Resting and stimulated mouse rod photoreceptors show distinct patterns of vesicle release at ribbon synapses

The exquisite visual sensitivity of vertebrates originates with the ability of rod photoreceptor cells in the retina to detect single photons. In darkness, rods have a resting membrane potential near −40 mV, sufficiently depolarized to activate Ca_V α subunit 1.4 (Ca_V1.4) L-type Ca²⁺ channels that control synaptic glutamate release (Pangrsic et al., 2018). Unlike most voltage-gated Ca²⁺ channels, Ca_V1.4 channels show minimal voltage- and Ca²⁺-dependent inactivation, allowing continued Ca²⁺ influx into rod terminals during maintained darkness (Baumann et al., 2004; Koschak et al., 2003; McRory et al., 2004; Waldner et al., 2018). The maintained release of glutamate from rods is also facilitated by the presence of a platelike presynaptic structure known as the synaptic ribbon (Heidelberger et al., 2005; Schmitz, 2009). Ribbons help to capture and tether vesicles along their planar surfaces, delivering them to release sites at the base (Snellman et al., 2011; Vaithianathan et al., 2016; Wen et al., 2018). In amphibian retina, it has been shown that synaptic release from rods varies linearly with Ca²⁺ influx, translating light-evoked voltage responses into changes in the rate of glutamate release that act on second-order neurons (Rieke and Schwartz, 1996; Thoreson et al., 2004). Release from mammalian rods is regulated by the exocytotic Ca²⁺ sensor molecule synaptotagmin 1 (Syt1), although there is also evidence for a second sensor (Grassmeyer et al., 2019). Syt1 is absent from amphibian rods (Heidelberger et al., 2003), suggesting that properties of release may differ among vertebrate genera.

Absorption of a single photon by rods produces voltage changes of only a few millivolts (Cangiano et al., 2012; Schneeweis and Schnapf, 1995). Second-order rod bipolar cells are faced with the challenge of distinguishing genuine light-evoked changes in release rate produced by these small voltage changes from random intervals in release. One strategy for overcoming this problem is to maintain a high rate of glutamate release so that small changes in membrane voltage cause a large enough reduction in release to be distinguished from basal rates (Rao-Mirotznik et al., 1998; Rao et al., 1994; van Rossum and Smith, 1998). Modeling studies suggest that success with this approach requires release rates of ≥100 vesicles/s/ribbon, but imaging studies with the activity-dependent dye FM1-43, suggest that release rates in gecko and salamander rods are only 2.5 and 18 vesicles/s/ribbon, respectively (Sheng et al., 2007). Release rates have not been measured in mammalian rods.

In this study, we measured voltage-dependent changes in rates of release from individual mouse rods that each possess a single ribbon synapse. Measuring release events postsynaptically is complicated by the convergence of many rods onto a single bipolar cell (Behrens et al., 2016; Tsukamoto and Omi, 2013). Furthermore, individual quantal release events cannot be observed in ON bipolar cell recordings due to the slow kinetics of the mGluR6 transduction cascade (Berntson et al., 2004; Berntson and Taylor, 2003). To detect individual release events, we took advantage of the fact that the presynaptic glutamate transporters in rod terminals (largely excitatory amino-acid transporter 5 [EAAT5]) are linked to an uncoupled anion conductance so that an anion channel is opened as glutamate is retrieved (Arriza et al., 1997; Schneider et al., 2014). The anion current is linearly proportional to glutamate in the cleft (Otis and Jahr, 1998) and can thus serve as a presynaptic measure of release (Hasegawa et al., 2006). Our measurements of glutamate transporter anion current (I_A(glu)) showed that release rates increased linearly with Ca²⁺ influx but rose only to ∼11 vesicles/s/ribbon at −40 mV (35°C). When rods were voltage clamped at −70 mV, we saw spontaneous release of single vesicles at random intervals, but release at −40 mV involved bursts of ∼17 vesicles apiece spaced at more regular intervals. These bursts required coordination by Syt1. Our experiments indicate that bursts at −40 mV involve vesicles in the readily releasable pool at the base of the ribbon and are triggered by opening of nearby Ca²⁺ channels. Lowering distant Ca²⁺ levels lengthened interevent intervals between bursts, perhaps by influencing ribbon replenishment rates. The presence of these bursts in rods but not cones and their appearance near the resting membrane potential in darkness suggest they may play a role in transmitting rod responses under scotopic conditions.

Control and mutant mice were kept on 12-h dark/light cycles. Animal handling and experimental protocols were approved by the University of Nebraska Medical Center Institutional Animal Care and Use Committee. Mice of both sexes aged 4–12 wk were euthanized in accordance with the American Veterinary Medical Association Guidelines for the Euthanasia of Animals by CO₂ asphyxiation followed by cervical dislocation.

Rod^Syt1cko mutants in which Syt1 was selectively eliminated from rods were bred by crossing Rho-iCre mice (RRID:ISMR_JAX:015850) with Syt1^flox mice (Syt1: MGI:99667) in which LoxP sites flank exon 6 of Syt1 (Quadros et al., 2017). Syt1^flox mice were originally created on a C57/BL6N background. Genetic screening showed that mice used for these experiments were 40% C57/BL6J.

Rod cells were visualized in a flat mount preparation on an upright fixed-stage microscope (Nikon E600FN) using a 60×, 1.0 NA water immersion objective. Rods were identified visually and targeted with positive pressure using recording electrodes mounted on Huxley-Wall micromanipulators (Sutter Instruments). Rod recordings were performed in whole-cell voltage clamp mode using a Multiclamp 700A amplifier (Axon Instruments/Molecular Devices), and signals were digitized with DigiData 1550 (Axon Instruments/Molecular Devices). Data acquisition and analysis were performed using pClamp 10 software (Molecular Devices).

Intracellular pipette solutions for I_A(glu) measurements contained thiocyanate (SCN⁻) to enhance anion currents and consisted of (in mM): 120 KSCN, 10 TEA-Cl, 10 HEPES, 1 CaCl₂, 1 MgCl₂, 0.5 Na-GTP, 5 Mg-ATP, and 5 phosphocreatine (pH 7.3). Intracellular Ca²⁺ was buffered with 5 mM EGTA unless otherwise noted. For I_Ca measurements, Cs-gluconate replaced KSCN, and Ca²⁺ buffering was reduced to 2 mM EGTA. Passive membrane properties measured with KSCN pipette solution showed an average membrane capacitance of 3.2 ± 0.2 pF and membrane resistance of 2.3 ± 0.04 GΩ (mean ± SD; n = 20). Voltages were not corrected for liquid junction potentials unless specifically noted (Cs-gluconate pipette solution, 12.3 mV; KSCN pipette solution, 3.9 mV).

I_A(glu) events were identified and measured for frequency, kinetics, and charge transfer with the event finder function in pClamp directed by a template averaged from ∼10 events. Data were acquired from one to three rods per animal. Statistical analysis was performed using GraphPad Prism 7 and 8 software, and data are reported as mean ± SD unless otherwise noted.

Fig. S1 shows data illustrating that CICR and feedback from horizontal cells are not required for bursting behavior in rods. Fig. S2 shows illustrations of large Ca²⁺-activated Cl⁻ currents that were occasionally activated during depolarizing stimulation.

Presynaptic recordings of I_A(glu) in voltage-clamped rods can provide a reliable presynaptic measure of glutamate release, reflecting the timing and magnitude of exocytotic events (Grassmeyer et al., 2019; Hasegawa et al., 2006; Szmajda and Devries, 2011). In several experimental systems, including rods and cones, the anionic charge transfer has been shown to be proportional to glutamate concentrations in the cleft and likewise proportional to excitatory postsynaptic currents (Hays et al., 2020; Koch et al., 2007; Otis and Jahr, 1998; Szmajda and Devries, 2011). To measure glutamate release presynaptically, we obtained whole-cell recordings from mouse rods using the permeant SCN⁻ as the principal anion in the patch pipette solution to enhance I_A(glu). When rods were voltage clamped at −70 mV, we observed occasional spontaneous presynaptic currents averaging 5.6 ± 1.7 pA (n = 187 events in 12 rods, 8 animals; Fig. 1). As illustrated by the average event shown in Fig. 1, these events exhibited a waveform typical of quantal postsynaptic currents but with slow kinetics, showing a 10–90% rise time of 9.3 ± 1.3 ms and decay time constant of 39.9 ± 9.9 ms (n > 10 events in each of 12 rods). The rising phase of I_A(glu) integrates rising glutamate levels in the synaptic cleft (Hays et al., 2020; Szmajda and Devries, 2011). The slow rate of decline reflects the slow cycle time of glutamate transport (Gameiro et al., 2011; Palmer et al., 2003). Event amplitude histograms assumed the shape of a single Gaussian function, suggesting that the distribution consists primarily of single-vesicle fusion events (Fig. 1 C).

I_A(glu) release events originated from the voltage-clamped rod. We predicted that if glutamate released from neighboring rods and cones could reach transporters in the voltage-clamped rod, then depolarizing the surrounding photoreceptors by bath application of a solution containing 30 mM K⁺ should increase the number of small events. However, application of 30 mM K⁺ did not increase event frequency at −70 mV compared with control extracellular solution in the same cells (0.87 ± 0.5 versus 0.62 ± 0.1 quanta/s; P = 0.28; n = 3 rods, 1 animal), nor did it change the 10–90% rise time (11.3 ± 8.0 ms versus 10.5 ± 6.5 ms; P = 0.36; n = 3 rods) or magnitude of unitary events (4.16 ± 1.82 versus 5.40 ± 2.00 pA; P = 0.47; n = 3 rods). Likewise, hyperpolarizing surrounding photoreceptors by applying a bright white light did not decrease baseline noise that would have been generated from extrasynaptic glutamate (mean ± SD = 0.50 ± 0.03 pA versus 0.52 ± 0.18 pA; P = 0.79; n = 5 rods, unpaired t test).

As illustrated in Fig. 2 A, the character of release changed dramatically as the membrane potential approached −40 mV. All of the spontaneous release events observed when rods were held at −70 mV were uniquantal, showing a unimodal amplitude distribution (Fig. 1 C). However, at −40 mV, rods consistently switched to a more regular bursting behavior with release of many vesicles in each burst (Fig. 2 A, arrowheads). Consistent with earlier findings from mouse rods (Grassmeyer et al., 2019), both bursts and uniquantal I_A(glu) events in mouse rods were blocked by bath application of the glutamate transport inhibitor threo-β-benzyloxyaspartate (TBOA; 100 µM), confirming that they reflect glutamate transporter activity (n = 5 rods, 2 animals; Fig. 2 B). Bursts were not a consequence of using KSCN in the patch pipette to enhance I_A(glu), since we observed similar, albeit smaller, bursts using a CsCl-based pipette solution (n = 4 rods, 3 animals).

To estimate the number of vesicles released during each bursting episode, the charge transfer of each burst was divided by the average charge transfer of occasional unitary events measured at the same potential in the same cell. Because of the high density of EAAT5 glutamate transporters in rods, glutamate reuptake increases even after saturation of postsynaptic mGluR6 receptors, indicating that transporters do not become saturated during release from rods (Hasegawa et al., 2006). We assumed that individual I_A(glu) events summed linearly and thus used charge transfer measurements to count the number of release events during 30-s steps at various holding potentials. The number of quanta in 10-s intervals shown in Fig. 2 is indicated at the right for each trace. In this example, there were 5 unitary release events at −70 mV during the 10-s period and 64 events at −40 mV, the latter occurring in periodic, coordinated bursts of release.

The overall rate of vesicle release measured in this way increased gradually with membrane depolarization (Fig. 2). The increase in release rate tracked voltage-dependent increases in I_Ca (Fig. 2 C), similar to the linear relationship between I_Ca and release observed in amphibian rods (Thoreson et al., 2004). I_Ca was measured in mouse rods using voltage ramps and Cs-gluconate rather than KSCN as the principal ions in the pipette solution. Fig. 2 C plots the voltage-dependent changes in release on the same graph as the voltage dependence of I_Ca after correcting for the different liquid junction potentials. Increasing temperature to 35°C from room temperature (20°C) nearly doubled peak Ca²⁺ influx and caused a proportional increase in release rate (Fig. 2 D). The Q₁₀ values for I_Ca and release rate at −30 mV were nearly identical (2.09 ± 0.78 for I_Ca, n = 9 rods, 3 animals; 2.10 ± 1.22 for release, n = 6 rods, 4 animals). Temperature can affect gating properties of voltage-gated channels, but as with Ca_V1.4 currents in transfected HEK cells, where the window current was unchanged (Peloquin et al., 2008), the midpoint activation voltage for I_Ca in our experiments was not shifted by temperature (Q₁₀ = 1.03 ± 0.02).

Interburst intervals were shortened at higher temperatures, largely accounting for the increase in release rate at depolarized potentials (interburst interval room temperature = 2,756 ± 596 ms, n = 14; 35°C = 1,130 ± 665 ms, n = 5 rods, 5 animals; P < 0.001, unpaired t test). The number of quanta in each burst at −40 mV (−43.9 mV after correcting for liquid junction potential) did not increase significantly (17 ± 7 at room temperature, n = 22 rods, 15 animals, versus 11 ± 6 at 35°C, n = 5 rods, 5 animals; P = 0.5, unpaired t test). Bursts showed faster kinetics with a narrower half-width at higher temperatures (room temperature, 216 ± 69 ms, n = 22; 35°C, 125 ± 33 ms, n = 5; P = 0.03, unpaired t test, n = 5).

Unitary events were occasionally interspersed between bursts. The example in Fig. 3 A comes from a rod that showed numerous small unitary events with a mean amplitude of 4 pA interspersed among larger bursts that averaged 12 pA in amplitude when the rod was held at −40 mV. The amplitude of unitary fusion events diminished with membrane depolarization because of the diminished outward driving force for SCN⁻ at more depolarized potentials (Arriza et al., 1997; Torres-Salazar and Fahlke, 2007; n = 6 cells, 6 animals; open triangles in Fig. 3 B). However, as illustrated by the filled circles in Fig. 3 B, when we plotted the mean amplitude of all events—both bursts and unitary events—from this cell together, average I_A(glu) amplitude initially increased as a function of membrane voltage and then decreased. This initial increase occurred despite a decline in anion driving force and likely resulted from an increase in the number of multiquantal bursts at more depolarized potentials. The average amplitude then diminished with a further decline in anion driving force with further depolarization above −40 mV.

Bursts showed inflection points during the initial inward current, suggesting they arose from sequential fusion of individual vesicles (e.g., Fig. 2, A and B). The examples in Fig. 3, C and D, show increases in I_A(glu) punctuated by inflections at intervals that were roughly the same amplitude as the corresponding single-vesicle events. This suggests that the rising phase consisted of rapid sequential fusion of three vesicles. One could also see individual peaks during the plateau phase of bursts, suggesting that additional fusion events contributed to maintaining the bursts. Event half-widths in room temperature varied widely (60–709 ms) but averaged 216 ± 69 ms (n = 22 rods, 15 animals).

Bursts were both longer in duration and larger in amplitude than individual events. Fig. 3 E shows an amplitude histogram for events measured at a holding potential of −40 mV from the rod shown in Fig. 3 A. This rod was chosen for this illustration because it showed an abundance of unitary events at −40 mV. When fit with a multiple Gaussian, the amplitude histogram showed two discrete peaks, one representing unitary events with a mean amplitude of 4.0 ± 0.7 pA and the other representing bursts with a mean amplitude of 11.7 ± 1.0 pA. Fig. 3 F plots the frequency distribution for charge transfer measurements of the same set of bursts and unitary events. A multiple Gaussian fit yielded a mean amplitude for unitary events of 154.5 ± 32.9 nC with bursts averaging 689.9 ± 73.8 nC. Thus, in this example, the amplitude of bursts was threefold larger than the single-quantal amplitude, but the charge transfer of bursts was fivefold larger than the single-vesicle charge transfer. This suggests that the bursts in this example consisted of an average of approximately five vesicles apiece. Bursts more often (81%; n = 22 cells) consisted of 10–20 vesicles with an apparent upper limit of 45–60 vesicles (3%; n = 22 cells). Quanta per burst calculated from charge transfer (17 ± 7; n = 22 rods) was always larger than the ratio of bursts to unitary events calculated from amplitude measurements (3 ± 1; P < 0.0001, n = 22 cells, t test). The time constant for decline was also two to three times slower for bursts (bursts, τ = 164.4 ± 49.8 ms, n = 31) than unitary events (τ = 71.9 + 40.7 ms, n = 29; P < 0.0001, unpaired t test). Together, these data indicate that multiple vesicles were released sequentially during each burst.

Cones voltage clamped at −40 mV did not show the same bursting behavior seen in rods, but instead showed both small and large individual release events with uniform kinetics, suggesting a more synchronous form of multiquantal release (Fig. 4). The example recordings and amplitude histogram in Fig. 4, A and B, show evidence for numerous large events along with many more small events in a rod held at −70 mV. Fig. 4, C and D, shows the amplitude histogram and response waveforms of release events measured in the same rod held at −40 mV. The insets show examples of large events measured at −70 (Fig. 4 A) and −40 mV (Fig. 4 C). Large and small events showed uniform kinetics, suggesting that large events arose from the synchronized fusion of multiple vesicles. As shown in Fig. 4, A and B, large multiquantal events were even seen during spontaneous release in cones held at −70 mV. These results differed from those in rods where multiquantal release was sequential, not synchronous, and only emerged at more depolarized potentials. Comparing events measured at −40 mV with those measured at −70 mV, the overall amplitude distribution showed a significant rightward shift to more positive values (Kolmogorov-Smirnov test, P = 0.006), despite the decrease in anion driving force at −40 mV (Fig. 4 C). This suggests that the number of multiquantal release events also increased with depolarization. Unitary events had a faster 10–90% rise time (7.4 ± 5 ms, n > 40 events in 3 cells, 3 animals) in cones than in rods (9.3 ± 1.3 ms, n = 12 rods, 9 animals; P < 0.001, t test), suggesting a faster rise in synaptic cleft glutamate. The decay time constant (49.8 ± 79.6 ms, n > 40 events in 3 cells), which is shaped by the glutamate transporter cycle time, did not differ significantly from that of unitary events in rods (P = 0.33, t test). These data show that the bursting behavior is specific to rods, suggesting it may play a role in signaling information under dim light conditions.

The number of quanta in each burst was sensitive to the length of the prior interburst interval. The example trace in Fig. 5 A shows a 30-s series of bursts. While bursts were relatively stereotyped in amplitude and duration, the arrow points to a slightly larger burst that followed a longer-than-average interval. The number of quanta in each burst is plotted against the interburst interval in the graph below. The filled red circles show data points from the example trace. The other data points came from 14 additional 30-s recordings in the same rod. When fit by linear regression, the slope of the straight line was significantly greater than zero (n = 12 rods), with an average correlation coefficient of r² = 0.29 (n = 12). While bursts generally increased in size with longer interevent intervals, they typically reached a plateau of 15–20 vesicles per burst.

Bursts in rods held steadily at −40 mV and release events evoked by brief, strong depolarizing stimuli originated from an overlapping pool of vesicles. With a strong depolarizing test step (25 ms, −70 to −10 mV) designed to release the entire readily releasable pool, release events averaged 21 ± 12 vesicles (n = 9) as measured from I_A(glu) charge transfer following termination of the step. Some bursts were larger than evoked responses, but some were also smaller, and there was no significant difference between the number of quanta in evoked responses and bursts (17 ± 7, n = 22 rods, n = 15 animals; P = 0.17, unpaired t test).

To compare vesicle pools involved in these two forms of release more directly, we tested for cross-depletion. To do so, we voltage clamped rods at −40 mV for 30 s, then applied a brief depolarizing step to −10 mV (25 ms) to deplete the releasable pool, and finally returned the membrane potential to −40 mV for another 30 s. In a few cells, including the cell illustrated in Fig. 6 A, we held cells at −30 mV to evoke more vigorous bursting. As shown in Fig. 6 A, the depolarizing step to −10 mV evoked an inward current that was similar in size and shape to bursts that occurred before and after the step. In the next trial, a burst occurred immediately before the step to −10 mV, and the depolarizing step failed to evoke release (Fig. 6 B). We defined a quantal ratio as the charge transfer in each burst relative to the charge transfer evoked by the step. We measured this ratio for bursts immediately preceding and following the step. The quantal ratio before the step was significantly >1 (1.475 ± 0.680, n = 32; P = 0.0004, one-sample t test), consistent with idea that the preceding burst depleted vesicles from the pool available for a subsequent evoked response. After the step, there was a pause of ∼500 ms during which no bursts were observed. Burst ratios then remained <1 for another 1 s. This is consistent with the previous experiment showing that burst size increased with interburst intervals, at least up to a point. Together, these data suggest that bursts involve a vesicle pool that significantly overlaps with the readily releasable pool of vesicles evoked by strong depolarizing steps.

The rate of release increased in parallel with the voltage-dependent increase in I_Ca, suggesting it was Ca²⁺ dependent. The essential role of Ca²⁺ influx was confirmed by adding the voltage gated Ca²⁺ channel blocker Cd²⁺ to the external solution (n = 3 rods, 2 animals; Fig. 7, A and B). This eliminated bursting, but some unitary release events persisted even in the presence of Cd²⁺ at all potentials, suggesting they arise by Ca²⁺-independent spontaneous release from rods, similar to spontaneous release in salamander rods (Cork et al., 2016; Hays et al., 2020).

As shown in Fig. 2 C, the membrane voltage of −40 mV approaches the midpoint activation voltage for rod I_Ca (Babai and Thoreson, 2009), so we investigated whether increasing the open time of L-type Ca²⁺ channels extended burst duration by recruiting additional vesicles. Lengthening Ca²⁺ channel openings by bath application of BayK8644 (2 µM; Koschak et al., 2003) did not significantly alter release properties at –40 mV. Bursts consisted of 22.2 ± 6.7 vesicles in control conditions compared with 19.6 ± 9.8 vesicles in BayK8644 (P = 0.24, n = 4 cells, 2 animals, paired t test). Burst event duration was also unchanged by extending Ca²⁺ channel open times (control half-width was 217.8 ± 30.6 ms versus 185.5 ± 45.5 ms in BayK8644; P = 0.4, paired t test, n = 4 cells). The frequency of individual uniquantal events in periods of 30 s (12.3 ± 13.8 events) was also not changed by BayK8644 treatment (12.3 ± 11.7 events, n = 4 rods; P = 0.2, paired t test). Thus, overall release rates at −40 mV were not increased with BayK8644 treatment (Fig. 7 C). These data suggest that while the initial influx of Ca²⁺ through Ca²⁺ channels is required to trigger a burst, the number of vesicles in each burst is not strongly shaped by further influx and diffusion of Ca²⁺ from voltage-gated Ca²⁺ channels. Evidence that no additional vesicles are available for release is consistent with the hypothesis that bursts empty the entire readily releasable pool (Thoreson et al., 2016).

Results from mouse and salamander retina have shown that Ca²⁺ released from internal stores by CICR promotes release when rods are tonically depolarized (Babai et al., 2010a; Cadetti et al., 2006; Suryanarayanan and Slaughter, 2006). Consistent with this, we found that blocking CICR by introducing an RYR inhibitor, dantrolene (100 µM), through the patch pipette caused a modest reduction in release rates (P = 0.006, mixed model ANOVA; Fig. 7 D). However, CICR was not required for coordinating bursts, because they persisted after introducing dantrolene or a high concentration of ryanodine (100 µM) into rods through the patch pipette (n = 4; Fig. S1; see text at bottom of the PDF). The replenishment of internal stores by store-operated channels also does not appear necessary for bursts, because application of the Orai/Stim1 blocker SKF96368 (10 µM) through the patch pipette did not eliminate bursting behavior (Fig. S1; n = 3 rods, 1 animal). We also tested whether feedback from horizontal cells to rods might be responsible for organizing bursts, but bursts at −40 mV appeared to be unchanged by bath application of 10 µM α-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid (AMPA) receptor blocker 2,3-dioxo-6-nitro-7-sulfamoyl-benzo[f]quinoxaline (NBQX; Fig. S1; n = 3 rods, 1 animal).

We manipulated intracellular Ca²⁺ buffering to probe the spatiotemporal Ca²⁺ distribution controlling release rates and bursting behavior. Typically, our pipette solution contained 5 mM EGTA as the principal Ca²⁺ buffer. Replacing EGTA with the faster Ca²⁺ buffer, 1,2-bis(o-aminophenoxy)ethane-N,N,N′,N′-tetraacetic acid (BAPTA), reduced the voltage-dependent increase in release rates. The reduction in rate was similar with both 1 and 10 mM BAPTA (P = 0.0004, n = 5–8, mixed model ANOVA; Fig. 8), but neither concentration totally eliminated bursting at −40 mV. Buffering Ca²⁺ with 10 mM BAPTA eliminated I_A(glu) evoked by strong depolarizing steps (25 ms; −70 to −10 mV; 1.04 ± 0.271 pA; n = 3 rods, 3 animals) compared with 5 mM EGTA (9.88 ± 1.77 pA; n = 16 rods, 14 animals; P = 0.049, unpaired t test), showing that BAPTA successfully reached the terminal during whole-cell recording. The number of quanta in each burst was unchanged by either concentration of BAPTA (5 mM EGTA, 17 ± 7, n = 22 cells; 1 and 10 mM BAPTA, 13 ± 4 and 11 ± 7; n = 5 cells each; P = 0.1289, one-way ANOVA). However, switching from EGTA to BAPTA lengthened the intervals between burst events from 2,756 ± 596 in control conditions with 5 mM EGTA (n = 14 cells) to 4,805 ± 1,524 ms with 1 mM BAPTA (P = 0.02, ANOVA; n = 5 cells, 3 animals) and 6,523 ± 2,535 with 10 mM BAPTA (P = 0.008, ANOVA; n = 4 cells, 3 animals). The number of unitary events between bursts was not reduced by 1 or 10 mM BAPTA buffering (10 mM; P = 0.6, t test; n = 5 cells of each condition), consistent with other evidence that these events result from Ca²⁺-independent spontaneous release. Large Ca²⁺-dependent Cl⁻ currents could sometimes be activated during maintained depolarizing stimulation even after strong buffering with 10 mM BAPTA, suggesting that Ca²⁺ buffers may become saturated by the continued influx of Ca²⁺ during 30-s steps (Fig. S2). Nevertheless, the ability of bursts to persist in the presence of strong and fast buffering by BAPTA suggests they involve release sites very close to ribbon-associated Ca²⁺ channels.

We next tested how the Ca²⁺ sensor Syt1 shapes sustained release. Selective elimination of Syt1 from mouse rods abolishes fast, synchronous, Ca²⁺-dependent exocytosis evoked by brief depolarizing steps, although some slower release remains, suggesting the presence of another sensor (Grassmeyer et al., 2019). Elimination of Syt1 also elevated the rate of spontaneous release in rods voltage clamped at −70 mV (Grassmeyer et al., 2019). This is consistent with evidence from other systems that Syt1 helps to clamp the SNARE apparatus and prevent spontaneous fusion (Courtney et al., 2019; Grushin et al., 2019; Xu et al., 2009). Along with an increase in spontaneous release at −70 mV (Fig. 9; 0.45 ± 0.23 events/s in control mouse rods, n = 8, 4 animals, versus 4.2 ± 2.3 events/s in mouse rods lacking Syt1, n = 9, 5 animals; P < 0.001, t test), we saw a further small increase in release rates with depolarization above −50 mV in rods lacking Syt1. The small voltage-dependent increase in release at −40 and −30 mV that remained after eliminating Syt1 from rods was abolished by replacing 5 mM EGTA with 1 mM BAPTA in the patch pipette (Fig. 9 B). This suggests that this residual release may involve nonribbon release sites that use a different Ca²⁺ sensor (Grassmeyer et al., 2019). Above −50 mV, roughly the same number of quanta were released over long periods of time (30 s) as in control rods, but events in rods lacking Syt1 were largely uniquantal and not organized in bursts (Fig. 9 A). Fig. 9, C and D, plots the frequency distribution for charge transfer measurements in a rod voltage clamped at −70 and −40 mV. Unlike control rods that showed a bimodal distribution of charge transfer when held at −40 mV (Fig. 3), rods lacking Syt1 showed a unimodal distribution at both −70 and −40 mV (Fig. 9, C and D). Furthermore, the mean single-event charge transfer was smaller at −40 mV (−396 ± 175 nC; n = 75 events) than at −70 mV (−577 ± 180 nC; n = 48 events). This suggests that unclamping Syt1 may facilitate fusion of vesicles that would normally be released only after an increase in Ca²⁺ and that the clamping function of Syt1 may be necessary to maintain inactive periods between bursts, allowing accumulation of a sufficient number of vesicles to initiate a burst at −40 mV.

An appreciation of the mechanisms underlying tonic release of glutamate-containing vesicles from rods, graded by membrane potential, is critical to understanding how synaptic transmission is regulated by light-evoked changes in rod membrane potential. Using I_A(glu) to assay release from individual rods, we measured the voltage-dependent increase in glutamate release at individual rod ribbon synapses. Release rates rose gradually in parallel with the voltage-dependent increase in I_Ca, attaining a rate of ∼11 vesicles/s (35°C) at the typical resting potential in darkness of −40 mV. Along with this increase in rate, the nature of release changed, shifting from uniquantal release at −70 mV to coordinated bursts of vesicles at −40 mV. In this study, we explored the mechanisms underlying these different components of release.

Our results showed a linear dependence of release on I_Ca, similar to earlier results in amphibian rods (Thoreson et al., 2004). Voltage-dependent changes in glutamate release rate tracked with I_Ca even when the temperature of the system was increased to physiological levels. Fusogenicity of the SNARE complex employed by rods for exocytosis is controlled by entropic forces, so complex formation and zippering are not aided by increases in temperature (Mostafavi et al., 2017). Thus, rate appears to depend chiefly on the increase in I_Ca that accompanies increasing temperature with Q₁₀ = 2.1. Voltage-gated Ca²⁺ channels in rods are L-type channels containing Ca_V1.4, with several splice variants that have different kinetics and gating properties (Haeseleer et al., 2016; Tan et al., 2012). While the gating properties of some voltage-gated ion channels are strongly temperature dependent (Q₁₀ = 5–10; DeCoursey and Cherny, 1998; Hille, 2001; Lee and Deutsch, 1990), our experiments agree with the relatively weak temperature dependence of Ca_v1.4 (Q₁₀ = 2–4) measured in various systems (Allen, 1996; Herve et al., 1992; Hope et al., 2005). Studies of synaptic release in bullfrog inner hair cells that also contain synaptic ribbons and employ L-type Ca²⁺ channels (Brandt et al., 2005) describe a stronger temperature dependence of release frequency but agree with the present study in that increases in release are likely due to enhanced ion currents rather than changes in the fusion machinery itself (Chen and von Gersdorff, 2019; Sabatini and Regehr, 1996). It is possible that other parts of the synaptic vesicle cycle are accelerated by temperature, but the large reservoir of cytoplasmic vesicles makes rods less immediately reliant on fast endocytosis for refilling the readily releasable pool (von Gersdorff and Matthews, 1997). The tight relationship between release and Ca²⁺ influx improves linearity in the transformation of light-evoked changes in membrane potential into synaptic release.

Spontaneous release can be important for circuit development, synaptic plasticity, and shaping spiking patterns (Andreae et al., 2012; Carter and Regehr, 2002; McKinney et al., 1999). At a hyperpolarized potential of −70 mV, where Ca²⁺ channels are minimally active, spontaneous release rates were slow, with rates of 0.45 ± 0.24 Hz at room temperature and 0.7 ± 0.6 Hz at 35°C. This is similar to rates of Ca²⁺-independent release in amphibian rods and cones of 1.3 vesicles/s/ribbon and 0.8–0.9 vesicles/s/ribbon (Cork et al., 2016; Hays et al., 2020). Spontaneous release can be triggered by the stochastic opening of L-type Ca²⁺ channels, but vesicles can also be released from salamander rods independent of Ca²⁺ influx (Cork et al., 2016). Our results show that mouse rods are capable of Ca²⁺-independent spontaneous release, since strong buffering by 10 mM BAPTA or inclusion of CdCl₂ in the extracellular medium did not eliminate spontaneous fusion events.

The increase in spontaneous release rates observed in Syt1-knockout rods is consistent with the hypothesis that Syt1 serves as a fusion clamp (Courtney et al., 2019). In the absence of Syt1, we observed a slight increase in release rates above −50 mV that was depressed by introducing the fast Ca²⁺ buffer BAPTA into rods. This suggests that the enhanced spread of Ca²⁺ that accompanies strong stimulation may recruit more distant release sites that use a different Ca²⁺ sensor. Syt7 and Doc2 have both been proposed as sensors for slower asynchronous release that is not tightly coupled to Ca²⁺ channel openings (Bacaj et al., 2013; Chung et al., 2010; Luo et al., 2015; Turecek and Regehr, 2018; Yao et al., 2011). Doc2 has also been proposed to be a sensor for spontaneous release (Pang et al., 2011; Ramirez et al., 2017; Yao et al., 2011), and RNA-sequencing experiments show high levels of Doc2B mRNA in rods (Kim et al., 2016).

Our data showed that while spontaneous release from rods held at −70 mV involved the stochastic release of individual vesicles, depolarizing rods to −40 mV caused the emergence of multiquantal bursts of release consisting of 10–20 vesicles apiece. In addition to their voltage dependence, bursts were blocked by Cd²⁺, indicating that they required Ca²⁺ influx through voltage-dependent Ca²⁺ channels. Furthermore, bursting at −40 mV was absent from rods lacking the exocytotic Ca²⁺ sensor Syt1 (Grassmeyer et al., 2019). While other Ca²⁺ sensors may also be present in rods and participate in coordinating bursts, it is clear that Syt1 is absolutely essential. The loss of Syt1 may directly interrupt bursting or act indirectly by depleting an overlapping pool of vesicles; nevertheless, these results indicate that the vesicles involved in bursting are released by Syt1-mediated mechanisms. In addition, because I_Ca and passive membrane properties were unchanged in rods lacking Syt1 (Grassmeyer et al., 2019), the absence of bursting in these cells with the same basic biophysical properties indicates that bursts were not due to a loss of voltage-clamp and regenerative activation of I_Ca.

Inclusion of 10 mM BAPTA in the patch pipette did not abolish the bursting that accompanied maintained depolarization. The persistence of bursts in the presence of BAPTA suggests that they involve release sites within Ca²⁺ nanodomains extremely close to Ca²⁺ channels. Since Ca²⁺ channels are clustered beneath ribbons (Lv et al., 2012; Mercer et al., 2011; Morgans, 2001; Specht et al., 2009), this in turn suggests that ribbon-associated vesicles participate in bursting. While the buffering capacity could be overwhelmed by the continued influx of Ca²⁺ during prolonged depolarizing stimulation, the cross-depletion between vesicle pools involved in bursts and depolarization-evoked release, along with evidence that the number of vesicles in bursts matched the number of vesicles released by a brief strong depolarizing step, suggests that the pool of vesicles involved in bursts overlaps considerably with the readily releasable pool of vesicles, which is thought to reflect vesicles tethered at the base of the ribbon (LoGiudice and Matthews, 2009; Mennerick and Matthews, 1996).

Unlike the bursting in rods that involved sequential fusion of vesicles, release events in cones varied widely in amplitude, but large events exhibited the same kinetics as small events, suggesting large events are due to the synchronized fusion of multiple vesicles. One aspect of release that differs between the two cell types is that rods, but not cones, are capable of release at nonribbon sites triggered by CICR (Babai et al., 2010b; Cadetti et al., 2006; Chen et al., 2014; Chen et al., 2015). The capability of nonribbon release in rods is akin to nonribbon release driven by global Ca²⁺ changes in retinal bipolar cells (Mehta et al., 2014). We examined whether CICR was necessary to coordinate bursting but found that bursts were not abolished by the CICR inhibitor dantrolene, blocking RYRs with a high concentration of ryanodine, or use of a store-operated channel inhibitor. These data suggest that release of Ca²⁺ from stores is not required for bursting behavior. We also considered the idea that bursts required the coincident opening of a certain number of Ca²⁺ channels. However, the intervals between bursts are not likely to be due to cessation of Ca²⁺ influx since Ca_V1.4 channels show minimal inactivation. If we assume a single-channel current of 0.3 pA (Bartoletti et al., 2011; Brandt et al., 2005), the amplitude of I_Ca at −40 mV suggests that seven or eight channels should be open at any given moment. Furthermore, lengthening Ca²⁺ channel openings with the dihydropyridine agonist BayK8644 did not reduce interburst intervals or extend burst durations. The finding that the increase in Ca²⁺ influx accompanying treatment with BayK8644 did not lengthen bursts is also consistent with the idea that each burst empties the entire pool of vesicles available for immediate release (Thoreson et al., 2016).

The finding that larger bursts were generally preceded by longer interburst intervals and evidence from cross-depletion experiments both suggest that bursts are shaped by vesicle replenishment kinetics. Ca²⁺ speeds replenishment at rod and cone ribbon synapses (Babai et al., 2010a; Van Hook et al., 2014; Van Hook and Thoreson, 2015), so the finding that inhibiting CICR with dantrolene slowed release rates is consistent with a role for Ca²⁺ in speeding replenishment. Interburst intervals were also extended by BAPTA, consistent with a slower rate of replenishment with less available intracellular Ca²⁺ (Babai et al., 2010a; Van Hook et al., 2014; Van Hook and Thoreson, 2015). Inhibiting calmodulin slows replenishment in rods and cones (Van Hook et al., 2014; Van Hook and Thoreson, 2015). Interactions between calmodulin and the Ca²⁺ sensor Syt7 can also regulate replenishment (Liu et al., 2014).

Burst sizes appear constrained by the number of vesicles available for release. While bursts were initiated by fusion of a few vesicles at the base of the ribbon located within Ca²⁺ channel nanodomains, the release of subsequent vesicles during the burst involved sequential fusion of additional vesicles. The sequential nature of vesicle fusion during these bursts looks superficially similar to multiquantal release in hair cells (Grant et al., 2010). In hair cells, it has been suggested that complex kinetics of release events arise from flickering of fusion pores (Chapochnikov et al., 2014). In our hands, unitary I_A(glu) events at −70 mV were uniform in both kinetics and magnitude, suggesting full vesicle fusion, releasing one whole quantum during spontaneous release. The burst events at −40 mV consistently had larger charge transfer than spontaneous unitary events, so it seems unlikely that bursts originated from only a single vesicle.

Bursting behavior was not observed previously in salamander rods, where multivesicular release instead involves synchronized fusion (Hays et al., 2020). Amphibians lack the rod bipolar cells that in mammals are specialized for transmitting rod light responses at scotopic threshold (Wu, 2010). Bursts were also not seen in cones of the mouse and salamander. These results suggest that bursts may have a particular role in transmitting scotopic signals. One strategy for achieving reliable transmission of single-photon responses at a rod synapse is to maintain extremely fast release rates; however, we found that release from rods at −40 mV and 35°C averaged only ∼11 vesicles/s/ribbon, substantially lower than the rate of ∼100 vesicles/s that is thought to be needed for reliable transmission, assuming a purely Poisson release process (Rao-Mirotznik et al., 1998; Rao et al., 1994; van Rossum and Smith, 1998). Our results in mouse rods are consistent with measurements made using the activity-dependent dye FM1-43, which showed release rates in darkness of 2.5 and 18 vesicles/s/ribbon from gecko and salamander rods, respectively (Sheng et al., 2007). Maintaining a high rate of release during long periods of darkness requires a tremendous amount of energy and other cellular resources (Linton et al., 2010; Okawa et al., 2008; Yuan et al., 2018). An alternative strategy for making detection more reliable is to make release more regular and predictable (Schein and Ahmad, 2005; Schein and Ahmad, 2006). At the dark resting membrane potential of −40 mV, we consistently observed large multivesicular release events that occurred at semiregular intervals. The replenishment and delivery of vesicles to release sites at the base of each ribbon may help to make bursting occur at more regular intervals, and this may improve postsynaptic detection of single-photon responses.

To improve the detection of single photons at the synapse from rods, rod bipolar cells employ a nonlinear thresholding mechanism that filters out small signals that are more likely to arise from noise (Field and Rieke, 2002; Sampath and Rieke, 2004). To implement this thresholding mechanism, glutamate levels in the synaptic cleft must be maintained at high levels in darkness (Sampath and Rieke, 2004). The periodic release of many vesicles in a burst may provide a means of keeping glutamate levels high in rod synapses during darkness.

In summary, our results described several key factors that shape voltage-dependent changes in synaptic release of glutamate-filled vesicles from rods in the mouse retina. At negative membrane potentials normally achieved only with extremely strong illumination and where Ca²⁺ channels show little or no activity, individual synaptic vesicles are released sporadically in a Ca²⁺-independent fashion. As Ca²⁺ channels open more frequently with increasing depolarization, vesicle release rates increase in parallel. However, the rate of vesicle release achieved at the typical resting potential in darkness of −40 mV remains slow, only ∼11 vesicles/s. At this membrane potential, release transitions from individual vesicle fusion events to coordinated bursts arising from the sequential fusion of 10–20 vesicles. These bursts involve the readily releasable pool of vesicles at the base of the ribbon and occur at more regular intervals than the stochastic fusion events seen at −70 mV. Intervals between bursts are shortened by elevating global Ca²⁺, suggesting they are shaped by rates at which the releasable pool can be replenished. This is consistent with the idea that delivery of vesicles down the ribbon may provide a mechanism for regularizing release rates (Schein and Ahmad, 2005). Deletion of the exocytotic Ca²⁺ sensor Syt1 virtually eliminated voltage-dependent increases in sustained release at membrane potentials up to −40 mV. Consistent with effects of Syt1 deletion in cones (Grassmeyer et al., 2019), this indicates that kinetics of release at photoreceptor ribbon synapses are shaped more by kinetics of vesicle delivery to release sites at the ribbon base than by reliance on different release sites and different Ca²⁺ sensors as seen at many conventional synapses (Kaeser and Regehr, 2014).

Jeanne M. Nerbonne served as editor.

We thank Cody Barta for invaluable assistance with mouse husbandry.

Support was provided by a University of Nebraska Medical Center graduate fellowship to C.L. Hays and National Eye Institute grant EY10542 to W.B. Thoreson.

The authors declare no competing financial interests.

Author contributions: C.L. Hays and W.B. Thoreson planned the study, conducted experiments, analyzed data, and wrote the paper. A.L. Sladek conducted experiments and analyzed data.