Transmission at rod and cone ribbon synapses in the retina

Abstract

Light-evoked voltage responses of rod and cone photoreceptor cells in the vertebrate retina must be converted to a train of synaptic vesicle release events for transmission to downstream neurons. This review discusses the processes, proteins, and structures that shape this critical early step in vision, focusing on studies from salamander retina with comparisons to other experimental animals. Many mechanisms are conserved across species. In cones, glutamate release is confined to ribbon release sites although rods are also capable of release at non-ribbon sites. The role of non-ribbon release in rods remains unclear. Release from synaptic ribbons in rods and cones involves at least three vesicle pools: a readily releasable pool (RRP) matching the number of membrane-associated vesicles along the ribbon base, a ribbon reserve pool matching the number of additional vesicles on the ribbon, and an enormous cytoplasmic reserve. Vesicle release increases in parallel with Ca²⁺ channel activity. While the opening of only a few Ca²⁺ channels beneath each ribbon can trigger fusion of a single vesicle, sustained release rates in darkness are governed by the rate at which the RRP can be replenished. The number of vacant release sites, their functional status, and the rate of vesicle delivery in turn govern replenishment. Along with an overview of the mechanisms of exocytosis and endocytosis, we consider specific properties of ribbon-associated proteins and pose a number of remaining questions about this first synapse in the visual system.

Introduction

Synaptic transmission from rods and cones to second-order retinal neurons is a key early step in the conversion of light responses into a neural code for vision. The task of this synapse is to convert light-evoked changes in membrane voltage into a sequence of synaptic vesicle release events. The properties of transmission at this first synapse in the retina shape visual capabilities in many fundamental ways [199]. For example, transmission at photoreceptor synapses acts as a bandpass filter [5, 230, 244]. Non-linearity in transmission at cone synapses preferentially enhances bipolar cell responses to small contrast changes [210]. Non-linear properties of transmission at rod synapses also filter out small noisy signals to increase the reliability of rod bipolar cells in detecting single photon responses [71]. Properties of release improve the detection of light decrements [106] and initiate the coding of changes in contrast vs. luminance [9, 158]. Modification of release by feedback from horizontal cells to cones yields spatiotemporal filtering that improves edge detection [212]. Inhibitory feedback from horizontal cells to cone synapses is also critical for performing spectral comparisons [211].

Our understanding of the mechanisms of release at photoreceptor synapses has come from studies using a number of animal models. Because of the relatively large size of its photoreceptor cells, salamander retina has served as a particularly useful model system. This review summarizes our understanding of photoreceptor synaptic function, focusing on results obtained in salamander retina with comparisons to results obtained in other animal models.

Anatomy and molecular composition of rod and cone ribbon synapses

Photoreceptor cell anatomy

Rod and cone photoreceptor cells consist of two largely independent compartments, inner and outer segments, separated by a thin non-motile cilium (Fig. 1a). The outer segment is specialized for phototransduction [7, 31, 69] while the inner segment contains ion channels that shape the light response along with the metabolic and synaptic machinery. A cluster of many mitochondria sits in the outer region of the inner segment just beneath the connecting cilium. These mitochondria provide ATP needed to sustain the activity of Na/K-ATPases as they work to counter the continuous influx of Na⁺ ions through cGMP-gated cation channels in the outer segment that remain open in darkness [104, 160]. The buffering of Ca²⁺ by mitochondria also segregates Ca²⁺ changes in the inner segment from changes in the outer segment [74, 120, 213].

Fig. 1.

Rods, cones, and synaptic vesicle pools at photoreceptor ribbon synapses. a Rod and cone anatomy. b Electron micrograph of a salamander cone synapse showing two electron-dense ribbons situated above their arciform densities. The readily releasable pool (RRP) of vesicles at the base of each ribbon is colored in yellow. The ribbon-attached reserve pool is shown in blue. Many vesicles surround both sides of the ribbon, forming a cytoplasmic pool. c Confocal image of a salamander cone (yellow) and horizontal cell (magenta) filled with Lucifer yellow and sulfarhodamine B, respectively, during paired whole cell recording. d Post-synaptic current (PSC) recorded in a horizontal cell evoked by a long depolarizing step applied to a presynaptic cone. Inset shows a magnified view of the initial transient component of the PSC (adapted from [19]). e Charge transfer from the PSC in D was fit with a function consisting of two exponentials and a straight line. The initial fast component corresponds to the RRP, the second slower component likely reflects a ribbon reserve pool, and the linear function likely represents replenishment from the cytoplasmic reserve

In mammalian retina, the synaptic terminals of rods and cones contain their own mitochondria to fuel transmission. However, the synaptic terminals of rods and cones in avascular retinas, including amphibians, lack mitochondria. These instead rely on energy supplied by mitochondria in the inner segments. In these animals, the mitochondria generate phosphocreatine to escape consumption by ATPases in the inner segment, allowing it to diffuse to the synaptic terminal where creatine kinase converts it to ATP [102].

Anatomy of photoreceptor ribbon synapses

Perhaps the most notable ultrastructural feature in the synaptic terminals of rods and cones is an electron-dense structure known as the ribbon [57] (Fig. 1b). In cross-section, ribbons appear as dark bars jutting into the cytoplasm perpendicular to the plasma membrane. In three dimensions, ribbons form flat planar structures with a thickness of ~ 35 nm [168, 193, 200]. Glutamate-filled vesicles attach to the ribbon surface in a hexagonal array. Each vesicle is tethered to the ribbon by 3–5 fine filaments [223]. Beneath each rod and cone ribbon rests a trough-shaped structure known as the arciform density (Fig. 1b).

Rod ribbons are generally larger than cone ribbons [168, 200]. In mouse and salamander rods, ribbons extend ~ 1 μm into the cytoplasm and ~ 1 μm along the membrane. Ribbons of mouse rods curl around the synaptic invagination, forming a horseshoe-shaped profile [81, 125, 173, 193, 250]. Mouse rods typically have only a single ribbon [41] but cat and primate rods often have two ribbons within each invagination [149]. Salamander rods can have as many as 3 terminals with each terminal possessing an average of 4 ribbons, yielding an overall average of ~ 7 ribbons per rod [222, 233].

Cone ribbons typically have dimensions of a few hundred microns along each side [3, 163, 168]. Unlike mouse rods that have only a single ribbon, cones have multiple ribbons with 10–15 apiece in mouse and an average of 13 apiece in salamander [19, 112, 163]. Ground squirrel cones [127] and foveal cones of primates have ~ 20 ribbons apiece, with the number of ribbons in primate cones increasing towards the periphery to ~ 40 ribbons apiece [48, 114].

Ribbons are also found in other sensory neurons that encode graded membrane potential changes. Retinal bipolar cells have smaller plate-like ribbons while hair cells of the auditory and vestibular apparatuses have small, ovoid ribbons [153].

Rods and cones form synapses with dendrites of second-order bipolar and horizontal cells. In both rods and cones, ribbons rest at the apex of an invagination into the synaptic terminal membrane (Fig. 1b). Mouse rod ribbons typically contact two ON-type rod bipolar cells [24] but as many as four rod bipolar cell dendrites may enter each invagination in primate rods [85]. In cones, ON-type bipolar cell dendrites occupy the central position within the invagination whereas OFF-type bipolar cells make flat contacts at the base of the synapse just outside the invagination [91]. A few flat, basal contacts from cone bipolar cells can also be seen just outside the invagination in rods [128, 140]. Unlike mammals, central bipolar cell dendrites in salamander retina often come from OFF bipolar cells whereas ON bipolar cells often make basal contacts [123]. Flanking the central bipolar cell dendrites within the invagination is a pair of horizontal cell processes that can provide inhibitory feedback to rods and cones [23, 178, 207, 212]. Bipolar cell circuitry is reviewed in detail elsewhere [68].

Molecular constituents of synaptic ribbons

Proteomic analysis has identified many different proteins associated with ribbons, but the most prevalent is a transcript variant of CtBP2 known as ribeye [111, 224]. The two variants of CtBP2 share a common B domain but ribeye has an additional ribbon-specific A domain. The shorter, more ubiquitous form of CtBP2 is present in the nucleus of many cells [196]. The shared B domain of CtBP2 contains a transcriptional repressor as well as lysophosphatidic acid acyltransferase (LPAAT) but the role that these domains play, if any, at ribbon synapses is not known [190]. The core structure of each ribbon is formed from tens of thousands of ribeye molecules [185, 186, 252] with the ribbon-specific A domain apparently facing the center [137]. The laminar appearance of photoreceptor ribbons is consistent with this arrangement (see Fig. 6 in [122]) and the 35-nm thickness of photoreceptor ribbons matches the dimensions of two parallel rows of ribeye proteins [154]. The related protein CtBP1 is also associated with ribbons, but deletion of CtBP1 had only mild effects on rod synaptic structure and function in mouse retina [225].

In heterologous expression systems, ribeye molecules coalesce into spherical bodies, roughly similar in shape to hair cell ribbons [186]. Spherical synaptic bodies are also found in the terminals of photoreceptors from hagfish, a primitive jawless fish [98]. The formation of the planar structure of retinal ribbons requires interactions with piccolino, a ribbon-specific piccolo splice variant [154]. The putative motor protein myosin Va [184] may also help with the construction of planar ribbons [129].

Interactions between adjoining ribeye molecules are sensitive to NADH and it is thought that metabolic changes in NADH levels that accompany changes in light and dark might be the mechanism controlling disassembly of rod ribbons with illumination in day-time and their re-assembly at night [1, 2, 59, 137, 186, 195]. Ribbons also disassemble during hibernation in ground squirrel cones and then rapidly re-assemble after waking [144, 178].

Kif3A is a kinesin motor protein associated with ribbons [155, 220] fueling speculation about motor-driven movement of vesicles. However, motor functions of Kif3A typically require heteromeric association with Kif3B or 3C [169], and these do not appear to be present at ribbons [111, 224]. Elimination of Kif3A causes photoreceptor degeneration but this appears to be a result of disturbed ciliary transport processes [108]. Furthermore, the release of vesicles on the ribbon can proceed without ATP suggesting that an ATP-dependent motor is not required for movement along the ribbon [95].

Composition and function of the arciform density

The trough-like arciform density that rests beneath each photoreceptor ribbon (Fig. 1b) is not found at other ribbon synapses. It is a complex of multiple proteins including bassoon, Munc13–2, Rim2, and ERC2/CAST1 [30, 49, 58, 88, 220, 221]. Loss of bassoon leads to detached ribbons and diminished synaptic transmission from rods [10, 63]. (We return later to discuss differences in the effects of bassoon and Ribeye knockouts.) After deleting CAST1, ribbons remain attached but the size of the active zone is reduced [220].

In addition to the particular functions performed by specific proteins that constitute the arciform density, a more general function is that it may shape the diffusion of Ca²⁺ ions entering channels positioned just beneath it. Simulations of salamander cones suggested that a diffusion barrier created by the arciform density could limit the spread of Ca²⁺ and thereby limit the ability of an open Ca²⁺ channel to trigger release of multiple vesicles [20]. This could in turn help to ensure a linear relationship between I_Ca activation and release [20, 113]. However, rods and cones are also capable of multivesicular release, and it was proposed for hair cell ribbons that constraining Ca²⁺ diffusion beneath the ribbon can enhance the likelihood of multiquantal release by promoting local buffer saturation [82].

Exocytotic proteins

Like most neurons, exocytosis of glutamate-filled vesicles at rod and cone ribbons synapses involves SNARE proteins. The vesicle-associated SNAREs, synaptobrevin and SNAP-25, are the same as those used at other synapses [238], but the t-SNARE on the plasma membrane, syntaxin 3B, is not used as widely [53, 151]. As discussed later, this SNARE protein promotes vesicle-vesicle fusion on the ribbon [92] but may also play other roles [53, 131].

In addition to this atypical SNARE protein, rods and cones also use complexins 3 and 4 rather than the more commonly used isoforms 1 and 2 [177]. The properties conferred by complexins 3 and 4 are qualitatively similar to properties of complexins at other mammalian synapses, e.g., limiting spontaneous release and facilitating evoked release, but they may also play a distinct role in regulating replenishment of vesicles to the ribbon [13, 152, 226, 228].

What are the molecular constituents of filaments that tether vesicles to the ribbon?

It was suggested that interactions between rab3A and partner molecules on the ribbon such as Rim (rab-interacting molecule) might form the fine tethers that attach vesicles to the ribbon [224]. Each vesicle has 10 or more copies of rab3A [245], which would support the observed number of tethers. More convincingly, peptides that interfere with rab3A function impaired vesicle replenishment [219]. However, while it was originally thought that Rim1 was present along the ribbon face [220], it was later shown that Rim1 antibody labeling on the ribbon was due to cross-reactivity with piccolino [133]. Rim2 is present at the arciform density but not along the ribbon face. The Rim2 variant in photoreceptors also exhibits weaker binding to rab3A than full-length Rim2 [133]. Although rab3a is not likely to interact with Rim proteins along the face of the ribbon, it may interact with other ribbon-associated proteins, such as rabphilin [224].

Release at rod and cone synapses

Ca²⁺ channels

Rods and cones maintain a relatively depolarized membrane potential in darkness, hyperpolarizing in response to increased light intensity and depolarizing to decreased intensity. Rods and cones lack voltage-gated Na⁺ channels and so light-evoked voltage changes spread passively from the outer segment to the synaptic terminal, modified by voltage-and Ca²⁺-dependent ion channels along the way [229]. These voltage changes regulate the activity of synaptic Ca²⁺ channels to control release of glutamate-filled synaptic vesicles [see review by 164], thereby converting light-evoked voltage changes into a train of vesicle release events. Unlike conventional synapses that use N or P type Ca²⁺ channels, ribbon synapses rely on L-type Ca²⁺ channels [187, 213, 246]. Photoreceptors use a particular type of L-type channel, Ca_V1.4, which shows minimal Ca²⁺- and voltage-dependent inactivation [22, 115, 142, 164, 240]. Over 100 mutations in Ca_V1.4 that have been identified, most of which cause some form of congenital stationary night blindness (CSNB) by impeding transmission of signals from rods [251].

Salamander cones possess 50–75 channels/ribbon or 3–4 channels for every vesicle [20]. The single channel properties of Ca²⁺ channels in salamander rods are similar to other L-type channels with a peak open probability approaching 0.3 and single channel conductance in Ba²⁺ of 20 pS [214] (Fig. 2a). By contrast, single channel recordings from heterologously expressed Ca_V1.4 channels recorded under similar conditions showed a single channel conductance of only 9 pS and peak open probability of 0.05–0.1 [32, 64]. Given these values, a macroscopic I_Ca in mouse cones of 45 pA would require 3500–7000 channels/cone or 200–700 channels per ribbon.

Fig. 2.

Changes in Ca²⁺ channel activity during light-evoked voltage responses of photoreceptors. a Ca²⁺ channel openings recorded in the cell-attached patch mode from a photoreceptor terminal in salamander retina with 82 mM Ba²⁺ in the pipette. b Overlaid series of voltage responses evoked in a salamander cone by light flashes of increasing intensity. c Whole-cell Ca²⁺ current (I_Ca) plotted against voltage evoked by a ramp voltage protocol in a cone. Voltage was corrected for access resistance and liquid junction potential [79]. The vertical dashed line indicates the typical photoreceptor resting potential in darkness of − 40 mV. The arrow shows the reduction in I_Ca that would accompany a bright light flash

The lack of Ca²⁺-dependent inactivation in Ca_V1.4 channels is due to an autoinhibitory domain in the C-terminus that competes with binding of apo-CaM (Ca²⁺-free CaM). Because apo-CaM and the autoinhibitory domain both compete for the same binding site, elevated CaM levels can enhance Ca²⁺-dependent inactivation. Phosphorylation can also promote apo-CaM binding and thereby enhance inactivation [87, 124, 182, 206]. In addition, some splice variants of Ca_V1.4 lack this autoinhibitory domain altogether [87]. Mechanisms such as these may explain the slow Ca²⁺-dependent inactivation seen in salamander rods (τ = 1.7s) [50, 172].

In addition to genuine Ca²⁺-dependent inactivation, depletion of Ca²⁺ ions from the synaptic cleft during maintained activation can also limit the amplitude of I_Ca. Similar to calyceal synapses [28, 197], the continued influx of Ca²⁺ ions at invaginating rod synapses as they remain depolarized in darkness can deplete the limited number of Ca²⁺ ions available in the tiny volume of the synaptic cleft causing a slow, overall decline in I_Ca [172].

Along with differences in the degree of Ca²⁺-dependent inactivation, some of the more than 20 splice variants of Ca_V1.4 differ in their voltage-dependence [87]. This variation can extend the range over which light-evoked voltage changes regulate channel activity. Voltage-dependence of Ca_V1.4 is also regulated by association with a Ca²⁺-binding protein CaBP4 that shifts the voltage-dependence of Ca_V1.4 to more negative potentials, providing a better match to the voltage range attained during rod and cone light responses [86].

Ca_V1.4 channels form a complex with intracellular β2 subunits and extracellular α2δ4 subunits [15, 249]. These subunits play a role in trafficking to the membrane but also regulate channel activity [65]. A common variant of the β2 subunit in human retina enhances voltage-dependent inactivation [124]. α2δ4 subunits are anchored to the extracellular membrane surface by glycosylphosphatidylinsolitol (GPI) links [56]. Interactions between α2δ4 and another extracellular protein, ELFN1, are critical for proper formation of rod synapses during development [112, 242]. Structural elements of the alpha subunit Ca_V1.4, independent of its ability to conduct Ca²⁺, also play a role in stabilizing rod synapses [136].

Ca²⁺-activated chloride channels

Ca²⁺-activated Cl⁻ channels are localized to synaptic terminals of rods and cones, but distributed more diffusely throughout the terminal than Ca²⁺ channels [135, 146]. The Ca²⁺-activated Cl⁻ channels in rods appear to be Ano2 channels while those of cones are a different subtype [55, 201]. In addition to regulating transmembrane voltage, chloride ions that move through these channels can regulate the open probability of L-type Ca²⁺ channels by interacting with anion-binding sites on the intracellular C-terminus and the beta subunit [11, 213, 214]. The chloride equilibrium potential in rods is relatively depolarized at ca. − 20 mV so activation of Ca²⁺-activated Cl⁻ channels at or below the resting membrane potential promotes an efflux of Cl⁻ that can in turn reduce Ca²⁺ channel activity [209]. This negative feedback interaction may provide a mechanism for adaptation and help to limit regenerative activation of Ca²⁺ channels. In ground squirrel cones with a resting membrane potential of ~ − 50 mV, E_Cl also has a relatively depolarized value of ~ − 30 mV [205], supporting the possibility of similar negative feedback interactions. In non-mammalian cones, E_Cl is closer to the resting membrane potential in darkness suggesting more limited effects of this mechanism to regulate Ca²⁺ channel activity in these cells [208].

Voltage-dependence of exocytosis

In most species, the resting membrane potential of rods and cones in darkness is around − 40 mV, close to the activation midpoint for Ca_V1.4 channels in vivo (Fig. 2c) [14, 80]. The limited inactivation of Ca_V1.4 channels allows them to remain active at the dark potential, smoothly altering activity to continuously encode light-evoked changes in membrane potential. The resting potential is near the steepest part of the activation curve, maximizing sensitivity to small changes in membrane potential. Channel activity diminishes with the membrane hyperpolarization evoked by light. The non-linear nature of I_Ca activation contributes to a greater change in cone output produced by small changes in contrast compared to larger contrast changes that produce larger voltage excursions [210]. With a very bright flash, the membrane can hyperpolarize to a level at which there is virtually no Ca²⁺ channel activity. This contributes to the “clipping” of rod signals in second-order neurons seen with bright flashes, in which the initial, transient component of the hyperpolarizing rod response evoked by bright light is transmitted quite poorly [8, 217].

Many substances can modulate Ca²⁺ channels at rod and cone synapses [164, 229]. One notable difference from Ca²⁺ channels at most other synapses is that the normal resting membrane potential in photoreceptors is near the midpoint for I_Ca activation. One consequence of this is that small differences in resting membrane potential between cells can cause significant cell-to-cell differences in sensitivity to the same light-evoked response. This arrangement also means that the activity of photoreceptor Ca²⁺ channels is unusually sensitive to small shifts in activation along the voltage axis. Thus, the local ionic environment can influence voltage-dependence significantly. For example, lowering the extracellular concentration of cations (e.g., divalent cations or protons) reduces positive charge on the outer membrane surface, altering the local electrical field experienced by Ca²⁺ channels and allowing them to be activated more easily [159, 167]. Thus, depletion of Ca²⁺ or H⁺ from the synaptic cleft produces a negative (leftward) shift in the activation curve for I_Ca. Changes in anion levels can also alter membrane surface charge, shifting activation in the opposite direction from changes in cation levels [216].

The retina has exploited the sensitivity of Ca²⁺ channels to protons [103, 159] as a means of regulating photoreceptor synaptic output [16, 17]. The co-release of protons with glutamate from synaptic vesicles causes rapid, transient inhibition of presynaptic I_Ca in cones [61, 101]. And while other mechanisms may contribute, an essential component of inhibitory feedback from horizontal cells to cones and rods involves proton-mediated changes in Ca²⁺ channel activation [212]. The hyperpolarization of horizontal cells in response to light alkalinizes the synaptic cleft, relieving proton-mediated inhibition of I_Ca and, more importantly, shifting I_Ca activation leftward to more negative potentials [35, 97, 236, 237, 241]. Together, these effects of horizontal cell feedback promote recovery of I_Ca when surrounding horizontal cells are hyperpolarized (e.g., by an increase in spatially-averaged luminance). And because the extracellular environments of invaginating synapses remain relatively isolated from one another, individual cone ribbons can be regulated independently by dendrites of different horizontal cells, fine-tuning the feedback signal in response to local changes in illumination [43, 80].

Ribbon-to-ribbon differences in ionic content arising from differences in the distribution of channels and transporters may also contribute to differences in the voltage-dependence of Ca²⁺ channels at different ribbons in the same cell [80]. The presence of multiple ribbons that differ somewhat in their sensitivity to light-evoked voltage changes, either because of the expression of different splice variants [87] or local differences in the ionic environment, may extend the voltage range for synaptic output available to a cone.

Release from cones occurs exclusively at ribbons

In collaboration with Dr. David Zenisek, we studied the role of ribbons in release using a technique known as fluorescence-activated laser inactivation (FALI). For these experiments, we introduced a fluorescein-conjugated peptide that selectively binds to ribeye into rods or cones through a patch pipette. The free radicals released upon activation of the fluorophore by bright laser light causes highly localized damage to the ribbon and nearby proteins. Including high antioxidant concentrations minimizes non-selective damage. Acutely damaging salamander cone ribbons in this way reduced both fast and slow components of release from cones equally, indicating that both forms of release occur at ribbons [194]. Lowering Ca²⁺ buffering and inhibiting Ca²⁺ extrusion to promote the spread of Ca²⁺ to non-ribbon sites also did not reveal any additional slower components to release from cones [233]. These data indicated that release from salamander cones occurs only at their ribbons. Evidence from ground squirrel cones where ribbons re-assemble rapidly after awakening from hibernation indicates an essential role for ribbons in mediating release [144]. Another study in ground squirrel retina analyzed miniature excitatory post-synaptic currents recorded simultaneously in different classes of bipolar and horizontal cells. The different cell types differed in distance from ribbon release sites, and it was found that synaptic events in more distant cells were smaller and showed slower kinetics. This indicated that the ribbon was the sole source of glutamate for all of the observed events [62]. The exclusive use of ribbons in cones differs from rods (see below) and retinal bipolar cells [143, 148] that are also capable of release at ectopic, non-ribbon sites.

Vesicle pools at ribbon synapses

The vesicle pools involved in release from cone ribbons in salamander retina have been analyzed using both membrane capacitance recordings and paired recordings from cones and horizontal cells (Fig. 1b) [19, 170]. As illustrated in Fig. 1d, stimulating a cone with a long depolarizing step evokes a fast inward current in horizontal cells that rapidly subsides to a smaller, more sustained current. There is little saturation or cross-desensitization among neighboring glutamate receptors so miniature post-synaptic currents evoked by release of individual vesicles sum linearly with one another at the cone-horizontal cell synapse in salamander retina [33, 36]. One can therefore measure the total amount of release during the test step by integrating the post-synaptic current to measure the total charge transfer (Fig. 1e). This approach revealed three components to release at salamander cone synapses: (1) A fast, transient component released within a few milliseconds consisting of ~ 15–20 vesicles/ribbon [19, 218] (dotted green trace, Fig. 1e). This represents a readily releasable pool (RRP) of vesicles and matches the number of ribbon-associated vesicles that contact the plasma membrane along the bottom two rows of the ribbon (yellow vesicles, Fig. 1c). Ground squirrel cones have a similar size RRP [77]. (2) With maintained depolarization, a second component consisting of ~ 90 vesicles per ribbon can be depleted from salamander cones within a few hundred milliseconds [19, 105] (dotted blue trace, Fig. 1e). The size of this secondary reserve pool matches the remaining number of vesicles further up the ribbon (blue vesicles, Fig. 1c). (3) After emptying these first two pools, release can be maintained at a linear rate more or less indefinitely (dotted gray trace, Fig. 1e), likely reflecting replenishment of ribbon-associated vesicles from the cytoplasmic pool [19]. Using similar approaches, we found that the RRP in salamander rods averages ~ 25 vesicles/ribbon [233]. Membrane capacitance recordings from mouse rods suggest an RRP of ~ 90 vesicles, also consistent with the number of membrane-associated vesicles at the base of mouse rod ribbons [76].

Graded voltage changes in rods and cones are converted to discrete vesicle release events. The number of discriminable intensities that can be encoded is thus limited by the size of the releasable pool and release rate [106, 199]). One way that cones address this limitation is by distributing release over many different ribbons, thereby increasing the effective size of the releasable pool. In addition, as mentioned earlier, differences in the voltage-dependence among ribbons can expand the intensity range over which an individual cone can encode voltage changes [80].

Ca²⁺-dependence of exocytosis

The quantal hypothesis of synaptic transmission postulates that the post-synaptic current is a product of three variables, n × P × Q, where n = the number of vesicles in the RRP, P = probability that a vesicle in that pool will be released, and Q = quantal content of a single vesicle. While we initially thought that stronger stimulation of cones might recruit more distant release sites and thus increase N, our analysis of post-synaptic currents in salamander cones showed that voltage-dependent changes in release were determined solely by changes in P and that the size of the RRP appears fixed [218]. A recent study came to a similar conclusion for ribbon-mediated release from mouse rods [76]. While cytoplasmic glutamate levels can influence Q, changes in transporter activity or cytosolic glutamate levels are generally too slow to encode light responses directly [21].

The number of open Ca²⁺ channels determines release probability, P, and release rises linearly with increases in I_Ca [215]. One important factor contributing to this linearity is that release at cone ribbons is regulated by highly localized Ca²⁺ changes within nanodomains < 100 nm from Ca_V1.4 Ca²⁺ channels [146]. As discussed earlier, this behavior may require the diffusion barrier formed by the arciform density. As I_Ca increases, more channels are activated allowing a larger number of vesicle release sites to be engaged. The amount of release thus rises in parallel with channel activity. The Ca_V1.4 Ca²⁺ channels that control release appear to be grouped in small clusters beneath the ribbon [134, 175] and release of a single vesicle in the RRP of salamander cones can be triggered by 2–3 Ca²⁺ channel openings [20]. Salamander rods are a bit less efficient, requiring the opening of 3–5 Ca²⁺ channels to trigger release of a vesicle [233].

Sensitivity of the exocytotic Ca²⁺ sensor molecules to intracellular Ca²⁺ levels may also contribute to linearity in release [66, 180, 215]. Ca²⁺ sensitivity of release is most accurately measured using caged Ca²⁺ compounds to abruptly elevate Ca²⁺ levels throughout the cell by photolysis with a bright UV flash. Photolytic uncaging elevates Ca²⁺ levels evenly throughout the cell so that average changes in cytoplasmic Ca²⁺ levels are the same as those experienced by the exocytotic Ca²⁺ sensor just beneath the plasma membrane. Using this technique, we measured release kinetics from the exocytotic rise in membrane capacitance accompanying vesicle fusion and by the kinetics of post-synaptic currents evoked in horizontal cells during paired recordings. Consistent with earlier findings [180], our results showed that the sensor in salamander rods and cones has a very high sensitivity to Ca²⁺, with a threshold of only a few hundred nanomolar. There is also evidence for a lower affinity sensor in salamander cones [118]. Our studies showed that the high affinity sensor has an intrinsic Ca²⁺ cooperativity of 2–3 [66, 215]. The high affinity and low cooperativity of this sensor differs from most synapses that typically show lower affinity and higher cooperativity (n = 4–5). Simulations indicated that the high affinity of rod and cone sensors arises from a slow off-rate for Ca²⁺ unbinding [66]. This would extend the time that the sensor is partially occupied with Ca²⁺ during which binding of one or two additional Ca²⁺ ions might be enough to evoke release. This property may also help to linearize the overall Ca²⁺ dependence of release at rod and cone synapses.

The exocytotic Ca²⁺ sensors that mediate fast release at most conventional synapses, synaptotagmin 1 (Syt1) and 2, are not found at photoreceptor synapses of salamanders or fish [27, 96] but Syt1 is present at rod and cone synapses in mammalian retina [72]. Selective genetic elimination of Syt1 from mouse cones totally abolished their release [79]. Eliminating Syt1 from mouse rods abolished fast, synchronous release evoked by brief stimuli but a slower component remained intact. One question is whether mouse rods and cones that use Syt1 also exhibit the high affinity and low Ca²⁺ cooperativity found in amphibians. It is generally thought that the cooperativity of n = 4–5 reflects the presence of 5 Ca²⁺-binding sites in Syt1 so it would be interesting to know if this property is retained at Syt1 synapses of rods and cones.

Vesicle cycle at rod and cone ribbons

Influence of replenishment on kinetics

Figure 3 illustrates the key steps in the vesicle cycle at photoreceptor ribbon synapses. By analogy with the behavior of electrical capacitors, plate-like ribbons “charge” with vesicles along their surface in light when release is minimal (i.e., when photoreceptors are hyperpolarized). Ribbons can then discharge vesicles in a rapid burst when photoreceptors depolarize upon return to darkness [106]. The large burst of release during an abrupt return to darkness expands the number of discriminable intensities 2.7 fold [106]. This enhancement of release at light offset may be one mechanism that contributes to the greater sensitivity of human observers to light decrements compared to increments [29]. After this initial burst, release slows to a rate governed by the rate at which vesicles descend the ribbon to re-occupy available release sites. The fraction of the RRP released at any given membrane potential is determined by the release probability and that is in turn determined by the fraction of open Ca²⁺ channels. At a constant membrane potential, the fraction of the RRP that that is available for replenishment (i.e., the number of empty release sites at the base of the ribbon) remains constant, neither growing nor declining [9]. This behavior in cones matches that of bipolar cells where the initial burst of release at light offset has been shown to provide a measure of the change in contrast, whereas the rate of sustained release provides a measure of luminance [158]. Slowing vesicle replenishment (by inhibiting endocytosis or using calmodulin inhibitors as described later) impairs the ability of the synapse to following flickering light stimuli showing that the rate at which vesicles can replenish release sites is important for shaping frequency response characteristics [230].

Fig. 3.

Diagram of the vesicle cycle at a photoreceptor ribbon synapse. Vesicles in the readily releasable pool (RRP)at the base of the ribbon dock at the membrane and can fuse within ~ 50 ms. Vesicular proteins and lipids released into the membrane must then be removed to prepare the release site for further release. This involves rapid endocytosis with a time constant (τ) of ~ 400 ms. After reforming and refilling vesicles with glutamate, they are returned to the large cytoplasmic pool of vesicles. Vacant sites on the ribbon are replenished from the cytoplasmic pool via both a rapid Ca²⁺-/CaM-dependent mechanism (τ ~ 700 ms) and slower Ca²⁺-independent mechanism (τ ~ 13 s). As vesicles move down the ribbon, they are primed for release. Illustration redrawn and adapted from [230]

Post-synaptic factors also play a role in recovery and kinetics [60, 205]. For example, the time constant for presynaptic replenishment of the vesicle pool in ground squirrel retina is slower than the rate at which post-synaptic AMPA receptors in OFF bipolar cells show recovery [77]. At this synapse, post-synaptic receptors are saturated by glutamate well before the entire RRP is released and so restoring only a portion of the RRP can be sufficient to produce large post-synaptic responses, thereby speeding recovery of post-synaptic responses during replenishment.

In salamander retina, the situation appears different. At salamander cone synapses, there is a linear correlation between pre-and post-synaptic measurements of release over much of the response range [170]. In addition, post-synaptic currents evoked in horizontal cell by depolarizing stimulation of cones can be predicted from a simple linear sum of individual quantal release events [33, 36]. These capabilities arise because the glutamate released from each vesicle is separated sufficiently in space and time from other single vesicle release events at the same synapse [163]. This avoids the receptor saturation seen in ground squirrel OFF bipolar cells and provides enough time for AMPA receptors to recover from desensitization before another release event impinges on the same set of receptors [163].

Endocytosis

Rods and cones have enormous cytoplasmic pools of vesicles, much larger than those of conventional neurons that typically have only 100–200 vesicles per synapse. Terminals of salamander rods and cones contain ~ 80,000 and ~ 194,000 vesicles, respectively [191]. Even mouse rods with only a single ribbon possess 5800–7500 vesicles [250]. In addition, unlike conventional synapses where most vesicles are immobilized to the cytoskeleton by interactions with synapsin, photoreceptor and bipolar cell ribbon synapses lack synapsin [138] allowing 85% of the vesicles to move freely within the synapse [176]. This means that there is a huge pool of cytoplasmic vesicles readily available to replenish ribbon release sites and accommodate changes in release rate. Although capable of higher rates when maximally stimulated, sustained release rates in darkness in rods and cones appear to range from 10 to 20 v/s/ribbon, as measured by a number of different techniques [26, 94, 191]. However, even at 10 v/s/ribbon, the entire pool of vesicles in a salamander rod terminal would be emptied after 45 min of darkness without vesicle recycling.

Endocytosis at rod and cone ribbon synapses involves both fast and slow retrieval mechanisms [51, 105, 231]. Fast retrieval has a time constant of a few hundred milliseconds and can be inhibited in rods by inhibiting the GTPase dynamin [231]. By contrast, the fast component of retrieval in salamander cones was insensitive to GTPase inhibitors [231]. Studies involving differences in the release and retrieval of large and small dye molecules suggested that this fast component may involve kiss-and-run mechanisms [243]. The slower component involves both clathrin-mediated endocytosis [192, 239] and clathrin-independent mechanisms. Key endocytic proteins involved in clathrin-independent retrieval at mouse rod synapses include dynamin 3, syndapin, calcineurin, endophilin, and synaptojanin [73, 99, 192, 239]. These proteins are enriched in the periactive zone surrounding the ribbon where endocytic vesicles arise after stimulation [73, 239].

In addition to replenishing the pool of available vesicles, endocytosis is needed to clear release sites of vesicular proteins and lipids deposited during prior fusion. This process is essential for restoring release sites to functional status. Imaging of single Ca²⁺ channels tagged with fluorescent quantum dots showed that immediately after vesicle fusion, the active zone briefly expands, thereby increasing the average distance between Ca²⁺ channels and release sites [145]. The active zone also expands after endocytosis is impaired (e.g., by the dynamin inhibitor, dynasore), presumably as a result of the additional, un-retrieved membrane. Increasing the distance between Ca²⁺ channels and release sites reduces release efficiency by increasing the number of Ca²⁺ channel openings needed to trigger release [147]. This may contribute to impaired fusion when endocytosis is blocked, but a more significant factor appears to be changes in the lipid and protein composition of the release site [244]. TIRF imaging of single vesicles showed that vesicles that descend the ribbon are released almost immediately or else dock at the membrane to await later release. However, when endocytosis was inhibited, fewer vesicles were released immediately and many of the docked vesicles turned around and retreated back up the ribbon, something that was almost never observed under normal conditions. These data suggested that endocytosis is needed to restore release site function, but inhibiting endocytosis did not impair vesicle docking. As mentioned above, slowing replenishment by blocking endocytosis also impaired the ability of second-order neurons to follow flickering light stimuli [244].

Glutamate refilling

After vesicles are retrieved by endocytosis, they must be refilled with glutamate. While this has not been measured directly in rods and cones, vesicle filling rates at central synapses show a time constant of ~ 15 s [100]. Vesicular glutamate uptake can influence kinetics at conventional synapses that possess many fewer vesicles [157], but rods and cones have a much larger repository of vesicles available for release while waiting for newly retrieved vesicles to be refilled. So, in the short run (seconds), refilling is not likely to be rate limiting for vesicle replenishment.

Ribbon replenishment

Vesicle replenishment rates are often measured using a paired pulse protocol. This protocol begins by applying a strong depolarizing test pulse from a holding potential where all Ca²⁺ channels are closed. The pulse amplitude and duration of this test pulse is chosen to release all of the vesicles in the RRP. After waiting a certain interval, a second test pulse is then applied to measure how much of the RRP has been replenished during that interval. By testing a range of interpulse intervals, vesicle replenishment in salamander cones was found to exhibit two components with time constants of ~ 600–800 ms and ~ 10 s (at room temperature) [105, 171, 218, 230]. The fast time constant for replenishment in ground squirrel cones shows a similar rate of recovery, with time constants of 450–700 ms (at 32 °C) [77]. In salamander retina, the slower replenishment mechanism is Ca²⁺-independent whereas Ca²⁺ acting through calmodulin regulates the fast component [230]. Unpublished data from our laboratory suggests the fast component also involves CaMKII that is known to be associated with ribbons [111, 224]. Ca²⁺/CaM acts to increase the proportion of vesicles that are replenished by the fast mechanism, but does not alter replenishment time constants [230]. This indicates that Ca²⁺ acts on the attachment process rather than individual vesicles. If Ca²⁺ acted on individual vesicles to alter their attachment probability, then the overall attachment probability would be a weighted average of Ca²⁺-dependent and Ca²⁺-independent attachment probabilities, yielding a single time constant that varied with Ca²⁺. However, if Ca²⁺ acts on the ribbon to alter attachment, then this analysis predicts two time constants with Ca²⁺ varying the proportion of the two components, as found in the data [230].

Vesicle movement along ribbons

Each vesicle is attached to a ribbon by 3–5 fine tethers [223]. We discussed the possibility earlier that the tethers may represent interactions between rab3A on vesicles and a ribbon-associated protein [219, 224]. Similar to results from bipolar cells [132, 227], experiments in rods indicate that vesicles can attach anywhere along the face of the ribbon and move freely along its surface but are released only at sites along the base [222, 244]. If vesicles can move freely on the ribbon, then each individual tether must provide a relatively weak binding force. A strong tethering force would restrict lateral movements while a weak force allows individual tethers to detach and be replaced by another nearby tether. This can allow vesicles to move freely along the face of the ribbon while the combined force from multiple tethers remains sufficient to prevent vesicles from popping off the ribbon prior to fusion. Vesicles appear to migrate along the surface of the ribbon in a type of movement characterized as “crowd-surfing” [83]. Fusion of vesicles at the base of the ribbon creates vacancies that can be re-occupied by vesicles further up the ribbon. And because new vacant sites are only created at the base, vesicles will tend to diffuse passively from top to bottom without requiring active molecular motors [83], consistent with evidence that ribbon-associated vesicles can be released in the absence of ATP [95].

Vesicle priming

In addition to storing vesicles and regulating their delivery, ribbons play a role in priming vesicles to make them competent for release. After selectively damaging ribbons by FALI, the existing RRP can be released but subsequent release of additional vesicles is impaired. Electron micrographs showed that although subsequent fusion was impaired by damage to the ribbon, vesicles were able to attach to the ribbon and occupy release sites along the base of the ribbon [194]. Together, these results suggested that the impairment was specific to an essential priming step. Further evidence for the importance of the ribbon in preparing SNARE complexes for release comes from experiments showing that use of a peptide to inhibit formation of SNARE complexes did not inhibit release of either the RRP or the slower component thought to reflect the ribbon-attached pool but impaired further replenishment [54]. These data suggest that SNARE complexes begin to form after vesicles attach to the ribbon, but before membrane association. Also consistent with the idea that vesicles are primed during descent are TIRF imaging experiments showing that vesicles arriving at ribbon-associated release sites are released almost immediately [46, 244]. This is illustrated in Fig. 4 that shows a dye-filled vesicle approaching a rod membrane and progressively brightening as it advances through the evanescent field generated by TIRF microscopy. The vesicle then fuses, disappearing within a single 40 ms frame as dye is released. Immediate fusion after reaching the membrane is analogous to “crash fusion” observed in neuroendocrine cells [235] and differs from synapses in bipolar cells and the calyx of Held where vesicles pause at the membrane for a few hundred milliseconds before release, presumably to await completion of a final priming step [110, 150, 253]. Furthermore, Munc13–2 and RIM proteins that prime vesicles at most other synapses do not appear to be necessary for priming vesicles at photoreceptor ribbon synapses, and the Rim2 variant in photoreceptors lacks interaction sites with Munc13–2 [49, 133]. Although Rim2 does not participate in vesicle priming, it is important for enhancing Ca²⁺ influx through Ca_V 1.4 Ca²⁺ channels [75].

Fig. 4.

Imaging synaptic vesicle approach to the membrane in a salamander rod terminal. Total internal reflectance fluorescence microscopy (TIRFM) was used to image a vesicle approaching the membrane and then fusing at a rod synapse. The vesicle was loaded with 3kD dextran-conjugated AlexaFluor488. The diagram at the top illustrates a vesicle becoming progressively brighter as it advances towards the membrane through the evanescent field generated by TIRF illumination (length constant = 60 nm). As shown by the TIRFM images at the bottom, after approaching the membrane, the vesicle fused and rapidly released its contents, disappearing within a single 40 ms frame [244]

Multivesicular release

A fundamental purpose of synaptic ribbons is to facilitate the ongoing release of multiple vesicles. In addition to maintaining release of vesicles during maintained depolarizing stimulation, rod and cone ribbons are also capable of the synchronized, simultaneous fusion of multiple vesicles [92]. We have studied the properties of individual release events at rod and cone synapses by recording post-synaptic currents in horizontal and OFF bipolar cells as well as by recording the presynaptic anion currents activated during glutamate re-uptake into photoreceptor cells [6, 67, 78, 89, 166]. In salamander rods and cones, release events measured in both ways show widely varying amplitudes but similar kinetics suggesting that they arise from the synchronized fusion of multiple vesicles [36, 52, 92]. While small, presumptive single vesicle fusion events are more common, ~ 1/3 of the spontaneous events in salamander rods involve synchronous multiquantal fusion [92]. In collaboration with Ruth Heidelberger, we found that the SNARE protein syntaxin 3B plays an essential role in synchronous multiquantal release at salamander rod ribbons [92]. We hypothesized that the presence of syntaxin 3B on adjoining vesicles may facilitate homotypic fusion between neighboring vesicles on the ribbon prior to their fusion with the plasma membrane. In studies on release from bipolar cells, variations in quantal amplitude have been shown to improve the coding of information about contrast [107]; multivesicular release events may play a similar role at photoreceptor synapses.

Spontaneous release events in mouse cones range widely in amplitude but with similar kinetics, akin to synchronous multiquantal events in salamander rods and cones [79]. However, multiquantal release in mouse rods is more asynchronous. When mouse rods were voltage-clamped at − 70 mV, below the activation range for Ca²⁺ channels, we observed occasional spontaneous fusion events that were relatively uniform in amplitude suggesting they arose from individual vesicles (Fig. 5). With depolarization, the rate of vesicle release increased in parallel with I_Ca [94]. As the membrane potential approached − 40 mV, release transitioned from unitary events to coordinated bursts of multiple vesicles (Fig. 5). These bursts typically involved the asynchronous release of 10–20 vesicles from the RRP at semi-regular intervals of roughly once per second at room temperature [94].

Fig. 5.

Spontaneous and evoked glutamate release from a mouse rod. Glutamate release was assessed in rods by recording anion currents activated during glutamate re-uptake through presynaptic glutamate transporters. When the rod was held at − 70 mV, release consisted of occasional spontaneous, uniquantal events (bottom trace). When this rod was held at − 40 mV, release transitioned to a semi-regular series of multiquantal bursts consisting of ~ 17 vesicles apiece (top trace)

Absorption of a single photon hyperpolarizes mouse rods by only ~ 3.4 mV and produces an even smaller 1 mV response in primate rods [38, 188]. A 3.4 mV change in mouse rods causes ~ 20–25% reduction in I_Ca and a similar reduction in release [94]. One strategy that has been proposed as a means of distinguishing light-driven changes in release rate from random pauses is to release vesicles at an extremely high rate [173, 174, 234]. However, this strategy requires release rates of 80–100 v/s/ribbon, well above those found in mouse rods where release rises only to about 10–12 v/s/ribbon at the dark resting membrane potential in darkness. A second strategy is to make release more regular, and thus more predictable. This can be achieved by implementing an Erlang distribution that waits a certain number of Poisson intervals before triggering a release event [183]. This approach has been termed the “clockwork” mechanism. The sensitivity to small voltage changes and regularity of multiquantal bursts in mouse rods may improve post-synaptic detection of single photons [93]. The high levels of glutamate achieved in the cleft by release of multiple vesicles during a burst may also help to maintain high levels of glutamate in the cleft, thus keeping mGluR6 receptors close to saturation in darkness. This is important for introducing a non-linearity at the synapse whereby only larger hyperpolarizing membrane voltage changes in a rod (that are more likely to reflect genuine responses to photon absorption) will cause a sufficiently large decline in glutamate levels to generate a detectable post-synaptic response [71, 162, 181].

Non-ribbon release in rods

Unlike cones where selectively damaging ribbons abolished both fast and slow components of release equally, damaging rod ribbons by FALI preferentially reduced fast release but left the slower component largely unaffected [44]. The initial fast component of release from rods thus appears to arise from the ribbon and involves similar mechanisms as ribbon-mediated release from cones [126, 233]. The major difference in release from rods and cones is that rods are capable of release at non-ribbon sites, whereas cones are not. In both salamander and mouse rods, non-ribbon release from rods can be triggered by Ca²⁺-induced Ca²⁺ release (CICR) from intracellular endoplasmic reticulum (ER) stores [12, 34, 202]. The Ca²⁺ released from intracellular stores in the rod terminal can help sustain release during lengthy depolarizing stimuli (e.g., in maintained darkness). Immunohistochemical staining for SERCA and labeling with ER tracker dye both showed the presence of ER in rod terminals [12, 45]. Electron micrographs also showed putative ER within 500 nm of ribbons [12] but similar structures have also been identified as endosomes [73]. The ER forms a continuous network throughout the cell, extending from the synaptic terminal to the soma. Imaging of ER Ca²⁺ levels showed that Ca²⁺ could move within the ER between the soma and terminal compartments [45]. Cytoplasmic Ca²⁺ changes that accompany opening of Ca²⁺ channels are typically restricted to regions close to Ca²⁺ channels in the terminal and do not travel up the axon. However, CICR triggered by local Ca²⁺ changes can deplete ER stores within the terminal. These stores can then be refilled by Ca²⁺ diffusing within the ER from other parts of the rod. As Ca²⁺ ions move from the soma to the terminal to restore intraterminal ER Ca²⁺, ER depleted in the soma can itself be refilled by the actions of SERCA and store-operated Ca²⁺ channels [119, 203, 204]. Inhibiting store-operated Ca²⁺ channels therefore has a similar inhibitory effect on release as inhibiting CICR in rods [12, 34, 202, 203]. An interesting question is whether the ability of Ca²⁺ ions to move within the ER also provides a means for intraterminal Ca²⁺ changes to influence activity in the soma and nucleus.

What is the role of CICR in release? Blocking CICR produces changes in baseline currents in bipolar and horizontal cells consistent with a reduction in basal glutamate levels in the synaptic cleft [12, 34, 121, 202, 247]. Unlike salamander cone synapses where the glutamate released from each vesicle acts independently and glutamate is rapidly removed from the cleft, glutamate levels at rod synapses rise and fall more slowly [33]. Elimination of Syt1 from mouse rods blocked fast release of the RRP but slow release evoked by long depolarizing steps, which likely reflects non-ribbon release, persisted [79]. However, the rod-driven ERG b-wave in mouse retina was almost completely abolished by elimination of Syt1 from rods suggesting that this slow, non-ribbon component of release plays at best a minor role in shaping post-synaptic light responses [79]. On the other hand, inhibiting CICR with a high concentration of ryanodine reduced bipolar and horizontal cell light responses and the ERG b-wave by 50% or more in both mouse and salamander retina [12, 34] suggesting a more significant role for non-ribbon release in shaping rod-driven light response.

Although further experiments are needed to reconcile these findings, we favor the idea that ribbon-mediated release is more important for signaling rapid light-evoked changes and non-ribbon release is more important for regulating basal levels of glutamate in the cleft. Changes in basal glutamate levels in the cleft could modulate synaptic transmission by altering the sensitivity of pre-and post-synaptic glutamate receptors. For example, non-ribbon release may help to maintain high levels of cleft glutamate needed for the non-linear thresholding mechanism in rod bipolar cells described above [71, 181]. Changes in basal glutamate could also act on presynaptic autoreceptors that may include both metabotropic and ionotropic glutamate receptors [90, 101, 116, 117]. For example, activation of mGluR8 by cleft glutamate can lead to a reduction in intracellular Ca²⁺ in rods [116, 117], providing a potential negative feedback mechanism for stabilizing Ca²⁺ levels. Resting Ca²⁺ levels in rods and cones at − 70 mV are maintained at low levels much like other neurons (ca. 100 nM) despite weak endogenous Ca²⁺ buffering in salamander rods and cones, equivalent to 50–100 μM EGTA [47, 146, 198, 203, 232].

Spontaneous release

Even at a membrane potential of − 70 mV where Ca²⁺ channels show little or no activity, individual release events can be seen in both rods and cones [52, 79, 92, 139]. Spontaneous release from cones at − 70 mV was not blocked by Co²⁺, Cd²⁺, or by intracellular introduction of the fast Ca²⁺ buffer, BAPTA (10 mM), indicating that it is Ca²⁺-independent [52, 139]. In rods, the frequency of spontaneous events is reduced but not eliminated by Cd²⁺ and BAPTA suggesting the presence of both Ca²⁺-dependent and independent events [52, 94]. Deletion of Syt1 in mouse photoreceptors elevated spontaneous release frequency in rods but not cones [79]. This increase in spontaneous release is consistent with findings in other systems where Syt1 clamps the SNARE apparatus to restrict Ca²⁺-dependent fusion [79]. The absence of an increase in cones may be because spontaneous release in these cells is entirely Ca²⁺-independent. In salamander rods, TIRF imaging experiments showed that spontaneous release involves both ribbon and non-ribbon sites [52]. During hibernation, ground squirrels disassemble the ribbons in their cones [178] and accompanying this change is a reduction in the size and frequency of individual spontaneous release events [144]. This suggests that, like evoked release, spontaneous release from cones involves ribbon release sites.

Rates of spontaneous release in salamander rods and cones at − 70 mV averaged 6 and 10 Hz, respectively [52]. Spontaneous release in mouse rods and cones held at − 70 mV averaged 0.9 and 5.8 Hz, respectively [79, 94]. If we normalize to the number of ribbons, spontaneous release averages ~ 0.9 Hz per ribbon in rods and 0.6–0.8 Hz per ribbon in cones of both mouse and salamander. Assuming the size of the RRP corresponds to the number of release sites along the base of the ribbon, then these numbers suggest that individual sites release vesicles spontaneously at 0.03–0.05 Hz in salamander cones, 0.04 Hz in salamander rods, and 0.01 Hz in mouse rods [19, 76, 233]. These rates are similar to those of hippocampal synapses (~ 0.01 Hz) [4, 156, 179] but tenfold higher than the frequency of ~ 0.002 Hz found at the calyx of Held [189]. It has been suggested that spontaneous release may be a means of providing an occasional “ping” signal to test whether a synaptic connection is still active. However, this function seems unnecessary at ribbon synapses where Ca²⁺-dependent release continues ceaselessly day and night. Perhaps Ca²⁺-independent spontaneous release is simply an unavoidable consequence of a sensitive fusion apparatus constantly poised for release.

Ribbon knockout mice

Two main genetic approaches have been used to examine the role of ribbons in release. In one set of studies, the anchoring protein bassoon was eliminated leading to photoreceptors with detached ribbons that floated within the cytoplasm. In these mice, the ERG b-wave, which reflects ON bipolar cell light responses, was severely reduced in both scotopic and photopic conditions [63]. The a-wave evoked by a flash of light (which reflects photoreceptor responses) was not reduced, showing impaired synaptic transmission from rods and cones. Whole cell recordings from cone-driven horizontal cells in mouse retina showed that after loss of bassoon, tonic glutamate release and post-synaptic light responses were both reduced. The slowing of tonic release, impaired replenishment and reduction in the size of the RRP in bassoon knockout mice [10] are generally consistent with the view of the ribbon obtained from studies in salamander retina.

Targeting the B domain of ribeye is lethal, so ribeye knockout mice were developed by targeting the ribbon-specific A domain [141]. Ribeye knockout mice lack ribbons entirely. These mice showed impaired release from rods and rod bipolar cells and a reduction in the scotopic b-wave, similar to bassoon knockout mice [70, 76, 141]. However, unlike bassoon knockouts, the photopic b-wave reflecting cone output, was unchanged in ribeye KO mice. This was surprising given evidence from other approaches (e.g., FALI) that release from cones occurs exclusively at ribbons. Direct measurements of release from cones lacking ribbons have not yet been reported but presumably the cone is able to release vesicles at non-ribbon sites in the absence of ribbons. Despite substantial deficits in release from rods and rod bipolar cells, recordings from large ON-α ganglion cells that receive strong rod and cone inputs in ribeye KO mice showed surprisingly modest changes in light-evoked spiking, even at scotopic light intensities. Ribeye KO mice showed a decrease in contrast sensitivity, slightly longer latency responses to light increments, and slightly weaker responses to light decrements [161]. While at first glance, these results might suggest that ribbons are largely dispensable for retinal signaling, the persistence of robust light-driven activity may instead reflect the ability of retinal ganglion cells to compensate for diminished excitatory rod and cone input by reducing inhibition in the inner retina [39, 40]. However, the unexpectedly small changes in ribeye knockout mice shows that we still lack a full understanding of the roles played by photoreceptor ribbons in shaping vision.

Conclusions and future directions

Because of the large cell size and robust physiology, salamander retina has served as a valuable preparation for analyzing properties of release from rods and cones. A number of studies have also been done using the cone-dominated ground squirrel retina and recent studies have begun to analyze release from mouse retina. Conclusions about the mechanisms of release employed by rods and cones in mammalian retina are so far generally in agreement with those from amphibian retina suggesting that many of these fundamental mechanisms are highly conserved. This is consistent with the conserved nature of many other aspects of the circuitry in vertebrate retina [248]. The differences that have so far emerged appear to relate to specialized requirements of bipolar cells. For example, AMPA receptors of OFF bipolar cells in salamander do not appear to saturate during release unlike those of ground squirrel [77, 163]. Amphibian retinas lack the specialized rod bipolar cell found in mammals and the input into these cells from mouse rods involves coordinated bursts of vesicle release that are not seen in salamander rods [94].

Many questions still remain concerning rod and cone synapses. For example, along with a general understanding of the role of ribbons in release, what are the specific roles of the enzymatic domains of ribeye? What are the roles played by ribbon-specific variants of proteins such as syntaxin 3B and complexins 3 and 4? How does the presence of an arciform density shape Ca²⁺ nanodomains and the properties of release? How are vesicles primed for release at rod and cone ribbons if they do not use Rim2 and Munc13? What is the molecular make-up of the tethers? What are the mechanisms of fast endocytosis at rods and cones? Do they incorporate kiss-and-run? What role, if any, does spontaneous release play at these synapses? Is the unusual Ca²⁺-sensitivity of amphibian rods and cones retained at mammalian synapses that use Syt1? If so, what accounts for this difference? What is the impact of synchronous and sequential multivesicular release on coding of light responses at rod and cone synapses? What is the role of non-ribbon release in rods? What are the Ca²⁺ sensors that regulate non-ribbon release in rods? Why are rods but not cones capable of non-ribbon release?

In addition to answering these and other questions concerning release at rod and cone synapses, a better understanding of these mechanisms provides opportunities for exploring disease-related changes. In the auditory system, various cochlear synaptopathies have been identified as contributing to hearing loss [130]. Synaptopathies affecting photoreceptors also contribute to vision loss [164]. The best recognized examples are the mutations in photoreceptor Ca²⁺ channels and associated proteins that can cause CSNB [251]. Mutations in other proteins found at photoreceptor synapses including Rim, Tulp1, and dystrophin are also associated with retinal degeneration [18, 42, 84]. In addition, retinal diseases that do not arise from specific mutations in synaptic proteins may also involve synaptic changes in photoreceptors. For example, changes in the ERG suggest impaired transmission from rods is present early in diabetes [165]. And in patients with age-related macular degeneration, photoreceptors show diminished expression of synaptic proteins, loss of ribbons, and altered synaptic structure [109]. Similar synaptic deficits have been seen in mouse models of age-related macular degeneration [254] and choroidal neovascularization [37]. We also observed a decrease in synaptic release from photoreceptors in a model of Stargardt’s disease [25]. Further study of photoreceptor synapses are likely to reveal other synaptopathies that contribute to vision loss.