inner retinal photoreceptors. The on-going analysis of this fascinating gene, including its functional expression, will determine whether melanopsin is OP⁴⁷⁹ or whether it performs some other function within the eye, perhaps acting as a photoisomerase and part of an as yet unidentified opsin photopigment system within these photosensitive RGCs. 

The Cryptochromes

An alternative hypothesis to the involvement of opsin-based photopigments is that the cryptochromes act as photopigments within the inner retina. It has been argued that the mammalian eye contains a class of flavoprotein-based photopigment called cryptochrome (CRY). ¹⁰³ The origins of this hypothesis have, to a large extent, arisen from the presumed photoreceptive role for the CRYs in other organisms. ¹⁰⁴  

CRY-like proteins are found in diverse phyla and belong to a group of proteins that bind both a pterin (methenyltetrahydrofolate [MTHF]) and a flavin (7,8-didemethyl-8-hydroxy-5-deazariboflavin [8-HDF]). On the basis of sequence similarity and defined function, these pterin-flavin-binding proteins can be subdivided into the photolyases and cryptochromes. ¹⁰⁵ Perhaps the best characterized are the photolyases, which repair UV damaged DNA and of which there are two types. The pyrimidine dimer photolyases harvest the energy of near-UV light to fix UV-induced cyclobutane pyrimidine dimerization in DNA. ¹⁰⁶ The other group is the (6-4) photolyases, which perform a similar function but act specifically on UV-induced pyrimidine (6-4) pyrimidine dimers. ¹⁰⁷ In contrast to the photolyases, the cryptochromes use near UV light for a non-DNA repair function. The cryptochromes were first named in the plants and appear to act as blue/UV photopigments and contribute to several physiological responses. ^{108 109 110 111} Two CRY proteins have been identified in plants, CRY1 and -2. ¹⁰⁵ Disruption of cry1 attenuates a number of blue light responses, including shortening of the circadian period by constant blue light. ¹¹² Disruption of cry2 again disrupts a number of photic responses, including the regulation of flowering by day length. ¹⁰⁹ Clearly, the plant CRYs are involved in a number of blue light-regulated processes, and it has been argued that these proteins must act as photopigments. ¹⁰⁵ However, we do not currently have sufficient experimental evidence to say whether the plant CRYs are acting as photopigments or as components of a light-input pathway, perhaps coupling a photopigment to a transduction cascade (see the following section, Critical Criteria). 

A single CRY-like gene has been identified in Drosophila (dcry) ¹¹³ and two CRY-like genes in mammals (cry1 and cry2). ¹¹⁴ Note that the naming of the animal cryptochrome genes is a little misleading. The animal CRYs show a much higher degree of identity with the (6-4) photolyases than to the plant cryptochromes, but despite their overall similarity to the (6-4) photolyases, they show no apparent photolyase activity. In addition, the mammalian CRYs have a COOH-terminal extension that is absent in the photolyases but is similar to the COOH extension found in the plant CRYs. These features (absence of photolyase activity and the COOH tail) were used to group the plant and animal CRYs together and then to suggest that the animal (Drosophila and mammal) CRYs are photopigments that mediate blue light’s effects on the circadian system. ^{104 105 115 116}  

Genetic manipulation of dcry has been shown to modify the effects of light on the circadian system. For example, underexpression of dcry reduces the size of phase shifts, ¹¹⁷ whereas overexpression of dcry leads to larger phase shifts. ¹¹³ Additional evidence that CRY contributes to photoentrainment comes from studies of a mutation in the dcry gene that has been called cry ^baby (cry ^b). ¹¹⁸ Although cry ^b mutants can entrain their locomotor rhythms almost normally and rhythmic clock gene expression can be entrained in the lateral neurons of the brain, ¹¹⁸ mutants are unable to reset their clocks to short light pulses, ¹¹⁸ and cry ^b mutants do not become arrhythmic under continuous light conditions. ^{115 116} In Drosophila, as in plants, CRY seems to act somewhere on the light-input pathway to the clock and may even act as a photopigment, acting both to absorb light and transduce light information directly to the molecular clock. However, critical photobiological experiments have yet to be undertaken that distinguish between CRY’s acting as a photopigment or as an element of the phototransduction cascade (see Critical Criteria). The obvious approach to resolve this issue would be to match action spectra in Drosophila to CRY absorbance spectra or to the absorbance spectra of other flavoprotein photopigments. ⁸³ Unfortunately, this has not yet been accomplished with any degree of precision. It is also important to note that although CRY may act as a photopigment in the entrainment of Drosophila circadian rhythms, it is not the only photopigment. Opsin-based photopigments in the compound eyes and the Hofbauer-Buchner eyelet are strongly implicated in photoentrainment in Drosophila ⁵⁴ and other insects. ¹¹⁹  

In contrast to the situation in Drosophila and despite claims that the CRYs mediate all of light’s effects on the mammalian clock, ¹⁰³ there is no direct evidence that these proteins contribute to photoentrainment or to any other photosensitive process in mammals. The mammalian CRYs appear to play a central role in the generation of the circadian oscillation itself. The strongest evidence for this comes from the work of van der Horst et al. ¹²⁰ This group showed that cry1 ^−/− cry2 ^−/− mice are completely arrhythmic, with no indication of a functional circadian clock under conditions of continuous darkness. When exposed to a light-dark cycle, cry1 ^−/− cry2 ^−/− mice confined most of their activity to the dark portion of the cycle but showed no evidence of anticipating light-dark transitions (as do normal mice). Suggesting that the behavior observed was simply masking (see earlier section on regulation of masking behavior) and driven by exposure to light and dark rather than being influenced by an endogenous clock. Significantly, the disruption of the cryptochrome genes in cry1 ^−/− cry2 ^−/− mice does not block the light-induced expression of the two clock genes mPer1 and mPer2 in the SCN. ¹²¹ Thus, on the basis of the current evidence, the mammalian CRYs appear to act as components of the molecular clock. ¹²² We speculated that the animal CRYs may contribute to clock function by mediating REDOX reactions. ¹²³ Exciting new results from the McKnight laboratory have provided evidence that this may indeed be the case. ^{124 125}  

Some researchers argue, however, that the mammalian CRYs have a dual function as both components of the oscillator and as photopigments. Indeed, a recent report appeared to support this possibility. In an attempt to define the photoreceptors that contribute to masking behavior, rd/rd mice (see foregoing section, Photoentrainment of the Mammalian Circadian System) were crossed with cry1 ^−/− cry2 ^−/− mice to generate mice without classic photoreceptors and CRYs (although rd/rd mice still have some cones). In contrast to young cry1 ^−/− cry2 ^−/− mice, which confine most (80%) of their activity to the dark portion of the light-dark cycle, the combined rd/rd cry1 ^−/− cry2 ^−/− mice spread their activity more evenly between the light and dark, with 37% to 40% of their activity occurring in the light phase. ¹²⁶ These results suggest that masking is abolished in rd/rd cry1 ^−/− cry2 ^−/− mice, and by extension, that classic and CRY photoreceptors contribute in a redundant manner to masking. This interpretation, however, does not take into account the following results. Approximately 30% of rd/rd cry1 ^−/− cry2 ^−/− mice are still able to mask. Furthermore, some mice that can mask at 5 to 6 months lose this ability by 9 to 12 months ¹²⁷ (Van Gelder RN, written communication, 2001). These results suggest that the loss of masking may be independent of the direct effect of losing CRY, the rod photoreceptors, and most cone photoreceptors. This conclusion is supported by two independent observations showing that masking is frequently lost in mice more than 6 months of age who lack only their CRY proteins (cry1 ^−/− cry2 ^−/−) (van der Horst GTJ and Mrosovsky N, written communication, 2001). Thus, the loss of masking in cry1 ^−/− cry2 ^−/− and rd/rd cry1 ^−/− cry2 ^−/− mice may be due to some secondary effect of CRY loss. One explanation could be that CRY loss precipitates an ocular disease that results in the secondary loss of the photoreceptors that mediate masking. These rather complicated arguments are summarized in Table 2 . 

Critical Criteria

The foregoing discussion emphasizes that the unambiguous identification of a photopigment has to rely on more than the disruption of candidate genes. Photopigment identification has traditionally been based on a number of criteria, and these are summarized herein. Ideally, all these criteria should be fulfilled in the characterization of a light response:

1.  
   The candidate should form a functional photopigment (e.g., Ref. ⁹⁰ ) responding to photic rather than nonspecific kinetic energy, such as heat or denaturing ultraviolet light. ⁸⁰ If a response were mediated by heat or infrared energy, then it would be abolished by filtering the light stimulus with a heat filter.
2.  
   The candidate photopigment should have an absorbance spectrum that matches the action spectrum of the response in question. Evidence from action spectra have been the single most useful way of assigning a response to a particular photopigment, and the use of action spectra in identifying photopigments has had a long and successful history. ⁸² For example, the recent action spectra in rd/rd cl mice demonstrate the involvement of a novel opsin-based photopigment rather than a flavoprotein, such as cryptochrome (Fig. 7) .
3.  
   The candidate molecule should be expressed in organelles, cells, tissues, or organs defined as photoreceptors using physiological assays. These assays would include direct illumination, electrophysiological recording, or ablation. In this respect, the coexpression of melanopsin with those RGCs showing endogenous light responses is very significant. ¹⁰²
4.  
   Genetic ablation of the candidate gene. When the candidate photopigment is genetically ablated, the response to light should be either lost or attenuated. If the response is merely attenuated, it is then critical to show that the action spectrum of the response is altered in a manner that would be predicted on the basis of the absorbance spectrum of the candidate photopigment. ⁶¹ In the absence of criteria (1) and (2), gene ablation studies can be used only to associate a gene with a light-dependent process and cannot distinguish between the loss of the photopigment and/or loss of an element in the phototransduction cascade.
5.  
   Chromophore identification or depletion. The chromophore of several photopigments is associated exclusively with photoreception. For example, 11-cis retinaldehyde is a specific form of vitamin A associated only with the opsin-based photopigments. 11-cis retinaldehyde can be readily identified by using HPLC, and its identification ^{128 129} or depletion ¹³⁰ has been helpful in defining the nature of photoreceptive pathways. However, great care must be taken in the interpretation of chromophore depletion experiments. As discussed previously, ¹³⁰ vitamin A depletion is not equivalent to vitamin A elimination. For example, Drosophila raised on diets without β-carotene showed a decline in visual photosensitivity of only 3 log units. ¹³¹ If all chromophore had been eliminated from these flies, then visual responses should have been abolished. Thus, the interpretation of vitamin A depletion experiments should be suitably cautious. ¹³²
6.  
   The candidate gene may share homology to known photopigment molecules. For example, during the course of vertebrate evolution, amino acid changes (along with gene duplications) have led to the formation of several distinct opsin families that share approximately 40% amino acid identity (Table 1) . But as discussed earlier (The Opsin Candidates), assigning function on the basis of sequence similarity alone can be very misleading.

Conclusions

We now have overwhelming evidence that a variety of NIF responses to light are regulated by non-rod, non-cone ocular photoreceptors within the mammalian eye, and evidence from pupillary action spectra show that at least one NIF response is regulated by a novel opsin-vitamin A-based photopigment. Furthermore, the recent findings showing that a subset of RGCs are intrinsically photosensitive in the rat retina supports the conclusion that the rods and cones are not the only cells within the mammalian retina capable of photoreception. However, these finding do not demonstrate that the classic rod and cone photoreceptors play no role in NIF responses to light. Indeed, there may be a close interaction between rods, cones, and novel photoreceptors in the regulation of the murine pupil. ⁷¹ Furthermore, several studies have indirectly implicated classic photoreceptors in photoentrainment. ^{133 134 135} Presumably, this multiplicity of photoreceptor inputs to the clock relates to the complex sensory task of twilight detection. ¹ A decade of research now allows us to move on from experiments designed to demonstrate the existence of novel ocular photoreceptors, to experiments directed toward an under standing of this unexplored photosensory system of the eye. It is clear, however, that this understanding will emerge only if the cellular and molecular analysis is undertaken in parallel with a consideration of those features of the light environment used to regulate circadian and other NIF responses to light. 

 Submitted for publication July 20, 2001; accepted August 21, 2001.

 Supported by the Biotechnology and Biological Sciences Research Council, United Kingdom, and the European Union BioMed2.

 Commercial relationships policy: N.

 

Figure 1.

 

Representation of the arrangement of the photocell used for the detection of (A) irradiance-illuminance and (B) radiance-luminance. Irradiance-illuminance measures collect radiant energy from all directions over a 180° field of view, whereas radiance-luminance measures collect radiant energy viewed from a specific direction or region in space. The role of NIF photoreception is analogous to the task performed by an irradiance detector, while mapping brightness in a point of space to generate an image (IF photoreception) is analogous to a radiometric measure. ⁸⁰

Figure 2.

 

Histologic sections from the retina of a wild-type (A) and congenic rd/rd cl (B) mouse in which functional rods and cones are lacking. ⁶³ Tissue was fixed with Bouins (75% picric acid, 25% formalin, 5% acetic acid) for 24 hours, paraffin embedded, and 8 μm sections treated with antibodies recognizing rod opsin photoreceptors. GCL, ganglion cell layer; IPL, inner plexiform layer; INL, inner nuclear layer; ONL, outer nuclear layer; IS, inner segments; OS, outer segments; RPE, retinal pigment epithelium. Magnification, ×640. ⁶²

Figure 3.

 

Irradiance response curves for pupillary constriction to 506-nm light. These plots show the relationship between the minimum pupil area attained during 60 seconds of light exposure (mean ± SEM % constriction for 4 to 9 mice) and the irradiance of a monochromatic (λ_max 506 nm, half bandwidth 10 nm) stimulus for wild type and rd/rd cl mice. The data are fitted with a sigmoidal function of slope 0.58 (wild type) and 1.05 (rd/rd cl). ⁷¹ The results show that pupillary constriction can occur in the absence of rod and cone photoreceptors. However, there is a clear impact of rod and cone loss in rd/rd cl mice. ⁷¹

Figure 4.

 

Photic inhibition of locomotor activity in wild type and rd/rd cl mice. (A) Actograms showing wheel-running records of rd/rd cl and wild-type mice over 2 days; each line represents 24 hours. The number of wheel revolutions per 10 minutes are plotted in 15 quantiles on the y-axis, with each quantile representing 26 revolutions. The bar at the bottom shows the entraining light:dark cycle where white represents lights on and black lights off, the shaded portion indicates the time (ZT14) at which a 1 hour light pulse (155 lux) was presented on day 2; (B) Mean ± SEM (n = 8 wild type and 5 rd/rd cl) for the suppression of locomotor activity by 1 hour light pulses of decreasing irradiance. Irradiance is depicted as the number of photographic stops by which a standard light source was attenuated using neutral density filters. There were no significant differences between the responses of wild-type (open circles, broken line) and rd/rd cl (closed circles, solid line) mice (2-way ANOVA, P > 0.1). ⁷⁸

Figure 5.

 

Opsin photopigments in the murine eye. ⁷¹ The action spectrum for the pupillary light reflex in rd/rd cl mice is poorly fitted by the absorbance spectra of the known murine photopigments (green cone opsin, R ² = 0.00; rod opsin, R ² = 0.38; and UV cone opsin, no fit), and indicates the presence of a previously unidentified photopigment in the murine eye (OP⁴⁷⁹). This vitamin A-based pigment has a λ_max around 479 nm (R ² = 0.88), and is spectrally distinct from the rod and cone photopigments with λ_max at 360 nm (UV cone opsin ⁸⁰ ), 498 nm (rod opsin ¹³⁶ ), and 508 nm (green cone opsin. ¹³⁷ ).

Table 1.

 

Vertebrate Opsin Identity (%) Encompassing Transmembrane Domains I–VII

	Rod	LWS	MWS2	MWS1	SWS	P	VA	PP	TMT	Pan	RGR	Per	Mel	Invert
Rod opsin	—
LWS cone opsin	42	—
MWS2 cone opsin	69	44	—
MWS1 cone opsin	52	43	55	—
SWS cone opsin	46	41	50	51	—
Pineal opsin	46	50	49	52	48	—
VA opsin	36	43	39	42	40	43	—
Parapineal opsin	40	40	40	41	40	48	43	—
Teleost MT opsin	33	35	35	36	35	39	37	39	—
Panopsin	30	28	30	32	27	30	31	28	41	—
RGR opsin	21	21	22	22	20	23	22	25	22	24	—
Peropsin	29	26	26	26	22	27	27	29	23	31	25	—
Melanopsin	27	25	27	26	28	28	26	26	30	28	26	30	—
Invertebrate	27	25	26	24	24	26	23	28	29	26	22	29	42	—

 Transmembrane domains were defined according to the model by Baldwin et al. ¹³⁸ Amino acid sequences were aligned using MacVector 7 (Accelrys Cambridge, UK) with default values. Sequence sources (GenBank accession numbers in parentheses): Rod, chicken (D00702); LWS, chicken red (M62903); MWS2, chicken green (M92038); MWS1, chicken blue (M92037); SWS, chicken violet (M92039); Pineal, chicken (U15762); VA, salmon (AF001499); Parapineal, catfish (AF028014); Teleost MT, Fugu (unpublished); Panopsin, human encephalopsin (AF140242); RGR, human (U14910); Peropsin, human (AF012270); Melanopsin; human (AF147788); Invertebrate, cuttlefish rhodopsin (AF000947). Table complied by Dr. James Bellingham, Faculty of Medicine, Imperial College.

Figure 6.

 

VA opsin retinal expression. In situ hybridization was performed on 10-μm sections from 4% paraformaldehyde-fixed, cryoprotected salmon eyes, using digoxigenin-labeled cRNA probes. Retinal section with VA opsin antisense probe showing expression in a subset of horizontal (H) and amacrine cells (A). Hybridization with sense cRNA probes gave no labeling (not shown), thus confirming the specificity of the procedure. A, amacrine cell; H, horizontal cells; INL, inner nuclear layer; IPL, inner plexiform layer; olm, outer limiting membrane; ONL, outer nuclear layer; OPL, outer plexiform layer; OS, photoreceptor outer segments; PE, pigmented epithelium. Scale bar, 40 μm. ⁹⁰

Table 2.

 

Summary of the Results and Conclusions of Experiments Designed to Investigate the Loss of Candidate Photoreceptors That May Mediate Masking Behavior in Mice

Genotype and Retinal Phenotype	Masking Ability	Interpretation
rd/rd; no rods, few cones.	Masking present 77	Rods, photoreceptors are not required. Masking may be mediated by small numbers of cones and/or novel photoreceptors.
rd/rd cl; no rods or cones.	Masking present 78	Neither rod nor cone photoreceptors are required for masking. Novel photoreceptors must contribute to masking.
cry1 −/− cry2 −/−; no CRY. Rods and cones present. No histologic examination of the retina in aged animals.	Masking present, 120 but reports that masking is lost in some mice older than six months (van der Horst and Mrosovsky, written communication).	Neither CRY1 nor CRY2 is required for masking. Rod, cone and novel photoreceptors may all contribute to masking. Loss of masking in mice older than 6 months may be due to induced ocular disease in cry1 −/− cry2 −/− (e.g., uveitis) that causes the secondary loss of the photoreceptors that mediate masking. Histologic examination of the retina of aged cry1 −/− cry2 −/− mice may help resolve these alternatives.
rd/rd, cry1 −/− cry2 −/−; no rods, few cones, no CRY. No detailed histologic examination of the retina.	Masking in some animals is abolished but is present in ∼30% of animals. Some animals that initially show masking may lose this ability with age (VanGelder, written communication).	Classic and CRY photopigments may contribute in a redundant manner to masking. But because some animals can mask, the interpretation of results is complicated. An ability to mask may be due to the uneven survival of cone photoreceptors in some rd/rd cry1 −/− cry2 −/− mice. Alternatively, the loss of masking may be due to an induced ocular disease in rd/rd cry1 −/− cry2 −/− mice that causes the secondary loss of novel opsin-based photoreceptors that mediate masking. Histologic examination of the retina of rd/rd, cry1 −/− cry2 −/− mice may help resolve these alternatives.

Figure 7.

 

A comparison of the pupillary action spectrum results (which best fit an unidentified vitamin A-based pigment with a λ_max around 479 nm), ⁷¹ with the absorbance spectrum of a flavoprotein photopigment. ⁸³ Figure compiled by Dr. Robert Lucas, Faculty of Medicine, Imperial College.

I thank all my colleagues who have collaborated on these studies during the past 10 years. I want, however, to extend my particular thanks to Robert Lucas and Ignacio Provencio, without whom much of the work outlined in this review could not have been undertaken. Furthermore, I offer my thanks to Robert Lucas, James Bellingham, and Mark Hankins for their help in the preparation of the manuscript and Nicholas Mrosovsky, Bert van der Horst, and Russell Van Gelder, for communicating to me their unpublished observations in transgenic mice without cryptochrome.