Journal of Neurophysiology

Release from the cone ribbon synapse under bright light conditions can be controlled by the opening of only a few Ca2+ channels

Theodore M. Bartoletti, Skyler L. Jackman, Norbert Babai, Aaron J. Mercer, Richard H. Kramer, Wallace B. Thoreson

Abstract

Light hyperpolarizes cone photoreceptors, causing synaptic voltage-gated Ca2+ channels to open infrequently. To understand neurotransmission under these conditions, we determined the number of L-type Ca2+ channel openings necessary for vesicle fusion at the cone ribbon synapse. Ca2+ currents (ICa) were activated in voltage-clamped cones, and excitatory postsynaptic currents (EPSCs) were recorded from horizontal cells in the salamander retina slice preparation. Ca2+ channel number and single-channel current amplitude were calculated by mean-variance analysis of ICa. Two different comparisons—one comparing average numbers of release events to average ICa amplitude and the other involving deconvolution of both EPSCs and simultaneously recorded cone ICa—suggested that fewer than three Ca2+ channel openings accompanied fusion of each vesicle at the peak of release during the first few milliseconds of stimulation. Opening fewer Ca2+ channels did not enhance fusion efficiency, suggesting that few unnecessary channel openings occurred during strong depolarization. We simulated release at the cone synapse, using empirically determined synaptic dimensions, vesicle pool size, Ca2+ dependence of release, Ca2+ channel number, and Ca2+ channel properties. The model replicated observations when a barrier was added to slow Ca2+ diffusion. Consistent with the presence of a diffusion barrier, dialyzing cones with diffusible Ca2+ buffers did not affect release efficiency. The tight clustering of Ca2+ channels, along with a high-Ca2+ affinity release mechanism and diffusion barrier, promotes a linear coupling between Ca2+ influx and vesicle fusion. This may improve detection of small light decrements when cones are hyperpolarized by bright light.

  • retina
  • synaptic transmission
  • computer modeling
  • electrophysiology

the exocytosis of synaptic vesicles is triggered by an elevation of intracellular Ca2+. The molecular machinery underlying exocytosis at the photoreceptor ribbon synapse has an unusually high affinity for Ca2+, requiring a Ca2+ concentration in the 10−7–10−6 M range (Duncan et al. 2010; Rieke and Schwartz 1996; Sheng et al. 2007; Thoreson et al. 2004) compared with the 10−5–10−4 M range for most other synapses. (Beutner et al. 2001; Bollmann et al. 2000; Heidelberger et al. 1994; Schneggenburger and Neher 2000) The high Ca2+ sensitivity in photoreceptors has led to the suggestion that the key Ca2+-dependent events controlling release might occur hundreds of nanometers from the mouth of voltage-gated Ca2+ channels, where the local cytoplasmic Ca2+ concentration is maximal. However, anatomical and physiological studies have shown that Ca2+ channels are located very close (<50–100 nm) to releasable vesicles at the base of the synaptic ribbon (Mercer et al. 2011a, 2011b; Morgans 2001; Nachman-Clewner et al. 1999; tom Dieck et al. 2005), consistent with tight functional coupling between the opening of Ca2+ channels and vesicle fusion events.

To better understand how exocytosis is controlled in the photoreceptor terminal, we determined the number of Ca2+ channel openings required to trigger release of a single vesicle. Previous studies at the neuromuscular junction (Shahrezaei et al. 2006), mammalian rod bipolar cell ribbon synapse (Jarsky et al. 2010), and calyceal synapse in the ciliary ganglion (Stanley 1993) indicate that a single Ca2+ channel opening can be sufficient to stimulate release. Several coincident Ca2+ channel openings are needed to stimulate vesicle fusion at the squid giant synapse (Augustine et al. 1991), GABAergic basket cell-granule cell synapse (Bucurenciu et al. 2010), mature calyx of Held (Fedchyshyn and Wang 2005), and hair cell ribbon synapse (Brandt et al. 2005). An even larger number of coincident channel openings (>5) are required for fusion at CA3–CA1 hippocampal synapses (Wheeler et al. 1994), goldfish bipolar cell (Coggins and Zenisek 2009; von Gersdorff et al. 1998), granule cell-Purkinje cell synapses (Mintz et al. 1995), immature calyx of Held (Fedchyshyn and Wang 2005), and adrenal chromaffin cells (Wu et al. 2009). The number of Ca2+ channel openings required for release at a given synapse depends on both the molecular makeup of the release machinery and the architecture of the synapse, including the distance between Ca2+ channels and vesicles.

Photoreceptors have evolved specific mechanisms that allow them to efficiently transmit graded signals about light intensity. For example, release from cones occurs principally at the synaptic ribbon (Snellman et al. 2011). In photoreceptors, the ribbon is a platelike, electron-dense structure that appears to tether the releasable pool of vesicles (Bartoletti et al. 2010; Gray and Pease 1971; Heidelberger et al. 2005; Lasansky 1973; Raviola and Gilula 1975; Usukura and Yamada 1987). The ribbon is anchored to the membrane by a troughlike structure lying at its base known as the arciform density. Ribbons in other sensory neurons, including retinal bipolar cells and hair cells, generally lack an arciform density (reviewed by Heidelberger et al. 2005; LoGuidice and Matthews 2009; Prescott and Zenisek 2005; Schmitz 2009; Sterling and Matthews 2005).

Another unusual property of photoreceptors is the relatively depolarized resting membrane potential in darkness (approximately −40 mV). This depolarized membrane potential keeps a fraction of the voltage-gated Ca2+ channels tonically open and raises the Ca2+ concentration at the base of the ribbon to ∼1 μM (Choi et al. 2008). As a consequence, the readily releasable pool of synaptic vesicles tethered at the bottom of the ribbon is depleted in darkness (Jackman et al. 2009). Under these conditions, the rate of release is limited by the rate of vesicle replenishment and not by the rate of Ca2+-dependent fusion. However, because vesicle replenishment itself is accelerated by raising Ca2+, fluctuations in intracellular Ca2+ levels can still exert control over the release rate in darkness (Babai et al. 2010). The protein that speeds replenishment is unknown, but Ca2+-sensitive sites of replenishment are located >200 nm from the base of the ribbon, too far to sense the “plume” of Ca2+ entering through an individual Ca2+ channel (Babai et al. 2010).

In bright light, cones are hyperpolarized and voltage-gated Ca2+ channels are closed, allowing the releasable pool of vesicles at the base of the ribbon to be replenished. Under these conditions, vesicles anchored at the base of the ribbon close to voltage-gated Ca2+ channels could fuse in response to individual channel openings. To investigate the coupling between Ca2+ channel opening and vesicle fusion at membrane potentials representative of the transition from light to dark, we carried out voltage-clamp experiments and computer simulations. Our results indicate that in bright light there is a close coupling between Ca2+ channel opening and vesicle fusion so that the opening of only two or three Ca2+ channels is needed to trigger release. This relationship appears to be partly ensured by diffusion barriers that restrict the spread of Ca2+ away from open Ca2+ channels. Restricting the spread of Ca2+ prevents promiscuous fusion of multiple vesicles by a single channel opening. The close coupling between Ca2+ channel opening and vesicle fusion in bright light contrasts with release in darkness, which involves synaptic ribbon-mediated replenishment regulated by changes in cytoplasmic Ca2+ far from plasma membrane Ca2+ channels (Babai et al. 2010; Jackman et al. 2009). The tight coupling of Ca2+ channel opening and fusion events in bright light conditions ensures precise timing of release, crucial for synchronizing synaptic signaling to decrements in light intensity.

MATERIALS AND METHODS

Retinal Slice Preparation

Male and female aquatic tiger salamanders (Ambystoma tigrinum, Kons Scientific, Germantown, WI and Charles Sullivan, Nashville, TN) 18–25 cm in length were handled humanely according to protocols approved by the Institutional Animal Care and Use Committee at the University of Nebraska Medical Center. Salamanders were maintained on a 12:12-h day-night cycle and were killed 1–2 h after the beginning of subjective night by decapitation with heavy shears followed by immediate pithing.

Retinal slices were prepared as described previously (Bartoletti et al. 2010; Thoreson et al. 1997). Briefly, retina was isolated under an amphibian saline solution on 5 × 10-mm pieces of nitrocellulose filter paper (type AAWP, 0.8-μm pores; Millipore, Billerica, MA). The retina was then cut into 125-μm-wide slices with a razor blade tissue chopper (Stoelting, Wood Dale, IL). Retinal slices were rotated 90° to see the retinal layers on an upright fixed-stage microscope (Olympus BHWI or Nikon E600FN, Tokyo, Japan) using a water immersion objective [×40, 0.7 numerical aperture (NA) or ×60, 1.0 NA].

Electrophysiology

Cones were voltage clamped simultaneously with adjacent postsynaptic horizontal cells with a Multiclamp 700A amplifier (Molecular Devices, Sunnyvale, CA). Some experiments were performed with an Optopatch (Cairn, Faversham, UK) amplifier for cones and Axopatch 200B (Molecular Devices) for second-order neurons. Cones and horizontal cells were identified by their morphology and response characteristics (Thoreson et al. 1997). Both recording pipettes were positioned with Huxley-Wall micromanipulators (Sutter Instruments, Novato, CA) and visualized through the eyepieces or with a video camera (Watec 502H, Orangeburg, NY) mounted on the microscope. Currents were low-pass filtered at 2 kHz and acquired at 0.1-ms intervals with a Digidata 1322 interface and pCLAMP 9.2 or 10.2 software (Molecular Devices). Acceptable access resistances for voltage clamp recordings were <50 MΩ. Photoreceptors were voltage clamped at −70 mV and horizontal cells at −60 mV.

Patch pipettes were pulled on a PP-830 vertical puller (Narishige USA, East Meadow, NY) from borosilicate glass pipettes (1.2-mm OD, 0.9-mm ID, with internal filament; World Precision Instruments, Sarasota, FL) and had tips of ∼1-μm OD with resistance values of 10–17 MΩ.

For variance-mean analysis of single-channel current amplitudes, we applied 100 test pulses (2 ms) from −70 to +50 mV. Currents were filtered at 5 kHz, and access resistance was compensated 80–90% with the Axopatch 200B amplifier. Retinal slices were superfused with a solution containing (in mM) 111 NaCl, 2.5 KCl, 2 BaCl2, 0.5 MgCl2, 10 N-2-hydroxyethylpiperazine-N′-2-ethanesulfonic acid (HEPES), 5 glucose, and 10 TEACl, with 100 μM niflumic acid and 5 μM BayK8644 (pH 7.8). The pipette solution contained (in mM) 42 CsCl, 48 Cs gluconate, 1.9 MgCl2, 32.9 HEPES, 9.4 TEACl, 9.4 MgATP, 0.5 GTP, and 5 BAPTA (pH 7.2). Most studies find that BayK8644 enhances the open probability of L-type Ca2+ channels with no effect on conductance (Brown et al. 1984; Hess et al. 1984; Kokobun and Reuter 1984; McDonald et al. 1994). Use of short test pulses, niflumic acid, TEA, BAPTA, and Ba2+ minimized activation of Ca2+-activated Cl and K+ currents. Trials were excluded if baseline currents changed significantly during the repetitive application of test pulses. We calculated the variance between subsequent pairs of trials to minimize effects of rundown or potentiation of the current. To measure current amplitude, we subtracted passive and capacitative currents, using a P/200 protocol in which we summed two trials involving 100 tests pulses of 1.2-mV amplitude recorded immediately before and after the test pulse series. The relationship between mean tail current amplitude and intertrial variance was determined at each time point and fit with a parabolic function: V=iII2/N+A(1) In this equation, I = mean current amplitude, i = single-channel current amplitude, N = channel number, A = offset, and V = variance.

For deconvolution experiments, we obtained paired simultaneous whole cell recordings from cones and postsynaptic horizontal cells. The presynaptic recording pipette was filled with a solution containing (in mM) 40 Cs glutamate, 50 Cs gluconate, 9.4 TEACl, 3.5 NaCl, 1 CaCl2, 1 MgCl2, 9.4 MgATP, 0.5 GTP, 5 EGTA, and 10 HEPES (pH 7.2). Postsynaptic pipettes were filled with a solution containing (in mM) 48 Cs gluconate, 42 CsCl, 9.4 TEACl, 1.9 MgCl2, 9.4 MgATP, 0.5 GTP, 5 EGTA, and 32.9 HEPES (pH 7.2). The osmolarity of pipette solutions was adjusted, if necessary, to ∼240 mosM. The liquid junction potential (LJP) was estimated from the calculator in pCLAMP 9.2 to be −13 mV for the cone pipette solution and −16 mV for the horizontal cell pipette solution. Retinal slices were superfused with a solution containing (in mM) 116 NaCl, 2.5 KCl, 1.8 CaCl2, 0.5 MgCl2, 10 HEPES, and 5 glucose (pH 7.8). Use of HEPES as a pH buffer limited effects of proton feedback (DeVries 2001; Hirasawa and Kaneko 2003; Hosoi et al. 2005). The osmolarity measured with a vapor pressure osmometer (Wescor, Logan, UT) was 242 ± 5 mosM. Solutions were bubbled continuously with 100% O2.

Deconvolution was performed with OriginPro 8 software (North Hampton, MA). For deconvolution, we used P/8 leak-subtracted ICa instead of the P/200 leak subtraction used for variance-mean experiments. To deconvolve single-channel events during ICa evoked by a step to −10 mV, we used a single Ca2+ channel probability density function with a peak amplitude of 0.31 pA and a decay time constant of 1.1 ms to represent variable channel open times. The channel open time constant of 1.1 ms was derived from single-channel measurements in photoreceptors (Thoreson et al. 2000) and is similar to open times for L-type Ca2+ channels in many other preparations (Fenwick et al. 1982; Fox et al. 1987; Hagiwara and Ohmori 1983; Lux and Brown 1984; Zampini et al. 2006). Single-channel current amplitude was determined from results of variance-mean experiments described below and is similar to single-channel currents measured directly from individual L-type Ca2+ channels in other preparations with physiological Ca2+ levels (Church and Stanley 1996; Rodriguez-Contreras et al. 2002). Nearly identical results were obtained by deconvolving ICa with a square function with amplitude of 0.31 pA and duration of 1.1 ms (not shown). For steps to −30 mV, we used a mean amplitude of 0.37 pA to account for the greater driving force. For experiments using a ramp voltage protocol (−90 to +60 mV, 0.5 mV/ms), we adjusted single-channel amplitude for driving force throughout the ramp. The number of channel openings was scaled by an average of 13 ribbons per cone (Bartoletti et al. 2010; Pang et al. 2008) to estimate the number of openings per ribbon.

EPSCs with peak amplitude >50 pA were deconvolved with an empirically determined average miniature EPSC (mEPSC) waveform (Bartoletti et al. 2010; Cadetti et al. 2008). The mEPSC waveform averaged from 16 horizontal cells had an amplitude of 5.7 pA and a total charge transfer of 15.5 pC. It had relatively slow kinetics, as expected from the large volume of the photoreceptor synaptic cleft (10–90% rise time of 780 μs and decay time constant of 2.4 ms; Cadetti et al. 2008). A single cone can contact a single horizontal cell at more than one ribbon, and paired cone-horizontal cell recordings showed that each ribbon contributes ∼46 pA to the peak amplitude of the PSC evoked by a step to −10 mV (Bartoletti et al. 2010). The number of ribbon contacts was therefore calculated by dividing the peak amplitude of the EPSC evoked by a step to −10 mV by 46 pA/ribbon (Bartoletti et al. 2010). Release events were shifted forward by 300 μs to compensate for the latency from the rise in Ca2+ concentration to the beginning of the EPSC (Duncan et al. 2010; Lisman et al. 2007). This was the latency to the beginning of a detectable PSC (i.e., the time when the first molecules of glutamate reach postsynaptic glutamate receptors) measured after instantaneous elevation of Ca2+ by flash photolysis of caged Ca2+ at the cone ribbon synapse (Duncan et al. 2010). The additional spread of glutamate through the volume of the synaptic cleft was captured by the mEPSC waveform. Results of deconvolution were low-pass filtered with a fast Fourier transform filter with 5–10 points of smoothing. The rate of Ca2+ channel openings was divided by the rate of release events to calculate the number of vesicle fusion events per channel opening at each time point.

Vesicle release was modeled with Mathematica (Wolfram Research, Champaign, IL). The immediately releasable pool of 20 vesicles was arranged in a square grid with 50-nm center-to-center spacing on the surface of a 250-nm-long ribbon in two rows of 5 vesicles on either side. Fifty-six Ca2+ channels were placed randomly on the plasma membrane evagination, an arc with 100-nm radius centered at the ribbon apex. A simulated step to −10 mV opened Ca2+ channels to a maximal open probability of 0.35, opening in time according to the equation Po = 0.35 × t/(t + 5 ms). Randomly chosen Ca2+ channels opened for 1.1 ms, supporting a local Ca2+ gradient as described by the linear approximation of nanodomains created by a point source of Ca2+ in the presence of a diffusible buffer (Neher 1998). Time-dependent Ca2+ gradients in the presence of a diffusion barrier (Ait-Haddou et al. 2010) were modeled in increments of 100 μs, with release binned into 1.1-ms intervals to reflect the opening of Ca2+ channels. The domains of all open channels were summed linearly. The Ca2+ dependence of release was R = 3,634 [Ca2+]2/(2 μM + [Ca2+])2 (Duncan et al. 2010), except when modeling release with the bipolar cell Ca2+ dependence (Heidelberger et al. 1994). The Kd for BAPTA and EGTA was 170 nM, with Kon for BAPTA of 6 × 108 M/s and Kon for EGTA of 9 × 106 M/s (Burrone et al. 2002). The diffusion coefficient for Ca2+ was 220 μm2/s, and the diffusion coefficient for Ca2+ buffers was 20 μm2/s. Each simulation was repeated for 1,000 trials.

Unless otherwise stated, chemicals were obtained from Sigma-Aldrich (St. Louis, MO). The criterion for statistical significance was chosen to be P < 0.05 and evaluated with GraphPad Prism 4.0. Variability is reported as ±SE.

RESULTS

Variance-Mean Analysis of ICa

Changes in the variance and mean amplitude of ICa can be related to one another by a parabolic function involving single-channel number and amplitude (Eq. 1). We determined trial-to-trial variance and mean amplitude of IBa tail currents recorded in the presence of 5 mM BayK8644. IBa was activated by a series of 100 brief (2 ms) test steps from −70 to +50 mV, and passive membrane properties were subtracted with a P/200 protocol (Fig. 1A). The relationship between mean and intertrial variance at different time points was fit with Eq. 1. In the example shown in Fig. 1B, the best fit parabolic function to the variance-mean relationship indicated that the tail current resulted from 1,164 ± 100 channels with a single-channel current averaging −1.29 ± 0.10 pA. On average, the results showed the presence of 1,034 ± 93 (n = 23) Ca2+ channels per cone with a single-channel current of 0.70 ± 0.08 pA. Given 13 ribbons per salamander cone (Bartoletti et al. 2010; Pang et al. 2008), this suggests up to 80 Ca2+ channels per ribbon if every channel is located at the ribbon. If, as suggested for hair cells (Brandt et al. 2005; Zampini et al. 2010), only 70% of channels are at the ribbon, ∼56 Ca2+ channels will be located beneath each ribbon.

Fig. 1.

Variance-mean analysis of cone IBa tail currents provides an estimate of the number of Ca2+ channels per cone terminal. A: overlay of 100 IBa tail currents evoked by 2-ms test steps in 2 mM Ba2+ and 5 μM BayK8644. Passive membrane properties were subtracted with a P/100 leak subtraction protocol. Inset: overlaid tail currents and region of nonstationary fluctuation analysis. The increase in variance can be seen from the thickening of the family of overlaid traces. B: variance and mean of IBa tail currents were plotted and fit with a parabolic equation (see materials and methods). In this example, the fit yielded a single-channel current (I) of −1.29 pA and 1,164 channels (N).

To estimate the open probability attained in control conditions when ICa was fully activated, IBa was measured with a ramp voltage protocol (−90 to +60 mV, 300 ms) without BayK8644. From the ohmic change in IBa between 0 and +20 mV, we extrapolated a reversal potential of +89 mV. With a reversal potential of +89 mV, the single-channel amplitude of 0.7 pA measured at −70 mV would be predicted to diminish to 0.44 pA at −10 mV. The amplitude of whole cell IBa measured at −10 mV averaged 162.6 ± 6.6 pA (n = 5), suggesting that 370 of the total of 1,034 Ca2+ channels were open when IBa was fully activated, yielding a peak mean open probability of 0.36. This is similar to the value found by single-channel measurements of L-type channels in chick hair cells from the semicircular canal (Zampini et al. 2006).

Application of a depolarizing step to −10 mV was found to stimulate release of the rapidly releasable pool of ∼20 vesicles (range 15–24; Bartoletti et al. 2010) with a time constant of 2.7 ms (Rabl et al. 2005). This suggests that 12.6 (9.5–15) vesicles (63%) fuse within that brief period. Each Ca2+ channel has a 36% chance of opening and remains open for an average of 1.1 ms (Thoreson et al. 2000). If every channel is located at the ribbon, then 68 Ca2+ channel openings should occur at each ribbon within 2.7 ms (0.36 × 77 × 2.7/1.1). This in turn implies ∼5.4 (4.5–7) channel openings per vesicle fusion event. If only 70% of the channels are found at the ribbon, then this suggests that ∼3.8 (3.2–4.9) channel openings per vesicle fusion event should occur during the first 2.7 ms.

For variance-mean experiments, we used Ba2+ as the charge carrier to minimize activation of Ca2+-activated K+ and Cl channels. For deconvolution experiments described in the next section, we used 1.8 mM Ca2+ as the charge carrier. We scaled single-channel Ba2+ currents for these Ca2+ levels by comparing the amplitude of the whole cell current measured with 2 mM Ba2+ as the charge carrier (IBa: 162.6 ± 6.6 pA) to the amplitude measured with 1.8 mM Ca2+ as the charge carrier (ICa: 116 ± 6.5 pA, n = 60). The smaller current amplitude observed with 1.8 mM Ca2+ suggests a single-channel amplitude of 0.31 pA at −10 mV, close to the amplitude of single-channel currents measured with physiological Ca2+ levels in L-type channels from other preparations (Church and Stanley 1996; Gollasch et al. 1992; Rodriguez-Contreras et al. 2002).

Simultaneous Measurements of ICa and Synaptic Release

The comparisons described above suggested that an average of 3–5 Ca2+ channel openings accompanied each vesicle fusion event during the first few milliseconds of release. However, fewer channel openings may be needed to stimulate fusion at the instant of peak release efficiency. We therefore compared the numbers of presynaptic Ca2+ channel openings and postsynaptic vesicle fusion events in individual pairs of cells by deconvolving individual channel openings from cone ICa and individual quantal release events from horizontal cell EPSCs. To stimulate release of the entire rapidly releasable pool, we applied a strong step depolarization to the voltage-clamped cone (−70 mV to −10 mV, 100 ms). For these experiments, cone ICa was leak subtracted with a P/8 protocol. Ca2+ channel opening rates were deconvolved from ICa by assuming a single-channel current amplitude of 0.31 pA and τopen = 1.1 ms (Thoreson et al. 2000) and scaling for 13 ribbons/cone (Bartoletti et al. 2010; Pang et al. 2008) (Fig. 2). Vesicle fusion rates were deconvolved from the simultaneously recorded EPSC with a waveform averaged from mEPSCs recorded in 16 horizontal cells (Cadetti et al. 2008). The number of ribbon contacts at each cone-horizontal cell pair was estimated from the size of the EPSC by assuming that each ribbon contributed 46 ± 23 pA to the peak amplitude of the EPSC (Bartoletti et al. 2010). Finally, the rate of fusion at each ribbon (Fig. 2B) was divided by the rate of Ca2+ channel openings per ribbon (Fig. 2B) to give the number of vesicles per channel opening (0.3 at the peak of release in Fig. 2C). Release and release efficiency quickly rose to a peak and then declined. Peak release efficiency (Fig. 2C) was attained 5.27 ± 0.54 ms (n = 18) after the beginning of the test step, before the actual peak of the PSC (Fig. 2A) that was attained at 9.78 ± 0.64 ms. The peak efficiency averaged 0.33 ± 0.040 (n = 18) vesicle fusion events per channel opening. The reciprocal value of 3.0 provides the number of channel openings per fusion event. If only 70% of the channels reside at the ribbon, then this suggests a peak efficiency of 2.1 channel openings per fusion event. As expected, the peak efficiency exceeded the average efficiency over the first few milliseconds of release.

Fig. 2.

Deconvolution of Ca2+ channel openings and release rates from the presynaptic ICa and excitatory postsynaptic current (EPSC) of a simultaneously voltage-clamped cone and horizontal cell provides an estimate of Ca2+ channel cooperativity. In these experiments, we used 1.8 mM Ca2+ without BayK8644. A: cone ICa overlaid on the EPSC recorded simultaneously from a horizontal cell. B: rate of Ca2+ channel openings per millisecond per ribbon obtained from deconvolution of ICa (see materials and methods). Deconvolution of the EPSC with the average miniature EPSC (mEPSC) waveform gives the release rate per ribbon. In this example, release rate was normalized for 6 ribbon contacts. C: dividing the release rate per ribbon by the rate of Ca2+ channel openings per ribbon gives the number of release events per opening. The reciprocal value provides the number of channel openings per fusion event. In this example, peak efficiency of 0.3 vesicle fusion events per channel opening (3.3 channel openings/fusion event) was reached at 20.9 ms, 4.7 ms after the beginning of the test step.

Application of a depolarizing test step rapidly activated ICa and thus generated fast EPSCs in horizontal cells. We were concerned about the possibility that small kinetic differences between the rise times for ICa and EPSCs might have a significant impact on calculations of release efficiency. To lessen the possibility of kinetic differences, we activated ICa more slowly by using voltage ramps (0.5 mV/ms, −90 mV to +60 mV). Ramp-evoked EPSCs were smaller in amplitude but broader than step-evoked EPSCs. ICa and EPSC from one cell pair are overlaid in Fig. 3A. The number of fusion events per channel opening is shown in Fig. 3B. By examining small ICa evoked during stimulation with voltage ramps, Jarsky et al. (2010) showed that single Ca2+ channel openings were capable of driving fusion. Deconvolution of ramp-evoked responses at the cone synapse sometimes suggested apparent release efficiency exceeding one vesicle per channel opening. However, abrupt increases in the apparent efficiency were also sometimes observed even without stimulation as a result of small baseline fluctuations in the cone membrane current and the occurrence of spontaneous mEPSCs derived from release by neighboring unclamped photoreceptors. Similar abrupt jumps in efficiency were sometimes seen at strongly hyperpolarized potentials during the voltage ramp, but efficiency typically rose steadily with increasing membrane depolarization during the ramp (Fig. 3). The peak efficiency was attained when the cone membrane potential was between −30 and −40 mV (uncorrected for LJP), well below the peak of ICa. The peak number of openings per fusion event averaged 0.44 ± 0.064 (n = 9) during a ramp voltage protocol. This was only slightly higher than ratios found with steps to −10 mV (P = 0.16, unpaired t-test). Thus differences in the activation kinetics of ICa and EPSCs did not appear to significantly influence the conclusion that ∼2 channel openings are needed to initiate fusion of each vesicle.

Fig. 3.

Efficiency of release evoked by activation of ICa with a ramp voltage protocol. A: overlay of cone ICa evoked by a ramp voltage protocol (−90 to + 60 mV, 0.5 mV/ms) with the EPSC recorded from a simultaneously voltage-clamped horizontal cell. B: deconvolution of ICa and the EPSC showed a peak cooperativity of 0.56 release events per opening (1.8 openings/simultaneous fusion event). Traces are plotted as a function of voltage applied during the ramp protocol (Vm).

We examined the possibility that activation of ICa with a strong test step may have stimulated unnecessary and redundant Ca2+ channel openings. We therefore compared the number of Ca2+ channel openings and vesicle release events during submaximal activation of ICa, using depolarizing steps to −30 mV rather than −10 mV. We scaled for the number of ribbon contacts, using the size of the EPSC evoked by steps to −10 mV. To account for the change in driving force with steps to −30 mV, we used the single-channel current amplitude of 0.37 pA for deconvolution of ICa. In the example shown in Fig. 4, the peak rate attained was 0.43 vesicles per Ca2+ channel opening. On average, steps to −30 mV stimulated 0.40 ± 0.06 (n = 8) vesicle fusion events per Ca2+ channel opening. This was not significantly different from the peak efficiency attained with ramp protocols (P = 0.66, unpaired t-test) or with steps to −10 mV (P = 0.39), suggesting that there are few unnecessary channel openings.

Fig. 4.

Efficiency of release evoked by activation of ICa with a step to −30 mV. A: overlay of cone ICa with the EPSC recorded simultaneously from a horizontal cell during a step depolarization from −70 mV to −30 mV (100 ms). Vh, holding potential. B: deconvolution of ICa and the EPSC showed a peak efficiency of 0.43 release events per opening (or 2.3 channel openings/fusion event).

We tested the effects of lowering Ca2+ channel open probability by inhibition of ICa with nifedipine (Brown et al. 1986; Hess et al. 1984; Worley and Kotlikoff 1990). In the presence of a concentration of nifedipine that partially blocked ICa (3 μM), steps to −10 mV evoked release with a peak efficiency of 0.19 ± 0.038 (N = 15) fusion events per channel opening, significantly lower than control release efficiency (P = 0.003; unpaired t-test). The inhibition of L-type Ca2+ channels by nifedipine can be rapidly unblocked with a bright ultraviolet (UV) light flash (Sanguinetti and Kass 1984). At the midpoint of a test step to −10 mV, we applied a bright UV flash to abruptly reduce nifedipine inhibition and thereby increase the number of open Ca2+ channels. Figure 5A shows the cone ICa and horizontal cell EPSC from one pair. Release efficiency immediately following the step is shown in Fig. 5B. The change in release efficiency following the UV flash is shown in Fig. 5C. ICa increased from 90.7 ± 9.3 pA to 120.0 ± 11.2 pA (n = 15) following the flash. The additional release stimulated by the increase in ICa exhibited an efficiency of 0.36 ± 0.056 (n = 15) vesicle fusion events per channel opening, significantly greater than release efficiency measured prior to the flash (P = 0.022, paired t-test). Application of a UV flash without nifedipine did not increase release (n = 3). If vesicle release could be consistently triggered by the opening of only a single Ca2+ channel, then reducing the probability of channel opening with nifedipine should cause a linearly proportional reduction in the number of vesicles released. The finding that release efficiency increased slightly with Ca2+ channel open probability suggests a weak cooperativity among Ca2+ channels during release. These data also provide further evidence that there are few unnecessary channel openings when the releasable pool is full.

Fig. 5.

Effects of changing Ca2+ channel open probability with 3 μM nifedipine. A: cone ICa overlaid on the simultaneously recorded horizontal cell EPSC evoked by a step from −70 mV to −10 mV (100 ms) applied to the cone in the presence of 3 μM nifedipine. A bright UV flash was applied 100 ms into the step to reduce nifedipine antagonism of Ca2+ channels. B: release events per channel opening deconvolved from the EPSC and ICa at the beginning of the test step. C: release events per channel opening deconvolved from the EPSC and ICa after reduction of nifedipine antagonism with the UV flash.

Because the spread of Ca2+ from open Ca2+ channels to release sites influences the efficiency of release, we examined the role of Ca2+ buffers in shaping release. To do so, we compared efficiencies obtained with 5 mM EGTA, 0.5 mM EGTA, and 1 mM BAPTA. Reducing Ca2+ buffering from 5 mM to 0.5 mM EGTA would be expected to allow Ca2+ to spread further from the mouth of an open channel and thereby increase release efficiency. However, as illustrated in Fig. 6, A and B, use of 0.5 mM EGTA did not significantly change release efficiency (0.30 ± 0.069 vesicle fusion events/channel opening, n = 6) compared with 5 mM EGTA (P = 0.67, unpaired t-test). BAPTA has a much faster ON rate than EGTA and therefore constrains the spread of Ca2+ to more narrow nanodomains. However, release efficiency was also unchanged by use of 1 mM BAPTA as the Ca2+ chelator (0.39 ± 0.066 vesicle fusion events/channel opening, n = 6; P = 0.50, unpaired t-test compared with 5 mM EGTA; Fig. 6, C and D). The latency from the beginning of the test step to the instant of peak efficiency also did not differ significantly in recordings obtained with 0.5 mM EGTA (6.30 ± 0.67 ms), 5 mM EGTA (5.27 ± 0.54 ms), or 1 mM BAPTA (5.77 ± 0.49 ms; P = 0.55, 1-way ANOVA). We assessed release efficiency in the presence of endogenous Ca2+ buffers by obtaining gramicidin-perforated patch recordings from cones (Thoreson and Bryson 2004). Release efficiency with intact endogenous buffering in cones (0.32 ± 0.050 vesicle fusion events/channel opening, n = 5; Fig. 6, E–G) was nearly identical to that observed with 5 mM EGTA (P = 0.86, unpaired t-test). Release efficiency measured with ramp voltage protocols was also not changed significantly by use of perforated patch techniques or different buffers (Fig. 6H). Ca2+ imaging experiments have established that chelators introduced through the patch pipette can successfully reach the synaptic terminal (Mercer et al. 2011a). We therefore interpret the resistance of synaptic release to effects of exogenous buffers as evidence that there may be diffusion barriers that limit access of these buffers to the base of the cone ribbon.

Fig. 6.

Effects of Ca2+ buffering on release efficiency. A, C, and E: cone ICa evoked by a step from −70 mV to −10 mV (100 ms) overlaid on the simultaneously recorded horizontal cell EPSC. B, D, and F: number of vesicle fusion events per channel opening obtained by deconvolution of the EPSC and ICa, respectively. In A and B, the cone pipette solution contained 0.5 mM EGTA. In C and D, the cone pipette solution contained 1 mM BAPTA. In E and F, we used a gramicidin-perforated patch recording technique to maintain endogenous Ca2+ buffering. G: average peak efficiency of release obtained when using a step depolarization to −10 mV and cone Ca2+ buffering provided by 5 mM EGTA (n = 18), 0.5 mM EGTA (n = 6), 1 mM BAPTA (n = 6), or endogenous Ca2+ buffers (perforated patch, n = 5). H: average peak efficiency of release with these same buffers obtained with a ramp voltage protocol (0.5 mV/ms; 5 mM EGTA, n = 9; 0.5 mM EGTA, n = 6; 1 mM BAPTA, n = 6; perforated patch, n = 5).

Simulations of Release at the Cone Synapse

To assess our understanding of the synapse, we simulated release at the cone ribbon synapse with a model that employed empirically determined values for Ca2+ channel number (present study), Ca2+ channel current amplitude (present study), Ca2+ channel mean open time (Thoreson et al. 2000), synaptic dimensions (Lasansky 1973; Pang et al. 2008; Raviola and Gilula 1975), vesicle pool size (Bartoletti et al. 2010), and Ca2+ dependence of release (Duncan et al. 2010). We simulated a step to −10 mV by assuming the stochastic opening of Ca2+ channels with a probability of 0.35. Each channel opened for 1.1 ms, establishing a local Ca2+ nanodomain with the gradient shaped by a diffusible buffer (Neher 1998). Overlapping domains of neighboring channels were summed linearly. As shown in Fig. 7, simulations with this model using 5 mM EGTA as the diffusible buffer predicted a transient burst of release that achieved a maximum release efficiency of 3.7 vesicle fusion events per channel opening. This is ∼10-fold higher than the experimentally observed peak efficiency of 0.32 fusion events per opening (or 0.4 vesicles/channel if only 70% of channels are located at the ribbon). Reducing EGTA to 0.5 mM produced a transient burst of release with an even greater peak release efficiency of 5.3 vesicle fusion events per channel opening (not shown). Simulations with 1 mM BAPTA produced a peak efficiency of 0.13 fusion events per channel opening (Fig. 7D) that was lower than the value observed in actual recordings. In addition, simulations with BAPTA predicted a relatively slow release profile, with peak efficiency attained 3.3 ms later than the peak predicted for 5 mM EGTA. By contrast, the time to peak efficiency in actual recordings using 1 mM BAPTA or 5 mM EGTA did not differ significantly (unpaired t-test, P = 0.47).

Fig. 7.

Simulations of release at the cone ribbon synapse. A: diagram of the geometric arrangements used for the original model without a diffusion barrier. B: release rates (dashed line) and Ca2+ channel openings (solid trace) per ribbon predicted by the model using 5 mM EGTA as the intracellular Ca2+ buffer in the cone terminal. C: release rates (dashed line) and Ca2+ channel openings (solid trace) per ribbon predicted by the model using 1 mM BAPTA as the intracellular Ca2+ buffer. D: release events per Ca2+ channel opening (5 mM EGTA, 1 mM BAPTA).

Simulations predicted peak release efficiency with EGTA higher than the measured efficiency and peak efficiency with BAPTA lower than the measured efficiency. We examined various elements of the model to see which might account for these differences.

Properties of the exocytotic calcium sensor.

Cones utilize a Ca2+ sensor with an unusually low cooperativity (n = 2) and high Ca2+ sensitivity (release threshold of ∼400 nM; Duncan et al. 2010) compared with other CNS neurons. To test the impact of sensor properties, we simulated release by using properties of the more conventional goldfish Mb1 bipolar cell Ca2+ sensor, which exhibits 5 Ca2+ binding sites and a threshold of ∼20 μM (Heidelberger et al. 1994). With this sensor, maximal release efficiency declined to 0.02 vesicle fusion events per channel opening with 5 mM EGTA and <0.0002 fusion events per opening with 1 mM BAPTA (Fig. 8, A–C). Thus the high affinity of the photoreceptor Ca2+ sensor appears to be important for maintaining high release efficiency at the cone synapse.

Fig. 8.

Effects of changing model parameters on simulations of release. In A–C, we used the exocytotic Ca2+ sensor properties found for Mb1 goldfish bipolar cells by Heidelberger et al. (1994). In D–F, we used Ca2+ channel properties (maximum open probability = 0.015 and single-channel current = 0.6 pA) reported for heterologously expressed CaV1.4 channels by Doering et al. (2005). In G–I, we simulated a 10-fold increase in channel number by lowering Ca2+ channel open probability to 0.035. In J–L, we increased the distance between Ca2+ channels and release sites by 50 nm. For all 4 of these manipulations, we show the predicted release rates (dashed lines) and Ca2+ channel openings (solid traces) per ribbon when using 5 mM EGTA (A, D, G, and J) or 1 mM BAPTA (B, E, H, and K). We also illustrate predicted release events per Ca2+ channel opening with both buffers in C, F, I, and L (5 mM EGTA, 1 mM BAPTA).

Combining a bipolar cell Ca2+ sensor with photoreceptor architecture, the model predicted lower release efficiency than has been observed empirically in bipolar cells (Coggins and Zenisek 2009; Jarsky et al. 2010; von Gersdorff et al. 1998). As a test of the model, we simulated release from bipolar cells by combining bipolar cell Ca2+ dependence with bipolar cell-like architecture. Because bipolar cells typically lack an arciform density (Raviola and Raviola 1982), we modeled release at this synapse by placing Ca2+ channels along a 20-nm strip immediately adjacent to release sites. The combination of this architecture with bipolar cell Ca2+ dependence yielded a peak efficiency of 0.44 fusion events per channel opening with 5 mM EGTA and 0.27 fusion events per opening with 1 mM BAPTA (not shown). This is similar to results obtained in goldfish retinal bipolar cells by von Gersdorff et al. (1998; 0.14–0.27 fusion events/opening with 0.2 mM BAPTA). These results suggest that Ca2+ sensor properties and architecture of the synapse both help to shape release at ribbon synapses.

Single Ca2+ channel properties.

Both variance-mean analysis (present study) and single-channel recordings (Thoreson et al. 2000) indicate that the properties of photoreceptor Ca2+ channels are similar to those of other L-type channels (Church and Stanley 1996; Fenwick et al. 1982; Fox et al. 1987; Hagiwara and Ohmori 1983; Lux and Brown 1984; Rodriguez-Contreras et al. 2002; Zampini et al. 2006). However, recordings of heterologously expressed CaV1.4 channels showed 5-fold lower conductance and >10-fold lower open probability (Po < 0.015; Doering et al. 2005). We simulated release with the use of these latter channel properties and obtained a peak release efficiency of 1.04 fusion events per channel opening with 5 mM EGTA and 0.26 fusion events per opening with 1 mM BAPTA (Fig. 8, D–F). While the efficiency predicted for BAPTA was similar to the peak efficiency observed empirically, the kinetics of release predicted by the model differed considerably from the kinetics observed in real recordings: the model predicted that release with 1 mM BAPTA was maintained at a nearly steady rate throughout the depolarizing step (Fig. 8E), whereas actual recordings showed an initial phasic burst of release followed by only a small amount of tonic release (Fig. 6C).

Ca2+ channel number or open probability.

We tested the impact of increasing the number of Ca2+ channels at the synapse 10-fold. This is equivalent to reducing peak open probability 10-fold to 0.035. With 5 mM EGTA, simulations using a maximal Po of 0.035 yielded a peak release efficiency of 4.7 fusion events per channel opening, far above observed values (Fig. 8, G–I). With 1 mM BAPTA, efficiency declined to 0.24 fusion events per opening, but as described above, the model predicted continuous sustained release (Fig. 8H) rather than the transient release kinetics observed in actual recordings.

Proximity of Ca2+ channels to release sites.

Physiological experiments have shown that Ca2+ channels are located within 50–100 nm of release sites (Mercer et al. 2011a). However, in principle, release efficiency could be reduced by placing channels further away from release sites. In model 1, channels were placed 25–100 nm from release sites. Expanding the active zone by an additional 50 nm so that channels could be as far as 150 nm from release sites reduced peak release efficiency to 2.2 fusion events per channel opening with 5 mM EGTA and 0.018 vesicles per channel opening with 1 mM BAPTA (Fig. 8L). Increasing the size of the active zone even further to 500 nm yielded a peak efficiency of 0.018 fusion events per channel opening with both 5 mM EGTA and 1 mM BAPTA. The release profiles with BAPTA or an expanded active zone were also much more sustained than that of actual recordings (Fig. 8K).

Ca2+ diffusion kinetics.

As described above, substituting alternative parameters for the empirically determined values used in the model tended to worsen the agreement between actual and predicted results. We therefore examined a key assumption of the model: that Ca2+ diffuses freely away from open channels, so that Ca2+ domains develop and collapse instantaneously with the opening and closing of Ca2+ channels (Simon and Llinás 1985). However, our experiments showed that introducing different exogenous diffusible buffers did not affect release (Fig. 6), suggesting that there may be a diffusion barrier that restricts access of buffers to release sites. Such a barrier could also slow the diffusion of Ca2+ away from channels and thereby slow the rise in intracellular [Ca2+]. The cone ribbon synapse possesses an arciform density, a dense proteinaceous structure situated between vesicles and Ca2+ channels, which provides a plausible anatomical basis for such a barrier. To test the impact of diffusion barriers on the spread of Ca2+ at the cone ribbon synapse, we developed a kinetic model that incorporated analytical, time-dependent solutions for the diffusion of Ca2+ away from open Ca2+ channels (Ait-Haddou et al. 2010). To simulate a diffusion barrier, we slowed the rate of Ca2+ diffusion and limited the concentration of diffusible buffer by the same factor. A diffusion barrier is mathematically equivalent to fixed endogenous buffers that act to lower the effective diffusion constant of Ca2+ until the fixed buffer is saturated. Slowing the rate of diffusion by a factor of 200 reduced peak release efficiency with 5 mM EGTA to 0.28 fusion events per channel opening and increased efficiency with 1 mM BAPTA to 0.43 fusion events per opening, close to values obtained from deconvolution experiments (Fig. 9, C and D). The model also predicted that the release kinetics and time to peak efficiency are similar for both 5 mM EGTA and 1 mM BAPTA (Fig. 9, A–C), consistent with results of paired recordings. These results suggest that constraints on the spread of Ca2+ (e.g., from a diffusion barrier) may be important for shaping release efficiency at the cone synapse.

Fig. 9.

Simulations of release in the presence of a diffusion barrier. A: diagram illustrating the cone ribbon synapse geometry with inclusion of a diffusion barrier that slowed diffusion (1/200). B: release rates (dashed lines) and Ca2+ channel openings (solid trace) per ribbon predicted by the model using 5 mM EGTA as the intracellular Ca2+ buffer in the cone terminal. C: release rates (dashed lines) and Ca2+ channel openings (solid trace) per ribbon predicted using 1 mM BAPTA as the intracellular Ca2+ buffer. D: release events per Ca2+ channel opening (5 mM EGTA, 1 mM BAPTA).

DISCUSSION

How Many Ca2+ Channels Are Present at Each Ribbon?

Variance-mean analysis of ICa tail currents indicated that there are ∼1,000 Ca2+ channels per salamander cone. Ca2+ channels are clustered beneath synaptic ribbons in photoreceptors (Morgans 2001; Morgans et al. 2005; Nachman-Clewner et al. 1999; Specht et al. 2009; Steele et al. 2005; tom Dieck et al. 2005; Xu and Slaughter 2005). Ca2+ imaging studies suggest as few as 70% of photoreceptor Ca2+ channels may be in the terminal (Steele et al. 2005; Szikra and Krizaj 2006). Excision of the rod terminal reduced rod ICa by 95%, suggesting that almost all of the Ca2+ channels are located in the terminal (Xu and Slaughter 2005) but not every channel in the terminal may be located at a ribbon. In adult hair cells ∼70% of Ca2+ channels appear to be localized with ribbons (Brandt et al. 2005; Zampini et al. 2010), and in immature hair cells only 27% of CaV1.3 channels localize with synaptic ribbons (Zampini et al. 2010). If every Ca2+ channel is located beneath a synaptic ribbon, then the average of 13 ribbons per cone (Bartoletti et al. 2010; Pang et al. 2008) suggests that there are ∼80 Ca2+ channels per ribbon. If only 70% of Ca2+ channels are located at the cone ribbon, then there may be as few as ∼56 channels/ribbon.

Freeze-fracture electron micrographs of macaque cone pedicles show ∼400 particles along the synaptic ridge that are thought to be Ca2+ channels (Raviola and Gilula 1975). Macaque cone ribbons are 700–1,000 nm long, whereas salamander cone ribbons are 150–350 nm long (Pang et al. 2008; Raviola and Gilula 1975). Scaling for the smaller size of salamander cone ribbons yields 80–140 particles per ribbon, somewhat greater than the number of Ca2+ channels suggested from variance-mean measurements.

We calculated a peak mean open probability for photoreceptor Ca2+ channels of 0.36. This is similar to values obtained from single-channel recordings in hair cells (Zampini et al. 2010) but higher than the value of 0.1 estimated from single-channel recordings in rods (Thoreson et al. 2000). However, measurements of maximal open probability by Thoreson et al. (2000) were limited by the small size of single-channel currents at strongly depolarized membrane potentials. The value of 0.36 is much higher than the mean open probability found for heterologously expressed CaV1.4 channels (∼0.015; Doering et al. 2005). However, the small single-channel open probability and conductance values found in this expression system predict an unrealistically large number of Ca2+ channels (∼6,700 channels/ribbon), suggesting that these properties may not be retained in situ.

How Many Ca2+ Channel Openings Are Associated with Release of Each Vesicle?

Measurements of the size of the immediately releasable pool, release kinetics, and Ca2+ channel properties suggested that an average of 3–5 Ca2+ channel openings are associated with fusion of each vesicle during the first few milliseconds of release. However, the peak efficiency is likely to be higher than this average value, so we also determined release efficiency by deconvolving individual Ca2+ channel openings and quantal synaptic currents from simultaneously recorded cone ICa and horizontal cell EPSCs. The only shared assumption between deconvolution analysis and comparisons between the average pool size and number of channels was the Ca2+ channel mean open time of 1.1 ms (Thoreson et al. 2000), a value that is consistent with single-channel measurements from L-type channels in many preparations (Fenwick et al. 1982; Fox et al. 1987; Hagiwara and Ohmori 1983; Lux and Brown 1984; Zampini et al. 2006). Deconvolution analysis of paired simultaneous recordings showed that at the peak of release ∼3 Ca2+ channel openings accompanied fusion of each vesicle. If only 70% of the channels are located at ribbons, then an average of ∼2 channel openings per ribbon may be sufficient to stimulate fusion of a single vesicle. Thus two different analytical approaches utilizing largely independent measurements suggest that only a small number of Ca2+ channel openings are capable of driving fusion of an individual vesicle. These results are consistent with those in hair cells and bipolar cells showing that a very small number of channel openings are necessary for fusion of each vesicle (Brandt et al. 2005; Jarsky et al. 2010).

Using measurements of channel properties, vesicle pools, calcium dependence of release, and active zone architecture, we developed a biophysically realistic simulation of release at the cone photoreceptor ribbon. The original version of our model assumed that Ca2+ gradients developed instantaneously upon Ca2+ channel opening and collapsed instantaneously after channel closure. However, that simulation predicted that with EGTA as the Ca2+ buffer Ca2+ entering through an individual channel should diffuse far enough beneath the arciform density to stimulate fusion of multiple vesicles. This prediction is a consequence of the high sensitivity of the photoreceptor release mechanism to Ca2+ (Duncan et al. 2010; Rieke and Schwartz 1996; Thoreson et al. 2004). The original version of the model also predicted that with BAPTA as the Ca2+ buffer Ca2+ would not spread far enough from individual channels to reliably stimulate vesicle fusion. By contrast with these predictions, our electrophysiological recordings showed that release efficiency was not changed by use of different exogenous Ca2+ buffers in the cone patch pipette. The insensitivity of release to changes in Ca2+ buffering (Fig. 6) suggested the possibility of a diffusion barrier at the cone synapse that limits access of diffusible buffers to release sites and may therefore also limit the spread of Ca2+ away from open Ca2+ channels. To examine the impact of a diffusion barrier, we modified our model to incorporate a time-dependent rise and fall of Ca2+ gradients (Ait-Haddou et al. 2010). We simulated a diffusion barrier by slowing the rate of Ca2+ diffusion away from open Ca2+ channels by 1/200. With this slower rate of diffusion, the model predicted release efficiencies with both BAPTA and EGTA of 2.3–3.6 channel openings per vesicle fusion event, similar to the efficiency observed empirically. Although our results do not allow us to identify the site of this diffusion barrier, one appealing possibility is the arciform density that lies just beneath the ribbon. Interestingly, this structure is absent from ribbon synapses in retinal bipolar cells as well as hair cells. Changes in the rate of diffusion improved the model's predictions, but changes in other model parameters worsened the agreement between predicted and actual results, suggesting that channel properties, properties of the release mechanism, and architecture of the synapse are all important for establishing efficient release at the cone ribbon synapse.

Although our results suggest an average peak efficiency of release of 2–3 Ca2+ channel openings per vesicle fusion event at each ribbon, it is possible that the opening of only a single channel may be sufficient to drive release. A depolarizing test step to −10 mV stimulates release of the entire immediately releasable pool of ∼20 vesicles per ribbon (Bartoletti et al. 2010). However, the findings that there are ∼56 Ca2+ channels at each ribbon that can open with a maximum probability of 0.36 suggest that a step to −10 mV stimulates the opening of only ∼20 channels per ribbon at any instant or 1 channel per vesicle in the immediately releasable pool. Assuming that the immediately releasable pool represents docked vesicles, not every channel may always be close enough to a docked vesicle to trigger release (Shahrezaei et al. 2006). The presence of channel openings that contribute little or no Ca2+ to release would reduce the average peak efficiency. For example, the small Ca2+ nanodomains predicted by the kinetic model suggest that the opening of a single channel located at the very edge of the active zone may not elevate Ca2+ sufficiently to stimulate vesicle release. Lower efficiency could also result from partial depletion of the ribbon allowing some Ca2+ channels to open beneath empty release sites.

In calculations of release efficiency, we did not include a delay for Ca2+ to diffuse from the open channel to the Ca2+ sensor. With a distance (x) of 100 nm and diffusion coefficient (D) of 0.22 μm2/ms, the relationship x2 = 2Dt predicts a diffusional delay of only 23 μs. However, slowing diffusion (e.g., with a barrier) might increase this delay. An additional diffusional delay would mean that the true efficiency is higher than the efficiency estimated from a step depolarization, since vesicle fusion was actually triggered by Ca2+ channel openings that occurred earlier when ICa was just beginning to be activated. However, if the true efficiency was substantially higher, then this should have been revealed by use of the ramp voltage protocol, which causes slower changes in ICa than those produced by a step protocol. But the ramp produced only a slightly higher efficiency (0.44 vesicles/channel opening).

The finding that release efficiency improved slightly when open channel probability was increased suggests that although the opening of one channel may sometimes be sufficient to stimulate release, the simultaneous opening of more than one Ca2+ channel is needed for consistent vesicle fusion. While some channels may be close enough to a docked vesicle to trigger release, other release events may require the opening of two or more adjacent channels, each contributing a portion of the Ca2+ needed to stimulate release. This appears similar to the situation at rod bipolar cell synapses, where single channel openings can trigger vesicle fusion but the average peak release efficiency is closer to two channel openings per vesicle fusion event (Jarsky et al. 2010).

Release Modes Differ in Light and Dark

In the present study, cones were voltage-clamped at a potential of −70 mV between test pulses. This is close to the membrane potential evoked in a cone by a saturating bright light. When the cone is strongly hyperpolarized, the releasable pool of vesicles tethered at the base of the ribbon is likely to be fully replenished (Jackman et al. 2009) and Ca2+ channel openings are tightly synchronized to depolarizing membrane potential changes. The ability of only 1 or 2 Ca2+ channel openings to trigger vesicle fusion can thus improve the ability of the cone synapse to signal the occurrence of a depolarizing decremental light stimulus with high precision.

Release from bipolar cells has also been reported to show a high efficiency (Jarsky et al. 2010; von Gersdorff et al. 1998). Although mouse rod bipolar cells may have a high-affinity Ca2+ sensor (Jarsky et al. 2010), goldfish Mb1 bipolar cells use an exocytotic Ca2+ sensor with a much lower affinity for Ca2+ (Heidelberger et al. 1994). Using the bipolar cell Ca2+ sensor and bipolar cell-like architecture without a diffusion barrier, the model predicted high release efficiency at the bipolar cell synapse. Why, then, do cones employ an unconventional high-affinity Ca2+ sensor? One answer may be that use of a high-affinity sensor promotes depletion of the releasable pool when a cone remains depolarized in continued darkness (Jackman et al. 2009). The present results indicate that the efficiency of release increases quickly when depolarization is initiated but then diminishes to very low levels as the releasable pool of vesicles is depleted. This low efficiency reflects the fact that, after depletion of the releasable pool, the sustained rate of vesicle release is not determined by individual Ca2+ channel openings but by the rate at which the releasable pool can be replenished (Babai et al. 2010; Jackman et al. 2009). This replenishment process has been shown to be Ca2+ dependent, and thus changes in intracellular Ca2+ levels driven by changes in cone membrane potential control the rate of release (Babai et al. 2010). However, the Ca2+-dependent sites involved in replenishment are further from Ca2+ channels (>200 nm) than are release sites (<50–100 nm). This links the sustained rate of release to average intraterminal Ca2+ levels rather than individual channel openings. This, in turn, may help to make sustained release in darkness less noisy by reducing synaptic noise introduced by the stochastic probability of individual channel opening.

The clustering of Ca2+ channels beneath the ribbon, the presence of diffusion barriers that limit the spread of Ca2+ to localized nanodomains, and the use of a high-Ca2+ affinity release mechanism at the photoreceptor synapse allow the cone synapse to shift between two signaling modes as the cone moves from bright light to darkness. In bright light, cones are hyperpolarized and synaptic output is tightly coupled to the opening of individual Ca2+ channels stimulated by decremental light stimuli. This enhances information transmission by improving the timing precision of release events and reducing the occurrence of redundant events. In darkness, cones are depolarized and release is governed by replenishment rather than individual channel openings. This enhances information transmission by reducing synaptic noise associated with the stochastic opening of individual Ca2+ channels in darkness.

GRANTS

This work was supported by National Eye Institute Grants EY-10542 (W. B. Thoreson) and EY-15514 (R. H. Kramer), Research to Prevent Blindness, and University of Nebraska Medical Center Graduate Student Research Fellowships to T. M. Bartoletti and A. J. Mercer.

DISCLOSURES

No conflicts of interest, financial or otherwise, are declared by the author(s).

AUTHOR CONTRIBUTIONS

Author contributions: T.M.B., S.L.J., R.H.K., and W.B.T. conception and design of research; T.M.B., N.B., A.J.M., and W.B.T. performed experiments; T.M.B., S.L.J., N.B., and W.B.T. analyzed data; T.M.B., S.L.J., and W.B.T. interpreted results of experiments; T.M.B., S.L.J., N.B., and W.B.T. prepared figures; T.M.B. drafted the manuscript; T.M.B., S.L.J., N.B., A.J.M., R.H.K., and W.B.T. edited and revised the manuscript; T.M.B., S.L.J., N.B., A.J.M., R.H.K., and W.B.T. approved the final version of the manuscript.

REFERENCES

View Abstract