Evolution and development of complex eyes: a celebration of diversity

 ‘The eye is like a mirror and the visible object is like the thing reflected in the mirror'

      Avicenna, Danish Nama-I ‘ala'I (11th century)

The eye has fascinated scientists and philosophers since the emergence of the discipline of biology. In each new technological era that has intersected biology (e.g. the advent of electron microscopy, advances in neurophysiology, developmental genetics and molecular phylogenetics), the visual system has been one of the first and most in-depth subjects studied. In each case, discoveries made about the visual system have been fundamental in furthering our understanding of the living world. There is no example clearer than that of the discovery of the transcription factor Pax6 (Cvekl and Callaerts, 2017).

Before the molecular revolution, biologists hypothesized that the eye evolved independently at least 40 times (Salvini-Plawen and Mayr, 1977). However, in 1994, it was discovered that the gene eyeless in Drosophila was a homolog of the gene small eye (Pax6) in mice and aniridia (PAX6) in humans (Quiring et al., 1994). This finding was expanded to a group of genes, the retinal determination network (RDN), that are required for eye development in Drosophila [eyeless (ey; Pax6 in mouse), twin of eyeless (toy; Pax6 in mouse), sine oculis (so; Six1/2 in mouse), Optix (Six3/6 in mouse), dachshund (dac) and eyes absent (eya)] and are directly associated with eye formation in other species (Gehring, 2014). This discovery fundamentally shifted our understanding of the evolutionary relationship between vertebrates and other distantly related organisms. It suggested the existence of ancient genetic networks that have a common relationship to patterning and organ formation across animals (Carroll, 2008; Shubin et al., 2009). This premise has ruled the study of evolutionary developmental biology for over 25 years.

Again, we find ourselves in a new technological era in biology. Barriers to study organisms beyond traditional genetic models have fallen, which enables us to better understand the diversity of life. To consider the evolution of form and function, it is crucial that we go beyond the study of homology. Organogenesis is undeniably a significant contributor to the diversity that we see across taxa, and photoreception is one of the most ubiquitous sensory modalities found on Earth. The eye provides an excellent comparative context to address issues regarding how morphogenesis and developmental genetics contribute to the evolution of both novelty and complexity.

In this Review, we discuss six areas in which a greater knowledge of visual organ development is advancing our understanding of evolutionary change (Fig. 1). Although the scope of this Review is limited by space constraints, fortunately, the field enjoys a wealth of excellent reviews (Arendt et al., 2009, 2016; Erclik et al., 2009; Jonasova and Kozmik, 2008; Kozmik, 2005; Lamb et al., 2007; Land, 2018; Land and Nilsson, 2012; Nilsson, 2013, 2009, 2004; Oakley and Speiser, 2015; Sanes and Zipursky, 2010; Suzuki and Grillner, 2018).

Single-chambered visual organs, like the vertebrate eye, have evolved multiple times. These complex organs are composed of multiple cell and tissue types that, through the process of morphogenesis, become arranged in an exacting morphology to enable specific visual functions. The components include a single lens (see Glossary, Box 1) in the anterior and a retina (see Glossary, Box 1) in the posterior. In addition, a pigment layer behind or within the retina is necessary, and a cornea (see Glossary, Box 1) is often found anterior to the lens. Animals have evolved varied morphogenetic processes to arrange these tissues during development. Here, we discuss eye morphogenesis in Vertebrates, Arachnids and Cephalopods (Fig. 2). This comparative understanding of morphogenesis can tell us about convergence in the evolution of tissue fates, organization and movement in development.

In all vertebrates, eye development begins with the bilateral evagination of optic primordia from the diencephalic region of the forebrain (Chow and Lang, 2001). The distal optic primordia elongate and adopt vesicle-like structures, subsequently invaginating to form a bilayered optic cup, whereas the proximal region constricts and forms the optic stalk (Fig. 2A). The distal aspect of the optic cup generates the neural retina (Fig. 3A), whereas the proximal aspect generates the retinal pigment epithelium (RPE) (Fig. 2B). Cranial neural crest and head mesenchyme form the periocular mesenchyme (POM), a cell type that directly contributes to multiple ocular structures but also interacts with the developing optic cup and provides extrinsic cues to modulate morphogenesis (Fuhrmann, 2010; Williams and Bohnsack, 2015; Gage et al., 2005). The lens placode, from which the lens arises, is generated through an inductive event between the optic cup and surface ectoderm (Fig. 2B-D; Cvekl and Ashery-Padan, 2014).

Much has been learned about optic cup morphogenesis in many vertebrate species, with perhaps the most detailed information coming from studies in zebrafish (Kwan et al., 2012; Cavodeassi and Wilson, 2019). Although these studies have identified distinct morphogenetic movements contributing to lens, retina and RPE formation, some of which are conserved between avian, mammalian and teleost species (Bernstein et al., 2018), our understanding of morphogenesis in mammals is limited given their in utero development. Interestingly, aspects of optic cup morphogenesis seem to be intrinsically programmed into neural tissues as optic cups derived from induced pluripotent stem cells, which lack lenses and POM, nonetheless undergo invagination to generate a cup-shaped structure (Eiraku et al., 2011).

Spiders belong to the subphylum Chelicerata within arthropods. Spider visual systems vary in complexity. Some spiders are almost blind and primarily web-dwelling, using vibrational cues as their principal sensory modality (Foelix, 2011; Land, 1985). Others, like jumping spiders, are wandering hunters with high-acuity vision (Jackson and Pollard, 1996). Most spiders have four pairs of camera-type image-forming eyes, each with a cuticular cornea, lens, cellular vitreous and photoreceptor and pigmented cells (see Glossary, Box 1). Across species, a huge diversity of placement and specialization of each pair of eyes exists.

The two front eyes, the anterior median or principal eyes, are the largest and often most complex (Land, 1985). The remaining eyes, called secondary eyes, often have specific functions. There are notable differences between the principal and secondary eyes beyond their complexity. First, the principal eyes have muscles that can move the retina laterally to enable directional vision (Strausfeld and Barth, 1993). Principal eyes never have a tapetum, a reflective layer enhancing sensitivity, which is common in secondary eyes (Homann, 1971). The orientation of the cells also differs between the two eye types. Principal eyes have outer segments (see Glossary, Box 1) pointing toward the direction of the incoming light, whereas secondary eyes have outer segments that are oriented posterior to their cell bodies. This is likely a result of differences in the morphogenesis of the two eye types.

Development of the principal and secondary eyes has been described in multiple species (Homann, 1971; Locy, 1886; Yoshikura, 1955). The principal eye is formed from an invagination of the prosoma, an anterior ectodermal tissue. Invagination starts from the bottom of the future eye, moving upward. This makes an ‘s’ shape of three layers of tissue; the most anterior generates vitreous cells, the middle generates photoreceptors and the posterior generates pigmented cells (Homann, 1971; Paulus, 1979) (Fig. 2E-G). Secondary eyes develop from a simple invagination of lateral ectoderm (Homann, 1971). These invagination events lead to an everted principal eye and an inverted secondary eye (Fig. 3B,C). In all spider eyes, the main refractive surface is the cuticular cornea and nothing is known about its development or the specific cuticular proteins it contains.

Coleoid cephalopods, including squid, octopus and cuttlefish, have a remarkable visual system with single-chambered eyes and large optic lobes for visual processing. The cephalopod retina has two identifiable cell types: photoreceptor cells and support cells (Fig. 3D). There are no interneurons (see Glossary, Box 1) found in the retina, as the photoreceptors synapse directly on the optic lobe.

The most in-depth description of cephalopod eye development comes from the squid, Doryteuthis pealeii (Koenig et al., 2016). The eye rudiments are among the first evidence of organ formation in the embryo. Two oval-shaped thickenings (placodes) form on each side of the embryo and, shortly thereafter, a lip of cells surrounding each placode emerges (Fig. 2H). Placodal cells contribute to the retina, whereas the lip cells contribute to the anterior segment (see Glossary, Box 1) and lens (Koenig et al., 2016). The lip begins to internalize placodal tissue and, once internalization is complete, the lip cells fuse and optic vesicles are generated (Fig. 2I). Shortly after vesicle fusion, lentigenic cells in the anterior of the vesicle begin to make the lens (Fig. 2J,K). The lens is made by long cellular processes wrapping around each other, forming fibrous layers (Meinertzhagen, 1990). These processes are made by three distinct populations of lentigenic cells arranged around the developing lens (Williams, 1909; Arnold, 1966, 1967; West et al., 1995). The lens itself is segmented, with distinct populations of lentigenic cells contributing to each segment (Arnold, 1967). The fibers from the lentigenic cells are eventually sheared from the cell bodies, making the lens acellular (Arnold, 1967).

What we see when comparing morphogenesis across species is that there is no universal way to execute this tissue organization. For example, cephalopod and spider lenses arise from neighboring tissue, are contiguous with the retina and are potentially neuroectodermally derived. The vertebrate lens is derived from the sensory pre-placodal ectoderm, is physically discontiguous from the retina and results from an inductive event. Beyond this tissue level of understanding, it will be interesting to identify cellular mechanisms of morphogenesis and their evolutionary conservation across the Bilateria.

The extensive knowledge of vertebrate eye development can be leveraged to better understand the evolution of complex organs. Exquisite anatomical data describing the visual organs in sister taxa, coupled to connectomic and molecular studies, have substantially enhanced our ability to assess homology across species. Here, we compare retinal structure and development between vertebrates, jawless fishes (lamprey and hagfish) and two invertebrate subphyla within Chordata, Cephalochordates and Urochordates (Suzuki and Grillner, 2018; Fain, 2019).

Retinal neurogenesis occurs in a stereotyped manner whereby neuroblasts are specified and differentiate into the seven cell types of the mature retina (Cepko, 2014). Retinal ganglion cells (RGCs; see Glossary, Box 1) are generated first, followed by cones, horizontal cells, amacrine cells, rods, bipolar cells, photoreceptors and Müller glia. These cells are arranged into the three laminae of the retina: the ganglion cell layer (GCL); the inner nuclear layer (INL); and the outer nuclear layer (ONL). The inner plexiform layer (IPL) separates RGCs and the INL, and the outer plexiform layer (OPL) separates the INL and ONL (Fig. 3A). Axons derived from RGCs make up the optic nerve, the sole output of the retina to visual centers of the brain. Rods and cones are the photoreceptive cells that receive and transmit visual information to RGCs via bipolar cells. Amacrine and horizontal cells serve as interneurons that modulate the activity of intraretinal circuits. The RPE is a melanin-containing cell monolayer that separates the retina from the choroidal blood supply. The RPE prevents light scatter and serves a number of crucial developmental and physiological roles in retinal homeostasis (Strauss, 2005). The transcription factor MITF is a key regulator of RPE development (Ma et al., 2019).

The molecular underpinnings of retinogenesis have been characterized in many vertebrate species, with key regulatory networks specifying retinal neuron classes largely conserved (Seritrakul and Gross, 2019; Bassett and Wallace, 2012; Cepko, 2014). Several of these are mentioned in descriptions of retinogenesis below, including Pax6, Pax2, Rx and Otx.

Lamprey possess cephalic eyes that develop biphasically: larvae possess immature eyes that develop into the adult image-forming camera-type eyes during metamorphic stages. Developmentally, the optic vesicles evaginate from the diencephalon (Murakami et al., 2001) and a rudimentary lens is formed in larvae (Kleerekoper, 1972; Meléndez-Ferro et al., 2002). Adult lamprey possess an ellipsoid multifocal lens, indicating that they can generate well-focused images, suggesting that a multifocal lens is ancestral in vertebrates (Bantseev et al., 2005; Gustafsson et al., 2008).

The larval (primary) retina is composed of only photoreceptors, RGCs and bipolar cells (Fig. 4A) (Dickson and Collard, 1979; Rubinson and Cain, 1989; Villar-Cerviño et al., 2006). This simple retinal circuitry, combined with the subdermal location of the eye, suggest that it is not image forming. Embryonic phases last ∼1 month and larvae then burrow into the sand for several years until they metamorphose into adults. In mid-stage larvae, the secondary retina proliferates and generates numerous neuroblasts that, with the exception of a few RGCs, remain undifferentiated (Fig. 4B) (Rubinson and Cain, 1989; Meléndez-Ferro et al., 2002; Villar-Cerviño et al., 2006; Villar-Cheda et al., 2008). This undifferentiated tissue encompasses ∼99% of the retinal area and gives rise to the adult retina (Villar-Cheda et al., 2008). The maintenance of this long-lived progenitor population in the secondary retina is not well understood.

Differentiation commences with RGCs in the secondary retina during late larval stages. During metamorphosis, amacrine and horizontal cells form first, photoreceptors next and bipolar cells last (Rubinson and Cain, 1989). The adult lamprey retina resembles that of other vertebrates, with the exception of the morphology and organization of RGCs (Fig. 4C; Fain, 2019). Lamprey possess inner RGCs and outer RGCs, the cell bodies of which are located in the IPL and the INL, respectively (Fritzsch and Collin, 1990; Jones et al., 2009). To generate the optic nerve, axons from these RGCs come together between the INL and IPL rather than along the vitreal surface of the eye, as with other vertebrates. There are two distinct retinotectal projections (optic nerves) that emerge at late larval stages; RGCs of the primary retina map to thalamic and pretectal areas, whereas many of those of the secondary retina map to the tectum (De Miguel et al., 1989; Jones et al., 2009).

Hagfish are the sister taxon to lamprey, but lack lens-containing eyes. It is likely that the eyes of extant hagfish are degenerate and have adapted to life in deep sea habitats. Evidence from a fossilized hagfish (Myxinikela) supports this model, with eyes appearing to contain a lens and a melanin-containing RPE-like structure (Bardack, 1991; Gabbott et al., 2016). This highlights one confounding aspect of using comparative embryology to understand the evolution of complexity: the incidence of loss in the phylogeny (Box 2).

Amphioxi possess four distinct light-responsive organs: the frontal eye, lamellar body, Joseph cells and dorsal ocelli (Fig. 5A; Lacalli, 1996; Lacalli et al., 1994). Ultrastructural and molecular studies suggest that the frontal eye is homologous to the vertebrate eye (Pergner and Kozmik, 2017). Structurally, the frontal eye is organized into four rows of neuronal cells and a pigmented cup (Fig. 5B,C). Pigmented cells express mitf, otx and a pax2/5/8 ortholog, and contain melanin, suggesting homology with the vertebrate RPE (Kozmik et al., 1999; Vopalensky et al., 2012). Dorsal and posterior to the pigment cup lie six Row1 ciliary photoreceptor cells that express otx and pax4/6 (Lacalli, 1996; Vopalensky et al., 2012).

Row2 cells are posterior to and contact Row1 cells. These cells extend projections to the tegmental neuropil of the cerebral vesicle, a region thought to bear homology to the vertebrate diencephalon and mesencephalon (Lacalli, 1996; Vopalensky et al., 2012; Albuixech-Crespo et al., 2017). Row2 cells might be homologous to RGCs; however, this model requires additional molecular and functional evidence. Row3 and Row4 cells are the most posterior in the frontal eye; Row3 cells are characterized by reciprocal contacts between one another, whereas Row4 cells extend axons to the tegmental neuropil (Lacalli, 1996). Molecularly, less is known about Row3 and Row4 cells but they express pax4/6 and Rx (Vopalensky et al., 2012), and potentially use glutamate as a neurotransmitter (Pergner and Kozmik, 2017). When compared to the vertebrate retina, their position supports their identity as interneurons, but this requires experimental validation.

Within tunicates, most studies have been performed on ascidians (sea squirts). Although adults do not have a discrete eye-like structure, the larval ocellus mediates phototactic and light-dimming responses (Tsuda et al., 2003; Horie et al., 2008; Salas et al., 2018; Bostwick et al., 2020). The cellular composition and organization of the ocellus in Ciona intestinalis has been resolved (Ryan et al., 2016). The ocellus houses two photoreceptive systems: PR-I and PR-II. PR-I consists of 23 photoreceptors that project outer segments into a single cup-shaped pigment cell. PR-II consists of seven photoreceptors, is located anterior to PR-I and is not associated with the pigment cell; rather, PR-II photoreceptors send outer segments into the brain ventricle. There are three lens cells located ventral to PR-I and PR-II that direct light to photoreceptors of both systems. Photoreceptors of PR-I and PR-II synapse onto relay neurons, which connect directly to the motor ganglion. Evidence supports a role for PR-I in mediating phototactic responses and for PR-II in the light-dimming response, suggesting that PR-I serves the function of cephalic eyes (Salas et al., 2018; Jiang et al., 2005; Kourakis et al., 2019).

The cell lineage of the ocellus has been determined, as have molecular mechanisms leading to its specification (Esposito et al., 2015). The three lens cells and photoreceptors arise from the medial neural plate (Cole and Meinertzhagen, 2004; Taniguchi and Nishida, 2004; Oonuma et al., 2016). Pax-6 and Mitf are expressed in cells that give rise to the ocellus (Glardon et al., 1997; Mazet et al., 2003) and Rx is required for ocellus formation (D'Aniello et al., 2006). The pigment cell derives from a Mitf-expressing cephalic melanocyte lineage located at the lateral edge of the neural plate, thought to potentially represent a rudimentary neural crest cell (Abitua et al., 2012; Yajima et al., 2003). In contrast, the vertebrate RPE is not neural crest-derived, indicating that the embryonic origins of these structures are different.

Cell type composition, anatomy and basic development of the visual systems within the chordate lineage have been fairly well resolved. However, even within vertebrates, there are many questions remaining. For example, we know almost nothing about the formation of cone and RGC-dense specializations that enable high-acuity vision. In primates, this is called the fovea, and similar high-acuity regions are found in species of birds and lizards. Expanding on recent progress (da Silva and Cepko, 2017), studying these high-acuity areas across species is interesting evolutionarily but also important for regenerative approaches in vision restoration. The cell biology of neurogenesis across species and its role in generating cell type diversity is another frontier in evolutionary developmental biology. Recent descriptions of neurogenesis in the squid retina have surprisingly shown a strikingly vertebrate-like neuroepithelium, which might be important to the generation of neural complexity (Box 3) (Koenig et al., 2016).

The euarthropod visual system is thought to have had a pair of lateral-faceted (compound) eyes and two pairs of median ocelli-type eyes, which have diversified significantly across arthropod lineages. With the expansion of molecular and genetic analyses of visual organs across arthropod species, we can identify molecular signatures that have been maintained and those that have changed to better understand how these changes impact morphology and function. Here, we look at three examples: Drosophila, Tribolium and arachnids.

Drosophila have seven visual organs, which include two compound eyes, three ocelli and a pair of Bolwig organs or larval eyes (Fig. 6A; Friedrich, 2006). The primary adult visual organs are the compound eyes, each composed of ∼750 ommatidia (see Glossary, Box 1). Their development has been extensively reviewed (Charlton-Perkins and Cook, 2010; Roignant and Treisman, 2009; Kumar, 2012; Quan et al., 2012; Treisman, 2013; Wolff and Ready, 1993; Kumar, 2010, 2011). Each ommatidium possesses eight photoreceptors and twelve supporting cells, including cone cells, pigment cells and mechanosensory bristles. The compound eyes and ocelli are lens-containing and are derived from the eye-antennal imaginal disc during larval development. RDN genes ey (Pax6), toy (Pax6), dac, eya, so (Six1/2) and eyg (Pax6) are required for eye development. Many of these genes can also induce eye formation if ectopically expressed in other parts of the embryo.

Drosophila are holometabolous insects (undergoing complete metamorphosis). The Bolwig organs are homologous to elaborate stemmata found in other holometabolous insects, such as fireflies and scorpionflies (Gilbert, 1994; Buschbeck, 2014). They are also considered homologous to the earliest-born photoreceptor cells in the adult eye of hemimetabolous species (lacking a pupal stage), such as grasshoppers (Liu and Friedrich, 2004). A small population of cells from the Drosophila embryo in the dorsal head neuroectoderm generates the larval Bolwig organs, adult compound eye and ocelli (Friedrich, 2006). After embryonic development, these cells invaginate to generate the eye-antennal disc. The three ocelli are derived from the dorsal anterior region of the disc. One ocellus is derived from each disc, and the third ocellus is a fusion of anlagen from each disc. After an early growth phase, the morphogenetic furrow initiates differentiation in the compound eyes, sweeping across the disc in a wave, posterior to anterior. The R8 photoreceptor cell is specified first, followed by R2 and R5, then R3 and R4, and ending with R1, R6 and R7. The Drosophila acellular lens, or pseudocone, is made in each ommatidium by four cone cells that differentiate after the photoreceptor cells. During pupal development, primary, secondary and tertiary pigment cells are added in sequence, ending with bristle cell differentiation. The primary refractive material in the Drosophila eye is the cuticular cornea, rather than the pseudocone. The cornea is made by cone and pigment cells (Cagan and Ready, 1989; Kimori et al., 1992; Janssens and Gehring, 1999; Stahl et al., 2017; Charlton-Perkins and Cook, 2010).

Tribolium, or the red flour beetle, is a member of the largest insect order, the Coleoptera, and diverged from Drosophila 350 million years ago (Klingler, 2004). Interest in Tribolium has grown because, unlike Drosophila, this genus has maintained more of its gene content relative to the urbilaterian ancestor, undergoes short germ-band development and has a larval head capsule. Tribolium have both larval and adult eyes. The larval eye is a simple eyespot composed of two photoreceptor cell populations that merge during embryonic development (Ho, 1961; Marshall, 1927; Liu et al., 2006; Friedrich et al., 1996). In total, the larval eye has ∼25 photoreceptor cells and the adult eye is composed of ∼80 ommatidia (Friedrich et al., 1996). Similar to Drosophila, Tribolium photoreceptor cells also differentiate in a wave, posterior to anterior, starting with R8 and ending with R7, followed by cone, pigment and bristle cells (Friedrich et al., 1996). Additionally, Tribolium also have a cuticular corneal lens. However, a survey of the Tribolium genome showed an absence of Drosocrystallin, the primary cuticle protein contributing to the Drosophila lens (Tribolium Genome Sequencing Consortium, 2008).

The first genetic work focused on eye development in Tribolium examined the RDN. Knockdown of Eyeless, Toy, Dac, Eya and So led to a reduction in size or loss of the adult eye (Schinko et al., 2008; Yang et al., 2009a,b; Luan et al., 2014; Posnien et al., 2011). In addition, Otd and Pph13 (Arx) might play similar roles in photoreceptor differentiation across insects. In Drosophila and Tribolium, orthologs of Otd and Pph13 are required for proper rhabdomere development (Mishra et al., 2010; Li et al., 1996). In both insects, Otd regulates short-wavelength opsin (see Glossary, Box 1) expression and Pph13 regulates long-wavelength opsin expression (Mahato et al., 2014; Li et al., 1996).

Although developmental homologies suggest that the mechanisms for generating photoreceptor diversity in insects might be conserved, the genetics of these eyes clearly differs (Friedrich et al., 2016). For example, in Drosophila, eyg, a Pax6 homolog, works downstream of Notch signaling to control cell proliferation and growth of the eye disc (Chao et al., 2004). In Tribolium, loss of Eyg makes the adult eye larger at the expense of the antennal region, suggesting a patterning or competence change (ZarinKamar et al., 2011).

The primary and secondary eyes found in arachnids have predicted homologies with Drosophila visual organs: the primary eyes are homologous to ocelli and the secondary eyes evolved from the ommatidium of lateral compound eyes (Fig. 6B). To address this hypothesis, two species of spider have been the focus of molecular investigation: the house spider, Parateatoda tepidariorum; and the wandering spider, Cupiennius salei. A whole-genome duplication in spiders has resulted in the generation of paralogs of many transcription factors likely involved in eye development, including atonal, dac, otd, pax6 and six3 (Fig. 6C) (Schwager et al., 2017). Gene expression in P. tepidariorum suggests that the primary and secondary eyes emerge from different areas. The six secondary eyes develop from two bilateral eye fields, with gene expression suggesting that each field splits into three individual secondary eyes over time (Schomburg et al., 2015). This supports the hypothesis that secondary eyes might have evolved from dispersed ommatidia of an ancestral compound eye and that the principal eyes evolved from the insect ocellus (Schomburg et al., 2015; Samadi et al., 2015; Morehouse et al., 2017). RDN gene expression in P. tepadariorum and C. salei shows varying combinations in primary and secondary eyes. Expression is also inconsistent across species (Schomburg et al., 2015; Samadi et al., 2015; Turetzek et al., 2016; Schacht et al., 2020) (Fig. 6C). These patterns might be evidence of neofunctionalization or subfunctionalization of these duplicated genes in the developing visual system. Current functional investigations suggest that six3 paralogs are redundant in P. tepidariorum but that soA/Pt-so1 is necessary for the development of both primary and secondary eyes (Schacht et al., 2020; Gainett et al., 2020 preprint)

Developmental constraint is the concept that the lineage of a tissue can impact or limit the phenotypes that can evolve (Smith et al., 1985). One of the shared developmental aspects of vertebrate and Drosophila eyes is their association with cephalization of the nervous system. Many complex visual organs are found in the head or associated in development with the central nervous system. However, there are multiple examples of non-cephalic or distributed image-forming eyes. To highlight the role of developmental context in the evolution of complexity, we describe three examples found in the Spiralia: Chiton, Scallop and Fan worms.

Polyplacophora (chitons) are a group of marine molluscs with eight overlapping shell plates made of aragonite. These plates are covered with thousands of pores containing sensory and secretory cells called aesthetes that lead to a canal system within the aragonite (Moseley, 1885; Sturrock and Baxter, 1993; Boyle, 1974; Baxter et al., 1987; Fernandez et al., 2007) (Fig. 7A,B). The aesthetes of chitons are thought to be multimodal sensors, but some species have evolved a subset into small lens-containing camera-type shell eyes (Moseley, 1885; Boyle, 1969a,b, 1976) (Fig. 7C). The lens in these eyes is composed of the aragonite shell material (Speiser et al., 2011; Li et al., 2015). This eye has the capacity to form an image and the animals show negative light responses (Speiser et al., 2011; Li et al., 2015; Kingston et al., 2018). The girdle tissue surrounding the shell plates is also photosensitive and lens-type tissue has been found in the sensory papilla of some species (Checa et al., 2017).

Although lineage tracing of adult tissues has not been performed in lens-containing chitons, evidence from other species suggests that both adult sensory cells and shell-making cells are ectodermally derived (Henry et al., 2004). In adult chiton, eyes on the shell plates are eroded over time, often losing lenses and becoming non-functional (Moseley, 1885). The mantle continually secretes shell at the periphery, adding new aesthetes (Haas, 1981; McDougall and Degnan, 2018; Checa et al., 2017).

No developmental work has been endeavored on the lens-containing chiton visual system. Recent studies in the eyeless polyplacophoran, Acanthochitona crinita, identified a dispersed population of Pax2/5/8-expressing cells in the developing shell field. They are labeled by the FMRFamide neuropeptide antibody and hypothesized to be the sensory cells contributing to the adult aesthetes (Wollesen et al., 2015a). The character of the shell that makes the lenses in chitons is more uniform in its alignment of crystalline (see Glossary, Box 1) components than the rest of the shell (Li et al., 2015); however, how these lenses are made remains unknown. Gbx is expressed in the developing shell field and possibly expressed in shell-secreting cells (Wollesen et al., 2015b). Surveys in many species have also shown SoxB1 and Engrailed expressed in the neuroectoderm and Hox genes expressed co-linearly in the developing shell field (Huan et al., 2020; Fritsch et al., 2015; Jacobs et al., 2000; Wollesen et al., 2018). These genes might also play a role in the development of adult sensory structures.

Within bivalves, both cephalic eyes and pallial eyes are observed, with examples ranging in structure and complexity (Morton, 2008; Nilsson, 1994). The scallop has between 30 and 100 pallial eyes at the tips of tentacles emerging from the middle mantle fold (Fig. 7D). Remarkable for their complexity, each is image forming, containing a lens, two retinas and a mirror (Dakin, 1909; Patten, 1887; Barber et al., 1967). The lens has little refractive power; the mirror, which is composed of square plates of guanine crystals, and cornea act as the primary focusing agents (Land, 1966a,b; Palmer et al., 2017; Speiser et al., 2016). The eye is single chambered, with light entering through the lens and hitting the proximal retina in the posterior of the eye. The light then reflects off the mirror (argentea) behind the proximal retina toward the distal retina at the anterior (Land, 1965; Palmer et al., 2017) (Fig. 7E). The proximal retinal photoreceptors express a rhabdomeric opsin and depolarize in response to light, whereas the distal retinal photoreceptors express a ciliary opsin and hyperpolarize in response to light (Wang et al., 2017; Gorman and McReynolds, 1969; McReynolds and Gorman, 1970; Land, 1966a; Hartline, 1938). It has been difficult to assess the function of either retina but the current hypothesis is that the distal retina forms an image, whereas the proximal retina might be responsible for acute peripheral vision (Palmer et al., 2017).

Scallop trochophore larvae have eye spots, but their relation to adult visual organs is unknown (Cragg, 2016). Early descriptions of scallop eye development identified that the retina, pigmented cells and cornea are ectodermally derived and the lens is mesodermally derived (Dakin, 1909; Küpfer, 1916). The first evidence of the eye anlage is a thickening on the middle mantle fold. This extends to form an undifferentiated optic papilla. Some cells within the papilla segregate to become the optic ganglion and as the papilla grows, cells organize into the optic vesicle (Butcher, 1930; Patten, 1887). Cells in the anterior become pigmented and will form the iris and cornea. The posterior cells proliferate, generating an undifferentiated cellular mass that organizes into layers. The proximal retina differentiates first, followed by the distal retina (Audino et al., 2015). Behind these retinae is the argentea and a pigmented cell layer. The argentea is cellular but eventually these cells degrade their nuclei (Patten, 1887). The lens is cellular and the last tissue to differentiate, with the nuclei arranging themselves at the periphery (Audino et al., 2015). RNA-seq shows that Pax2/5/8, Brn3, Lmx1b and Six4/5 are enriched in the adult pallial visual system (Wang et al., 2017). Their contribution during ontogeny remains unknown.

Annelids are a phylum of segmented worms found in the Spiralia that includes leeches, earthworms and fan worms, among others. Fan worms that live in tubes, including sabellid and serpulid annelids, have evolved a remarkable set of visual organs. They have long fan-like radiole tentacles used for feeding, with compound eyes along the tentacles to detect predators (Nilsson, 1994, 2013). These radiolar eyes show incredible diversity in placement and complexity (Bok et al., 2016; Bok and Nilsson, 2016). The most complex are found in Acromegalomma, which have two large compound eyes composed of hundreds of ocelli (Gil and Nishi, 2017). These animals have cerebral eyes as well, with rhabdomeric photoreceptors. The radiolar ocelli photoreceptors are ciliary (Eakin and Hermans, 1988; Verger-Bocquet, 1992; Purschke et al., 2006; Bok et al., 2017a). The use of this type of photoreceptor in image-forming eyes is unusual because in other annelid species, ciliary photoreceptors are associated with nonpigmented photoreception and the central nervous system (Bok et al., 2016). The radioles are constituents of the branchial crown, a specialized part of the annelid mouth (Bok et al., 2016, 2017b). Development of these visual organs has not been well documented, but the ocelli are derived from epidermal cells in the outer aboral corners of the radioles (Bok et al., 2017b). Transcriptomic data from the eyes of adult serpulid fan worms show expression of eya, six4, so, pax6, glass and dpp (Bok et al., 2017b).

As discussed elsewhere, one of the ways we understand evolutionary relationships between organs, tissues and cells in development is the identification of characteristic patterns of gene expression and function. The RDN is the primary group of genes used to understand these relationships in visual organs; however, there are numerous genes and networks associated with eye development in animals. Here, we discuss some of these regulatory factors in the context of Spiralian visual organ development.

With the explosion of Spiralian genomic and transcriptomic sequencing, we can assess expression of canonical visual system-associated transcription factors in the third major branch of the Bilateria (Fig. 1). Early surveys show canonical eye-related gene expression during visual organ development and homeostasis; however, no exclusive combination of genes appears to be required. We see this in the eyeless chiton species, Leptochiton asellus. All larval eye spots express Six1/2, Eya, Dach, Lhx2/9, Irx, Ovo and FoxQ2; however, Pax6 and Six3/6 were only found in some, but not all, larval photoreceptor populations or associated cells (Vöcking et al., 2015). Cephalopods, a class of molluscs that does not undergo metamorphosis, have shown a significant number of canonical transcription factors expressed in the developing eyes (Table 1). However, without a clear lineage of these tissues, it is difficult to assess homology across these visual organs.

Within Annelids, the adult cerebral eye of Platynereis dumerilii is the most in-depth description of lens-containing eye development (Rhode, 1992). In Platynereis, the larval eye is composed of two cells and emerges early in development. The adult eye forms in the larva, both organs existing simultaneously (Rhode, 1992; Arendt et al., 2002). Molecular work in P. dumerilii demonstrated variation in gene expression between larval and adult cerebral eyes, with Pax6, eya, six1/2, and otx found in the larval eye and Atonal and Six1/2 found in the adult (Arendt et al., 2002; Steinmetz et al., 2010). In addition, recent studies in Capitella telata, an annelid species with eyes that do not contain a lens, showed that knockdown of Ct-pax6 leads to loss of the larval eye and downregulation of Ct-eya (Klann and Seaver, 2019).

Some of the most in-depth molecular descriptions of the eye in Spiralia is during regeneration in planaria (Lapan and Reddien, 2012, 2011). Eye regeneration is independent of Pax6 (Pineda et al., 2002), but Eya and Six1/2 are required (Mannini et al., 2004; Pineda et al., 2000). Access to embryos in planaria is difficult but expression studies have confirmed that eye development is also likely Pax6 independent (Martín-Durán et al., 2012).

These surveys are highlighting that no single gene appears to be required for visual system development across all species; yet, many canonical eye development genes are shared. With still incomplete data and little information about the function of any of these genes, the evolutionary implication(s) of this shared expression remains unclear. Much will be learned with the expansion of functional genomic methods paired with further genetic manipulations.

The transition to bilateral symmetry correlates with an explosion of innovation in the animal lineage, including the cephalization and increasing complexity of the central nervous system. However, interesting innovations in the nervous system found outside of Bilateria include multiple instances of lens-containing visual organs in the Cnidaria.

Cnidarians, including corals, sea anemones and jellyfish, are the sister phylum to bilaterally symmetric animals in the Metazoa. Cnidarians have a large range of visual organ complexity, from eye spots to lens-containing eyes (Martin, 2004). The majority of these photoreceptive organs are associated with the free-swimming medusa stage within the Medusazoan clade. Three classes of Medusazoa have evolved lens-containing eyes independently: Hydrozoa, Cubozoa and Scyphozoa (Picciani et al., 2018). The medusa stage in these animals emerges from a sessile polyp and the visual organs develop during this metamorphosis, a process that differs in each group. Cubozoa and Scyphozoa have evolved specialized sensory structures called rhopalia at the bell margin of the medusa. Depending on the species, the sensory modality and complexity of the rhopalia varies, but can include a statocyst and photoreceptive organs. Hydrozoans do not have rhopalia; their visual organs are usually found at the base of the tentacle (Martin, 2002).

The Cubozoa or box jellyfish have the most sophisticated visual system of any cnidarian. In each of the four corners of the medusa, the associated rhopalium has six visual organs, each animal having 24 in total (Nilsson et al., 2005) (Fig. 8A). Two eyes in each corner have a lens, the others are simple pigmented eye cups. The two complex eyes have a cornea, iris, lens, retina and pigmented layer (Conant, 1898; Yamasu and Yoshida, 1976; Martin, 2004; Nilsson et al., 2005; O'Conner et al., 2009; Laska, 1982) (Fig. 8B). The upper eye is used for navigating mangrove forests of the cnidarian habitat and the lower eye is used for viewing the underwater environment (Nilsson et al., 2005; Hamner et al., 1995; Garm et al., 2007, 2011).

The medusa in Cubozoa is formed from a whole-body metamorphosis of the polyp stage. Early in development, the polyp tentacles move from radial symmetry to tetraradial symmetry (Werner, 1983, 1975; Werner et al., 1971; Stangl et al., 2002). These polyp tentacles form the developing rhopalia, which are highly proliferative, and the lens and retina emerge from transdifferentiated tentacle muscle (Laska-Mehnert, 1985; Piatigorsky et al., 1989; Gurska and Garm, 2014). Shortly thereafter, a cup primordium and melanin are apparent (Stangl et al., 2002; Valley, 2011). The lens is cellular and, early in development, cells that will make the lens divide in parallel with the epithelial plane. One cell changes its orientation, turning toward the developing retina, becoming wrapped by the other epithelial cells (Fig. 8C). This ‘roll morphogenesis’ continues until a circular lens is formed (Laska-Mehnert, 1985). During this period, pigment granules appear in the photoreceptor cells and the morphology of these cells elongates. These eyes are functional once the animal has budded from the polyp base (Laska-Mehnert, 1985). Molecular analysis of the Cubozoan eye is limited. The adult Tripedelia retina expresses a Pax homolog, PaxB, Mitf and two opsins, as well as three crystallins in the lens (Liegertová et al., 2015; Piatigorsky et al., 1989; Kozmik, 2005; Kozmik et al., 2008).

Scyphozoa lens-containing species have been identified, but the development of their eyes has not been described (Maas, 1903; Hertwig and Hertwig, 1878). However, we do have some knowledge of nervous system development in the related moon jellyfish, Aurelia sp.1. Aurelia do not have lenses, but have pigmented eye cups (Nakanishi et al., 2009). The pigmented cells are derived from endoderm and the sensory cells from ectoderm (Nakanishi et al., 2009; Yamasu and Yoshida, 1976; Hyman, 1940). Pit1, Brn3 and Otx are expressed during early neurogenesis in the rhopalia (Nakanishi et al., 2010). Later, Brn3, Six1/2 and Eya are found in the developing optic cup (Nakanishi et al., 2010). Pax genes and Six3/6 are not expressed during the stages surveyed in the developing visual organ, supporting its independent evolution (Nakanishi et al., 2015).

The last group, Hydrozoa, bud off multiple medusae from a sessile polyp, and many lens-containing eyes have been described in these species. In Cladonema radiatum, the ocellus is ectodermal, with the retina composed of a layer of photoreceptor cells and a layer of pigmented cells (Martin, 2002; Weber, 1981). The lens is cellular and composed of two units. Each unit is a distal cytoplasmic extension of a pigmented cell. No developmental surveys of gene expression have been performed, but Six1/2, Six3/6, Six4/5, Eya and PaxA are expressed in adult eyes (Suga et al., 2010; Stierwald et al., 2004; Graziussi et al., 2012).

Cnidarians independently evolved the eye multiple times (Picciani et al., 2018) and a greater understanding of these organs may reveal the developmental basis of elaborate neural tissues and optical structures in all animals. Furthermore, the regenerative capacity of cnidarians make them a particularly attractive model to understand stem-cell biology and the regeneration of sensory structures in an evolutionary context.

In this Review, we have discussed the visual organs of animals across broad phylogenetic distances, with a focus on ontogeny. We have highlighted six areas in which existing comparative studies and future work will help us understand fundamental questions in evolutionary biology. Although gene expression analyses in development have suggested homologies and potential evolutionary relationships between the eyes of diverse animals, it is now evident that no single gene regulates eye formation in all animals. In fact, the diversity found within the visual organs of animals has much more to offer us than an understanding of homology. Although we must keep gene expression in mind, we should expand our studies to include genetic function, cell biology and the morphogenetic mechanisms underlying visual organ formation across the Metazoa. The ability to study (and appreciate) this diversity is a powerful new and unbiased approach to understand the emergence of phenotype.

The areas discussed here are just a miniscule sampling of what the visual system has to offer. An additional example includes the developmental diversity of lens formation across animals, and how this can pattern the refractive index (Land, 2012). Another is the generation of reflective materials, such as a tapetum, to enhance photoreceptor sensitivity, or the retinal mirror found within scallops (Schwab et al., 2002; Land, 1966b). These materials have evolved multiple times but how they are generated is largely unknown. We know astonishingly little about the developmental basis of iris diversity or the cell biology of intracellular filters in photoreceptors, such as pigmented oil droplets in the avian retina (Banks et al., 2015; Toomey and Corbo, 2017). The list of newly accessible questions is endless.

In conclusion, much like we should celebrate the diversity among ourselves, so too should we celebrate the diversity throughout the Metazoa. We are in an era in which ‘model systems’ are no longer limited to those few that were once tractable to molecular techniques. We should leverage this diversity to uncover shared and unique developmental mechanisms underlying complex organ evolution and, in doing so, shed light on cell and developmental processes that are beyond our current imagination.

We thank Ana Gabriel for generating several figures; Katie Crutchfield and Andy Gill for editorial support; and Dave McClay, Susana da Silva and the anonymous reviewers for helpful comments on this Review. We apologize to any authors whose contributions to this broad and diverse field have been omitted from this Review due to space constraints.