How do horizontal cells ‘talk’ to cone photoreceptors? Different levels of complexity at the cone–horizontal cell synapse

Abstract

The first synapse of the retina plays a fundamental role in the visual system. Due to its importance, it is critical that it encodes information from the outside world with the greatest accuracy and precision possible. Cone photoreceptor axon terminals contain many individual synaptic sites, each represented by a presynaptic structure called a ‘ribbon’. These synapses are both highly sophisticated and conserved. Each ribbon relays the light signal to one ON cone bipolar cell and several OFF cone bipolar cells, while two dendritic processes from a GABAergic interneuron, the horizontal cell, modulate the cone output via parallel feedback mechanisms. The presence of these three partners within a single synapse has raised numerous questions, and its anatomical and functional complexity is still only partially understood. However, the understanding of this synapse has recently evolved, as a consequence of progress in understanding dendritic signal processing and its role in facilitating global versus local signalling. Indeed, for the downstream retinal network, dendritic processing in horizontal cells may be essential, as they must support important functional operations such as contrast enhancement, which requires spatial averaging of the photoreceptor array, while at the same time preserving accurate spatial information. Here, we review recent progress made towards a better understanding of the cone synapse, with an emphasis on horizontal cell function, and discuss why such complexity might be necessary for early visual processing.

Abbreviations

AC

amacrine cell

BC

bipolar cell

HC

horizontal cell

STED

stimulated emission depletion microscopy

VGCC

voltage gated calcium channel

Introduction

The retina is a specialized part of the central nervous system; it detects light, and processes and transmits visual information about the outside world to the brain. Photons hitting a photoreceptor are transduced into an electrical signal and fed into parallel bipolar cell (BC) pathways that extract different visual features and convey this information to the retinal ganglion cells, the output neurons of the retina (Euler et al. 2014). The signals along this ‘vertical’ pathway – from photoreceptors to BCs to retinal ganglion cells – are modulated by inhibitory neurons at two different levels, horizontal cells (HCs) and amacrine cells (ACs) in the outer and inner retina, respectively (Fig. 1 A).

Figure 1. Organization of the mouse retina and the cone photoreceptor synapse.

A, schematic representation of a vertical section of the mouse retina, consisting of five neuronal classes organized in different layers. AC, amacrine cell; BC, bipolar cell; C, cone photoreceptor; GC, ganglion cell; GCL, ganglion cell layer; HC, horizontal cell; INL, inner nuclear layer; IPL, inner plexiform layer; ONL, outer nuclear layer; OPL, outer plexiform layer; R, rod photoreceptor. B, volume‐rendered individual HC (red) and contacting cone axon terminals (purple); HC traced by Yue Zhang using a published EM dataset (Helmstaedter et al. 2013); 3D volume rendering by Christian Behrens. C, schematic organization of the cone synapse: Simplified cone axon terminal (purple) with a single exemplary ribbon (black vertical line) surrounded by numerous synaptic vesicles (white), with two invaginated lateral HC dendritic processes (red) and an ON cone BC dendrite (dark grey). OFF cone BC dendrites (light grey) are located at the base of the cone axon terminal. Note that an individual mouse cone axon terminal contains ∼10 ribbon synapses. Scale bar: 10 μm.

Horizontal cells modulate the output of photoreceptors and play many roles in early visual processing contributing to contrast enhancement, colour opponency, and the generation of centre–surround receptive fields in cone photoreceptors (cones) and BCs. In non‐mammalian vertebrates, HCs are functionally even more complex; e.g. the fish retina possesses mono‐, bi‐ and triphasic HC types (Kamermans et al. 1991). For many decades HCs have been mainly considered as an electrically coupled network, primarily responsible for providing global feedback (reviewed by Thoreson & Mangel, 2012). However, recent studies suggest that HC feedback can also act on a much smaller spatial scale – between a single HC dendritic tip and a cone axon terminal (Jackman et al. 2011; Vroman et al. 2014), indicating that HC dendrites may play a role in both local and global visual signal processing. Combining computational processes at different scales into a single type of neuron is a recurrent motif in the retina (e.g. in ‘dendritic processing’; reviewed by Schubert & Euler, 2010). The retina needs to use space efficiently and avoid metabolic overheads (Grimes et al. 2010); this requirement is supported by dendritic processing.

In this review, we discuss the retina's first synapse, with an emphasis on cone and HC connectivity. We focus on the different functions of HCs; with respect to details on distinct HC feedback mechanisms, we direct the interested reader to recent reviews (e.g. in Thoreson & Mangel, 2012; Vroman et al. 2013; Kramer & Davenport, 2015). Despite important differences between cone and rod photoreceptors (rods), e.g. with respect to the glutamate release mechanisms (Babai et al. 2010; Chen et al. 2013), the underlying principles of HC feedback at the rod–HC synapse appear to be similar to those at the cone–HC synapse (Thoreson et al. 2008; Szikra et al. 2014). Therefore, some principles discussed here may also hold for the rod–HC synapse.

Horizontal cells – the multi‐purpose interneuron of the outer retina

Like the majority of mammals, mice possess one type of rod that mediates low‐light (‘night’) vision, and two types of cone mediating vision in bright light (‘daylight’): a short (S‐, ‘blue’) and a medium (M‐, ‘green’) wavelength‐sensitive cone (Szél et al. 1992; Baden et al. 2013b). For transmitting information to HCs and BCs, photoreceptor axon terminals contain specialized structures called ribbons, which allow fast and sustained release of glutamate (reviewed by Baden et al. 2013a). In contrast to rod axon terminals, which feature only a single ribbon (Tsukamoto et al. 2001), cone terminals usually have multiple ribbons. The number of ribbons varies between species. While mouse cones contain approximately 10 ribbons, cones in the peripheral monkey retina feature more than 50 ribbons (reviewed in Sterling & Matthews, 2005).

Horizontal cells are GABAergic interneurons that receive information from cones and rods. Aside from a few rodent species, such as rats and mice, that possess a single type of HC (the axon‐bearing type), most mammals have two HC types. In mouse, HC dendrites receive input from all cones within their dendritic field (Feigenspan & Babai, 2015) (Fig. 1 B), and each cone is contacted by several HCs, whereas their axon terminals receive rod input (Peichl & González‐Soriano, 1994). The dendritic arbour and the axon terminal system each constitute independent, large and electrically coupled networks through Connexin57‐formed gap junctions (Hombach et al. 2004). Due to this strong coupling, the receptive field of each HC is larger than its dendritic field (Shelley et al. 2006). The extent of HC coupling is modulated by dopamine, whose levels are regulated both by the circadian rhythm and the light adaptation state of the retina (Tornqvist et al. 1988; Hampson et al. 1994; Xin & Bloomfield, 1999; He et al. 2000). For example, when daylight increases at dawn, dopamine is released by dopaminergic ACs in the inner retina, diffuses to the outer retina and decreases electrical coupling in the HC network (reviewed by Witkovsky, 2004). Despite the anatomical segregation of cone and rod contacts in the HC dendrites and axon terminal system, respectively, these structures are not completely isolated from each other: cones and rods are electrically coupled at the level of their axon terminal (Hornstein et al. 2005; Asteriti et al. 2014). Thus, cone signals reach HCs via rod axon terminals and rod signals may enter HCs via cone axon terminals. Additionally, it has been shown that cone signals can travel along the axon of an HC from its dendrites to its axon terminal system (Trümpler et al. 2008; Szikra et al. 2014). However, whether rod signals can also travel from the axon terminal system to the dendrites remains controversial (Trümpler et al. 2008; Szikra et al. 2014). Of course, BC dendrites also participate in the photoreceptor synapse: a single ON cone BC dendrite – accompanied by two HC dendritic tips – invaginates the synaptic cleft of the cone axon terminal, whereas OFF cone BC dendrites occupy its base (Haverkamp et al. 2000) (Fig. 1 C).

Synaptic interactions at the cone synapse

The highly sophisticated cone synapse is formed by three different neuron classes and is conserved across vertebrates. Horizontal cells play a pivotal role by modulating cone output via reciprocal feedback. In addition, they may also affect BC dendritic excitability directly via GABAergic feedforward signalling.

Cones release glutamate, which binds to postsynaptic receptors expressed on HC and BC dendrites. Horizontal cells express ionotropic AMPA‐ and kainate‐type glutamate receptors and form sign‐conserving synapses with the cones (Schultz et al. 2001; Schubert et al. 2006; Kreitzer et al. 2009; Ströh et al. 2013; Feigenspan & Babai, 2015). In turn, HCs feed back onto cones and modulate their glutamate release. In addition, it has been suggested that HCs modulate BCs via a feedforward pathway (Yang & Wu, 1991; Duebel et al. 2006; Puller et al. 2014). For example, as discussed below, HCs probably release GABA (Liu et al. 2013) and GABA receptors have been found on BC dendrites in at least some species (Puller et al. 2014; Hoon et al. 2015). Hence, depending on the local chloride equilibrium potential in BC dendrites, HCs may provide these cells with feedforward inhibition or excitation (Vardi & Sterling, 1994; Vardi et al. 2000; Duebel et al. 2006). Nonetheless, the signalling between HCs and BCs is poorly studied and still a source of controversy (Schubert et al. 2008; Purgert & Lukasiewicz, 2015).

Hypothetically, BCs may in turn modulate the cone synapse. Glutamate transporters are present in BC dendrites (Tse et al. 2014) and may affect the glutamate concentration below individual ribbons. Additionally, interplexiform ACs that are driven by BC input may shape HC responses by providing GABAergic output in the outer plexiform layer (Dedek et al. 2009).

Feedback mechanisms employed by horizontal cells

Horizontal cell‐to‐cone feedback may employ three different mechanisms: ephaptic, proton (or pH)‐mediated and GABA‐mediated feedback. Although these feedback mechanisms have been extensively studied, it is unclear whether and how they interact (reviewed by Kramer & Davenport, 2015). Both ephaptic and proton‐mediated feedback act on voltage‐gated Ca²⁺ channels (VGCCs) expressed at the active zone of the ribbon synapse in the cone axon terminal. Feedback via these mechanisms results in a shift of the VGCC activation range and therefore regulates cone glutamate release (Kamermans & Fahrenfort, 2004; Vroman et al. 2014). In addition, both feedback pathways appear to be modulated by a third one; this pathway is referred to as GABAergic feedback (Liu et al. 2013).

Ephaptic feedback is fast and mediated by cations flowing through hemichannels into the distal HC dendrites when HCs hyperpolarize to light. For example, the hemichannel‐forming protein Connexin55.5 was shown to be involved in ephaptic feedback in zebrafish (Klaassen et al. 2011). Another possible candidate protein is Pannexin1, which is present in the dendritic tips of mouse and zebrafish HCs (Prochnow et al. 2009; Kranz et al. 2013). The resulting drop in voltage between the cone axon terminal and the synaptic cleft is sensed by the VGCCs in the cone, promoting Ca²⁺ influx and thereby increasing glutamate release. In theory, ephaptic feedback should be an instantaneous process, and indeed, Vroman and colleagues (2014) detected no latency for ephaptic feedback measured in cones when using full‐field light stimuli. This, however, is in contrast to a study in which feedback signals in cones displayed a latency of ∼10 ms (Warren et al. 2016b). This discrepancy may have resulted from the different ways HC feedback was evoked in the two studies: Vroman et al. (2014) employed a full‐field light stimulus to activate feedback from the entire HC network to an individual recorded cone. Warren et al. (2016b) also recorded individual cones, but in contrast to Vroman et al. (2014), they manipulated the membrane potential of the presynaptic HC directly. When hyperpolarizing a single HC (and possibly a few electrically coupled HCs weakly) via a patch pipette, only the activity state of a single HC is strictly controlled, while the state of the neighbouring HCs is likely to be less constrained. The different approaches to manipulating HC activity are likely to have contributed different filter/delay components to the feedback measured in the cone. If and how the pattern of stimulation ‘seen’ by the HC network affects feedback kinetics remains to be investigated.

Compared to the ephaptic pathway, the kinetics of proton‐mediated feedback is slow (with a time constant of ∼35 ms for ephaptic vs. ∼200 ms for proton‐mediated light‐induced feedback responses in cones. Note that these time constants include the response time constants of both cones and HCs; Vroman et al. 2014). The general underlying mechanism is an acidification in the synaptic cleft caused by proton concentrations that exceed the extracellular pH buffer capacity. Protons then bind to negative residues present at the VGCC pore and reduce the channel's conductance (Chen et al. 1996). These protons derive from multiple sources, including protons co‐released with glutamate from the cone axon terminal (DeVries, 2001), the hydrolysis of ATP extruded by HCs through hemichannels (Vroman et al. 2014), Na⁺/H⁺ exchangers, proton–bicarbonate permeable channels (Warren et al. 2016a), plasmalemma membrane Ca²⁺/H⁺‐ATPases (Kreitzer et al. 2007) and proton pumps (Wang et al. 2014).

Also, the third feedback mechanism is rather unusual: here, GABA released from HCs upon depolarization binds to ionotropic GABA receptors on the HCs themselves (auto‐reception; Liu et al. 2013). Therefore, it is conceivable that this pathway may modulate the other two feedback mechanisms (Kamermans & Werblin, 1992; Kemmler et al. 2014). Nevertheless, the question of a common GABA‐mediated pathway in vertebrates is still controversial, as its contribution to HC feedback appears to depend on the experimental condition (Verweij et al. 2003; Tatsukawa et al. 2005) as well as on the species. For example, in mice, HC feedback is clearly modulated by GABA released from HCs, because after elimination of the vesicular GABA transporter, uptake of GABA in HC vesicles is abolished and feedback modulation absent (Hirano et al. 2016). The GABA‐mediated feedback is likely to be indirect, as mouse cones lack ionotropic GABA receptors (Kemmler et al. 2014). In fish, activating GABA_A receptors reduced the HC‐to‐cone feedback (at a time scale of several seconds; Endeman et al. 2012), supporting a modulatory role for GABA on the other feedback mechanisms. In contrast, application of GABA did not have any consistent effect in the primate retina (Verweij et al. 2003).

Note that at least ephaptic and proton‐mediated feedback is fast (Kamermans & Werblin, 1992; Vroman et al. 2014; Warren et al. 2016b), which argues against an involvement in light adaptation, a function that is often ascribed to HCs. Adaptation to the ambient light level mostly happens as a modulation of the phototransduction cascade (reviewed by Pugh et al. 1999; Fain, 2011) and of gap‐junctional coupling between photoreceptors and between HCs. The speed of feedback highlights that HCs should be well suited for tasks, such as contrast enhancement, that require fast adjustment of cone output.

Lateral signal spread in the horizontal cell network or local feedback?

Recent studies suggest that HCs can act at two different spatial scales, global and local (Jackman et al. 2011; Vroman et al. 2014). This raises the interesting possibility that global and local modulation of cone output act in concert, with local modulation minimizing a potential loss of local stimulus features that is introduced by spatial averaging in the coupled HC network.

‘Traditionally’, HCs are thought to be involved in global signal processing (Fig. 2 A), for example in the context of contrast enhancement (VanLeeuwen et al. 2009). The basic principle is lateral inhibition – averaging the input from many cones and subtracting this mean signal (which approximates the ambient light level) from the local cone response. The signal integration over distances beyond the size of an HC's dendritic arbour strongly relies on the low‐resistance gap junctional coupling of the HC network (Fig. 2 A). Expression of VGCCs and voltage‐gated sodium channels in HCs may promote lateral spread of signals within the network (Schubert et al. 2006; Mojumder et al. 2007). More generally, global signal processing can range from integrating the input from a few cones by one individual HC to large cone population by multiple, electrically coupled HCs.

Figure 2. Global and local signal processing at the cone‐to‐horizontal cell synapse.

A, schematic representation of global feedback in an individual HC or in the electrically coupled HC network. Cones (purple) release glutamate (black arrows) which is received by HCs (red) that in turn provide global feedback to cones (grey arrows) due to spread of signals in the HC network via gap‐junctions (black zigzag). B, schematic representation of local (cone synapse‐specific) HC feedback. HC dendritic processes may feed back onto individual cones independently from neighbouring dendritic processes (grey/black double arrows). C, hypothesized ribbon‐specific presynaptic microdomains at different synaptic clefts in a single cone axon terminal. Two ribbons contacting dendrites of two different types of ON cone BC (dark and light green) differ in glutamate release rate (black circles) and synaptic strength (arrows). BC, bipolar cell. D, overview of the connectivity between cones and two neighbouring HCs (orange and red). HC, horizontal cell. The black box corresponds to the enlarged schemata shown in E and F. E and F, two alternative hypotheses for the processing of cone signals in HC distal dendrites (postsynaptic microdomains). E, two electrically coupled processes from two distinct HCs (orange and red) act as a single input–output structure (double arrow) due to their gap‐junctional communication (black symbol). F, the two lateral HC processes may be independent of each other and act as individual ‘post‐synaptic microdomains’ (grey and black arrows). Note that only two ribbon synapses are shown for the mouse cone axon terminal in C, E and F instead of ∼10 ribbon synapses per cone.

Recently, a more complex picture of HC processing has started to emerge. In this picture, HCs may also perform local signal processing by providing individual cones with a highly ‘personalized’ feedback at the level of a single HC dendrite (Fig. 2 B). First evidence supporting such local feedback came from a study by Jackman et al. (2011). They proposed that HCs can provide positive (excitatory) feedback that acts at a local scale and depends on intracellular Ca²⁺ signalling, which in turn triggers the release of a retrograde messenger acting at the cone terminal. However, the nature of this messenger remains elusive. Moreover, it is unclear how such positive feedback could be prevented from functionally destabilizing the synapse by driving it out of its operational range. Here, one possibility may be that local positive feedback is counterbalanced by global (or maybe even local) negative feedback. More recently, it was proposed that local feedback – at the level of an individual HC‐cone synapse – may be implemented via the proton‐mediated pathway (for discussion, see Vroman et al. 2014; Wang et al. 2014).

If HCs do indeed feature distinct spatial ‘modes’, further interesting questions arise, for example, whether these modes co‐exist in the same HC or if they represent different general states of an individual HC. Jackman and colleagues (2011) proposed that global and local signals are functionally segregated by their ‘carrier’ (intracellular Ca²⁺ for local vs. changes in membrane potential for global). In principle, this separation cannot be complete, as Ca²⁺ and membrane potential are interlinked by VGCCs, which are present in HCs. In this respect, a working model for local signal processing in HCs may be the A17 AC (Grimes et al. 2010). This inhibitory neuron employs a distinct combination of anatomical features (thin dendrites, consistent minimal distances between synaptic buttons), active channel complement, and locally restricted Ca²⁺ signalling to provide local reciprocal feedback in the inner retina.

Another level of complexity at the cone‐to‐horizontal cell synapse?

Evidently, the cone synapse features a highly stereotypical and conserved anatomical organization (Fig. 1 C). At the same time, it is remarkably complex, not only anatomically (Haverkamp et al. 2000) but also functionally, e.g. with respect to the diversity of unusual signalling mechanisms it employs (see above). But have we really understood the main working principles of this synapse? Recent work suggests a further, ‘subsynaptic’ level of organization (Tang et al. 2016), which, if confirmed, has interesting consequences for the integration of cone signals and for HC feedback.

‘Presynaptic microdomains’ in the cone axon terminal

Synaptic multiplexing refers to the ability of a neuron to provide different postsynaptic partners with distinct input. Multiplexing typically requires structural specializations, such as the birds‐on‐a‐wire organization of A17 AC dendrites (Grimes et al. 2010) or the compartmentalized axon terminals of some fish BCs (Baden et al. 2014). Because cone axon terminals are so compact, it seems unlikely that cones support multiplexing. Nevertheless, it cannot be excluded that the individual ribbons of a cone can act, to some degree, independently from their neighbours as previously shown for ribbons in mouse inner hair cells (Frank et al. 2009). As shown in a recent lizard and salamander study, the local presynaptic Ca²⁺ signals can be restricted to individual glutamate release sites (ribbons), resulting in independent output units (‘presynaptic microdomains’) in a single cone axon terminal (Choi et al. 2008). Indeed, since the activation of a few VGCCs at the active zone is sufficient to initiate glutamate release (Bartoletti et al. 2011), different sizes of VGCC clusters directly at a ribbon (Morgans et al. 2005; Mercer et al. 2011) may result in the release of different numbers of vesicles and, thus, different glutamate concentrations beneath the ribbon (Fig. 2 C). This functional diversity could be complemented by the stereotypic connectivity at the cone synapse: here, each ribbon is opposed by a single dendrite of a different (ON‐) cone BC type (Behrens et al. 2016). Distinct glutamate concentrations at ribbons presynaptic to distinct BC types would contribute to shaping their responses in a type‐specific manner – probably in concert with different sets of postsynaptic glutamate receptors with different affinities and kinetics (DeVries, 2000; Puller et al. 2013). Glutamate‐sensing biosensors such as iGluSnFR (Marvin et al. 2013) in combination with synaptic targeting to the synaptic cleft (Wang et al. 2014), and high‐resolution microscopy may provide the experimental approach to test the hypothesis of presynaptic microdomains at individual ribbons and their potential effect on postsynaptic neurons.

‘Postsynaptic microdomains’ in horizontal cell dendritic processes

Why is each synaptic cleft in the cone axon terminal occupied by exactly two dendritic processes (Haverkamp et al. 2000) from two (likely) distinct HCs? So far, it is not known to what extent these two distal processes are electrically coupled (Fig. 2 D–F). If the electrical synapse was not at the distal dendritic tip – as previously proposed (Janssen‐Bienhold et al. 2009; Puller et al. 2009) – the two HC processes could function as independent input–output structures (Fig. 2 D and F). Such ‘postsynaptic microdomains’ would add yet another level of complexity to this synapse, particularly, if the two contacting HCs were in different activation states (or ‘processing modes’), or if they expressed different types of ionotropic glutamate receptor, as was suggested by Deng et al. (2006). Moreover, the regulation of glutamate release at each presynaptic microdomain would strongly depend on the dominant feedback mechanism in the postsynaptic microdomain (e.g. fast ephaptic vs. slow proton‐mediated; Vroman et al. 2014; Warren et al. 2016b). In fact, the presence of a local proton‐mediated feedback at the cone–HC synapse, as described by Wang and colleagues (2014), generally supports the idea of postsynaptic microdomains.

In addition, the alignment of pre‐ and postsynaptic microdomains (Tang et al. 2016) could play a role in shaping the glutamate release at each ribbon (independently of neighbouring ribbons). As discussed above, this would directly affect BC responses and their kinetics, as a function of the feedback mechanism involved.

Horizontal cell function and implications of local processing

If cones and HCs featured multiple pre‐ and postsynaptic microdomains, respectively, how can this be reconciled with the ‘classical’ view, in which HCs integrate signals on a global scale? In what follows, we speculate on the consequences of such microdomains on specific known and proposed HC functions.

Contrast enhancement and colour opponency

In the primate retina, HCs are involved in centre–surround antagonism (Verweij et al. 2003) in cones, in particular in the context of colour opponency (Packer et al. 2010). These functions clearly represent global processing modes of HCs. In contrast to the mouse retina, which possesses a single type of HC, primates have two HC types: the HI type, which samples signals from long (L‐, ‘red’) wavelength‐sensitive cones and M‐cones but largely avoids S‐cones (Dacey et al. 1996; Goodchild et al. 1996), and the HII type, which samples predominantly S‐cones and receives weak input from L‐ and M‐cones (Dacey et al. 1996). For large stimuli, HIIs are thought to sum up M‐ and L‐cone input and provide lateral inhibition to S‐cones, thus generating colour opponency as early as in the cone (Packer et al. 2010) (Fig. 3 A). Mouse HCs are more similar to HII cells, as they contact all cones within their dendritic field (Feigenspan & Babai, 2015) and thus likely receive mixed S‐ and M‐cone inputs. Thus, it is conceivable that mouse cones receive chromatically antagonistic input from neighbouring cones via lateral HC inhibition (Fig. 3 B). However, such chromatically opponent responses have not (yet) been detected in mouse cones (Baden et al. 2013b), but were proposed for other non‐primate mammals such as rabbits (Mills et al. 2014). In addition, the HII type transmits feedforward excitatory signals from M‐cones to blue‐ON cone BCs via GABA release, likely providing them with a yellow‐OFF input (Puller et al. 2014), also a feature not yet found in mice.

Figure 3. Speculative chromatic horizontal cell feedback in primate and mouse retinae.

A, schematic representation of colour opponency in an S‐cone generated by an HII cell in the primate retina when a green/red light stimulus is presented. M‐/L‐cones (green/red) provide glutamatergic input (thin black arrows indicate weak synaptic contacts) that is summed by the HII (dark grey), which in turn provides lateral feedback to the S‐cone (blue) (thick grey arrow indicates strong feedback). B, scheme showing the speculative colour opponency in S‐cones in the mouse retina generated by an HC when a green light stimulus is given. M‐cones (green) release glutamate (black arrows) that is received and integrated by the HC (light grey), which in turn provides feedback (grey arrows) to S‐cones (blue). C, schematic representation of global feedback generated by a primate HI cell. Only M‐/L‐cones (green/red) but no S‐cones (not shown) provide glutamatergic drive (black arrows) to the HI cell (dark grey) that feeds back to M‐/L‐cones (grey arrows). D, speculative global S‐cone‐specific feedback generated by a primate HII cell. S‐cones (blue) make strong contacts (thick black arrows) with an HII cell (dark grey) whereas M‐/L‐cones (green/red) make weaker contacts (thin black arrow). S‐cone signals are predominantly converted into feedback to S‐cones (blue arrows). E, speculative local feedback to S‐ (blue) and M‐ (green) cones in the mouse retina. Glutamatergic input from cones (black arrows) is not globally processed in mouse HC dendrites (light grey). Instead, feedback is generated locally in distal HC dendritic tips and provided in a cone‐specific way (green and blue arrows).

From a more general perspective, the largely differential cone connectivity of HI and HII cells suggests that primates possess two (mostly) independent HC networks with distinct functions: with HI cells shaping M‐ and L‐cone output, and HII cells predominantly shaping S‐cone output and supporting S vs. M/L opponency (Fig. 3 C and D). Note, however, some synaptic interaction between these networks is required, for example, with respect to colour constancy (discussed in Kamermans et al. 1998). Nonetheless, evidence for such a ‘division of labour’ exists also for non‐primate mammals with two HC types (Sandmann et al. 1996; Mills et al. 2014). It is not clear how could this be implemented in species like mice that possess merely a single HC type (Peichl & González‐Soriano, 1994) – here, local processing may be a way out, with cone synapse‐specific feedback modulating different cones (Fig. 3 E).

Setting the cones’ operational range

Horizontal cells have been proposed to be responsible for adjusting the operational range of cones (Burkhardt, 1995) by enabling them to respond with graduated glutamate release over a broad range of light intensities (Thoreson & Mangel, 2012). However, HC receptive fields are much larger than those of single cones (Shelley et al. 2006), and therefore the mean input an HC receives may differ tremendously from that of the cone it provides with feedback. In non‐mammalian vertebrates this difference is particularly pronounced when different chromatic tuning with opposite polarity comes into play (Kamermans et al. 1991; Pottek et al. 1997). As a consequence, HCs and cones may respond very differently to the same light stimulus, arguing against a critical role of (global) HC feedback in controlling the cones’ operational range. However, with local cone synapse‐specific feedback, independent control of each cone's output by the postsynaptic HC dendritic tips would be possible (Fig. 2 B). In this case, HCs could keep each cone within its operational range, which may also preserve cone signal properties independent of a global subtraction of the mean ambient light level. In fact, cone synapse‐specific gain control may also help address different requirements of postsynaptic BCs. For example, the high frequency‐encoding BC type in the ground squirrel expresses high‐affinity glutamate receptors, which saturate quickly and therefore do not profit from a broad glutamate concentration range (Grabner et al. 2016).

Discussion

To get a deeper insight into HC function at the cone synapse, one must take advantage of the powerful genetic tools which induce biosensor expression in outer retinal neurons for measuring voltage (Nakajima et al. 2016), chloride concentration (reviewed by Arosio & Ratto, 2014) or calcium changes (Akerboom et al. 2012) in subcellular and synaptic structures as well as glutamate release (Marvin et al. 2013). This could be complemented by the reconstruction of individual cone axon terminals and the contacting HCs, using data from serial‐scanning electron microscopy (e.g. in Helmstaedter et al. 2013; Behrens et al. 2016) (Fig. 1 B) or super‐resolution light‐microscopy techniques such as stimulated emission depletion microscopy (STED). Both methods could give experimental access to the precise anatomical organization of the synaptic circuitry in the outer plexiform layer. Furthermore, the development of computational models has begun to provide a quantitative view of HC feedback mechanisms (Wang et al. 2007; Gardner et al. 2015). However, currently these models only capture specific features of HC function; more complex aspects, such as the interaction between feedback mechanisms or the spatial scale of signalling (i.e. local vs. global modes of HC processing), have yet to be addressed. For such a comprehensive biophysical HC model, it is essential to know the exact morphology, including the location of synaptic contacts, as well as the spatial distribution of the different types of channels and receptors.

The approaches outlined above would also allow investigation into whether microdomains exist in cone axon terminals or HC dendrites, and if so, what mechanisms support their functional segregation. Here, studying the electrical properties of HCs is a good starting point: even in bright light, when electrical coupling between HCs is reduced (Xin & Bloomfield, 1999), signals may still easily spread within an HC's dendritic arbour (and in neighbouring HCs). Therefore, the mixture of passive and active membrane properties in the distal HC dendrites – such as small process diameters (‘spine necks’) or expression of potassium channels (Feigenspan et al. 2009) – may reduce the spread of electrical signals, analogously to dendritic spines in the brain (reviewed by Parajuli et al. 2017). Additionally, the absence (or down‐regulation) of VGCCs in more proximal dendrites would help to support local signalling in the dendritic tips, which then could serve as postsynaptic microdomains.

In summary, in view of decades of research on HCs and the fact that less than 20% of the retina's neuron types contribute to the circuits in the outer retina, it is somewhat surprising that a comprehensive picture of HCs and their precise function – a picture that is not restricted to a particular species – is still elusive. Even considering only two admittedly extreme cases – primate and mouse – offers a glimpse of the broad of functional diversity in a retinal cell class with less than a handful of types.

Additional information

Author contributions

All authors have approved the final version of the manuscript and agree to be accountable for all aspects of the work. All persons designated as authors qualify for authorship, and all those who qualify for authorship are listed.

Funding

The authors have received the following funding: Deutsche Forschungsgemeinschaft (DFG): T.E., T.S., EXC307; Deutsche Forschungsgemeinschaft (DFG): T.S., SCHU 2243/3‐1.

Biographies

Camille A. Chapot obtained her MSc in Neurosciences at the University of Strasbourg (France) and is currently a PhD student at the Institute for Ophthalmic Research/Centre for Integrative Neuroscience at the University of Tübingen (Germany).

Thomas Euler obtained a diploma in Biology and a PhD in Neuroscience at the University of Mainz and the Max‐Planck Institute for Brain Research in Frankfurt. After being a postdoctoral fellow at the Harvard Medical School in Boston and the MPI for Medical Research in Heidelberg, he is now a full professor for Ophthalmology at the Institute for Ophthalmic Research/Centre for Integrative Neuroscience in Tübingen.

Timm Schubert studied Biology in Oldenburg, where he also obtained his PhD in Neuroscience. He was a postdoctoral fellow at the University of Washington in Seattle, and is now a senior researcher at the Institute for Ophthalmic Research/Centre for Integrative Neuroscience in Tübingen.