Presynaptic Calcium Transients in Bipolar Cells Exhibit Phase-Locking In Vivo
To investigate how spikes in retinal bipolar cells transmit information about a visual stimulus, we began with an in vivo preparation—zebrafish expressing SyGCaMP2 under control of the ribeye A promoter [12, 21, 22
]. SyGCaMP2 monitors presynaptic calcium transients that control neurotransmitter release. A full-field stimulus was modulated at 1 Hz (100% contrast, square wave) and summarizes results from 1821 terminals in five fish in which all strata of the inner plexiform layer (IPL) were sampled equally. Although some terminals generated purely graded responses to light (A), 93% were also capable of generating fast calcium transients, probably reflecting voltage spikes in the presynaptic compartment (B and Figure S1
, available online).
Phase-Locking of Calcium Spikes in the Synaptic Terminals of Bipolar Cells Observed In Vivo
Synaptic terminals were placed into four functional groups, and examples of each are shown in B. The response to the onset or offset of steady light identified ON and OFF terminals, which constituted 21% and 52% of the total, respectively. The remaining terminals did not respond clearly to a light step, but 8% responded to contrast with spikes. Finally, 19% of terminals generated spikes that were not modulated by either stimulus. Bipolar cells generating strong, graded responses continuously followed the fluctuations in light intensity, with ON and OFF terminals in antiphase (A, bottom). These graded calcium signals could be distinguished from spikes by their slower rate of rise, as illustrated in Figure S1
. Notably, spikes occurred sparsely (B and 1C), and the fraction of stimulus cycles in which a spike occurred (reliability) varied widely between different terminals, as shown by the histogram in D. The mean reliability in the ON, OFF, and contrast-sensitive groups of terminals were 5.9%, 7.4%, and 11.4%, respectively.
Although presynaptic calcium spikes occurred sparsely, their timing was strongly dependent on the phase of the stimulus. For instance, in the ON cell in C, most spikes were initiated while the luminance was high, whereas the OFF cell spiked while luminance was low. The tendency of a neuron to respond at a particular phase of a periodic stimulus is termed phase-locking, and in the auditory system this is one of the strategies by which neurons encode the frequency of sound [7
]. The degree of phase-locking can be quantified as the vector strength [23
], a metric that varies between zero (spikes independent of the stimulus) and 1 (perfect synchronization of all spikes at one particular phase-lag; Experimental Procedures
). Vector strengths were indistinguishable in ON, OFF, and contrast-sensitive populations of terminals firing spikes, and averaged 0.5 ± 0.23. In the 10% of terminals with the highest reliability (averaging 19.9% ± 7.9%), the average vector strength increased to 0.75 ± 0.12 in ON terminals and 0.71 ± 0.12 in OFF terminals.
Phase-locking will cause a population of neurons to respond at a particular phase of a stimulus if those neurons tend to be synchronized. Such synchronization was found within the population of ON and OFF cells, as shown by the distribution of spike times relative to the stimulus cycle (E). For instance, in the 10% most reliable terminals, the OFF group were almost completely silent at a phase of the stimulus in which the ON group spiked with the highest probability (F). Synchrony within these bipolar cell populations probably arises from shared synaptic inputs received from photoreceptors. However, strikingly different behavior was observed in the 8% of terminals sensitive to contrast but of no clear polarity: spikes were distributed relatively uniformly as a function of phase-lag, indicating that shared photoreceptor input was not sufficient to synchronize activity within this population (E).
Light-Driven Spikes in Retinal Bipolar Cells: Electrophysiology
The visual systems of fish and mammals detect stimuli at frequencies of 20 Hz or more. What is the temporal precision with which spikes in bipolar cells encode these frequencies? This question was difficult to investigate by imaging SyGCaMP2 because the calcium transient detected by a high-affinity reporter decays with a time constant of about 1 s in the synaptic terminals of bipolar cells [12, 24
]. To achieve the appropriate temporal resolution, we therefore turned to electrophysiological recordings in slices of goldfish retina.
Whole-cell recordings from a total of 128 bipolar cells were made with the pipette on the soma; 81% of the cells responded to current injection with spikes. A 500 ms step of light modulated spiking in 37 of 47 cells (79%), confirming that the population of bipolar cells made wide-spread use of spikes to encode a visual stimulus. Both graded and spiking responses occurred with high temporal precision, as shown by the six examples in A–2F. For instance, in the ON cell in A, spikes occurred at a mean interval of 1.6 s in the dark, but in 8/10 trials a single spike occurred 75 ± 9 ms after light onset. Injection of a small hyperpolarizing current to prevent the membrane potential crossing spike threshold revealed the underlying generator signal as a transient depolarization just after light onset and a transient hyperpolarization at light offset (blue trace). Other examples of cells generating spikes include a fast ON (B), sustained OFF (C), delayed ON (D), and transient OFF (E). Finally, F shows a cell generating a damped voltage oscillation at both light onset and offset, similar to the electrical resonance reported in isolated bipolar cells [14
]. The large majority of spikes in bipolar cells were generated by voltage-dependent calcium channels (L and/or T type), but some cells also fired sodium spikes blocked by TTX (Figure S2
Light Modulates Spikes Recorded in Bipolar Cells in Retinal Slices
A number of studies have failed to clearly observe spikes in recordings made from bipolar cells in eyecup or slice preparations from a number of species [9, 10, 25
]. Two factors may have made detection of spikes difficult in fish. First, recordings in the cell body do not faithfully detect spikes that originate in the synaptic terminal: the voltage signal detected in the soma is attenuated and filtered in time, at least to some degree, by the high resistance of the connecting axon and the capacitance of the soma (Figure S2
). Second, the spike-generating mechanism could be destroyed by whole-cell dialysis. When recordings were made with a control intracellular solution, spikes were typically destroyed within 30–60 s (G and 2H and Experimental Procedures
). But when the solution in the pipette included 10 mM creatine phosphate, rundown could be prevented. Creatine phosphate is a mobilizable reserve of high-energy phosphates found in muscle and brain, commonly used to preserve the function of calcium channels, mitochondria, and synapses during electrophysiological recordings [26, 27
]. It appears that spikes in bipolar cells are very sensitive to these energy stores, probably reflecting a regenerative mechanism dependent on voltage-sensitive calcium channels [14, 16
In most neurons, spikes are generated in the axon initial segment, close to the soma. We could not, however, record spikes in any of 21 axotomized bipolar cells, as judged by dye-filling. In contrast, spikes occurred in 104 of 128 cells in which the terminal remained attached to the axon (A and 2D). These results provide strong support for the idea that fast presynaptic calcium transients observed in vivo are caused by voltage spikes generated in the synaptic terminal ( and [12
]). Our observations are consistent with previous reports of spikes in an ON class of bipolar cell in the retina of goldfish [13, 16, 20
], and further reveal that spikes are used to encode light in OFF cells and in neurons with varying kinetics, including those generating sustained signals.
Spikes in Bipolar Cells Phase-Lock with Millisecond Precision
To explore the temporal precision with which spikes in bipolar cells encoded a visual stimulus, the intensity of full-field light was modulated sinusoidally at 100% contrast and the frequency swept up from 0 to 20 Hz, and then back down. This chirp stimulus was applied 10–50 times, and A illustrates how the response of an individual cell varied from trial-to-trial. An expanded view of the response of an ON and OFF cell is shown in B. In both channels, spike times were consistent between presentations, as can be appreciated by the raster plot (top), and the superimposition of voltage recordings (below).
Temporal Precision of Spikes in Retinal Bipolar Cells
Responses to the chirp stimulus were analyzed with the approach illustrated in C–3E, which show ON and OFF cells as examples. First, the phase shift between each spike and the stimulus cycle in which it occurred was plotted as a function of frequency (C). Phase shift varied linearly with frequency, as shown by the lines fitted to the points, indicating that the phase composition of the signal from both cells was preserved. As a result, ON and OFF responses remained in antiphase at frequencies up to 20 Hz (cf. E and 1F). In other words, the mean time delay between the firing of the ON cell and OFF cell was maintained at half the stimulus period, independent of changes in frequency.
The next step in the analysis was to investigate how the temporal precision of spiking varied as a function of frequency. The time deviation of each spike from the line describing the response of the linear system best fitting the data was measured (C). D shows that the dispersion of this timing deviation fell as frequency increased for both the ON and OFF cell. Finally, the standard deviation in spike times over bandwidths of 4 Hz was calculated. This quantity, the temporal jitter, is plotted as a function of frequency in E, with results averaged over eight cells (four ON and four OFF). Temporal jitter fell as frequency increased according to a power function, as shown by the log-log plot in F (slope −0.72). Over the 16–20 Hz window, the temporal jitter was 2.2 ms in the OFF cell shown in B and 2.8 ms in the ON cell, and averaged 3.8 ± 1.6 ms (n = 8). The strength of phase-locking was then quantified through the relation between vector strength and temporal jitter (Experimental Procedures
). Vector strength was constant across frequencies at a value of 0.9 (E), which is similar to that observed in auditory nerve fibers [6
Spike Generation in Bipolar Cells Was Less Reliable than in Postsynaptic Amacrine Cells and Ganglion Cells
Low Reliability of the Spike Code in Bipolar Cells
A key difference in the spike code of neurons in the inner retina emerged when we compared the reproducibility of their responses to repeated presentations of the chirp stimulus: spikes in amacrine cells and ganglion cells were generated much more reliably than spikes in bipolar cells. Example records in an amacrine cell and ganglion cell are shown in A, in which each cycle of the chirp elicits a single or double spike with a probability approaching 100%. In contrast, when the response to this stimulus was recorded in a bipolar cell, no single cycle of the chirp stimulus consistently elicited a spike, as can be seen in the examples in . To quantify the reliability of signaling, we counted the fraction of stimulus presentations in which a given cycle of the stimulus generated one or more spikes. Across a sample of eight bipolar cells reliability fell as a function of frequency and averaged ~15% above 15 Hz (B, black). In contrast, the reliability of postsynaptic amacrine cells and ganglion cell did not fall below 80% (blue). Despite these differences in reliability, the temporal precision of spikes were similar across all three types of neuron (C).
Spike Generation Did Not Significantly Alter Frequency Tuning
Temporal filtering of the visual signal can be shaped by active conductances within neurons [28–30
]. For instance, comparison of the generator potential with the spike output demonstrates that the spike-generating mechanism acts as a high-pass filter in cat retinal ganglion cells [31
] and as a band-pass filter in ocellar neurons in the cockroach visual system [32
]. In the case of bipolar cells, interactions between voltage-sensitive calcium channels and calcium-activated potassium channels in the synaptic terminal can generate both spikes and an electrical resonance [14, 20
]. Does the spike-generating mechanism in bipolar cells alter the time course of the visual signal?
To investigate the transfer function of bipolar cells, the retina was stimulated 10–50 times with a chirp. In the OFF cell in A, the probability of firing peaked at a stimulus frequency of 11–12 Hz, and when spiking was prevented by injection of hyperpolarizing current, the graded voltage signal also exhibited bandpass tuning with maximum amplitude at the same frequency (C). The ON cell in B also exhibited bandpass behavior at the level of both spikes and the underlying generator potential, but with peak gain at 7–8 Hz. The relation between the peak frequency for transmission of the generator potential and spikes is plotted for each of seven cells in E. The line through the points has a slope which is not significantly different from equality (1.02 ± 0.06). We conclude that the spike-generating mechanism in bipolar cells does not significantly affect frequency tuning but simply converts an analog (graded) signal to a digital (spiking) format.
Spikes and Graded Responses Represent Similar Temporal Filters
Switches in the Signaling Mode of Individual Terminals
To investigate how graded and spiking signaling modes interact in the synaptic terminal of bipolar cells, we analyzed SyGCaMP2 signals in the time-frequency domain with a Wigner Distribution Function [33
]. Of 1818 terminals imaged in vivo, only 39 generated a purely graded signal that followed a stimulus modulated at 1 Hz (100% contrast), and an example is shown in A. The accompanying spectrogram confirms that the modulation in presynaptic calcium was locked on to the 1 Hz stimulus, as can also be seen in the power spectrum. However, the vast majority of terminals (1400) generated spikes as well as a continuous presynaptic calcium signal, and an example is shown in B. Although this terminal began responding to the stimulus in a graded manner (i.e., simply following the 1 Hz stimulus), after about 30 cycles it also began to spike at 0.25 Hz. The next transition involved a complete switch in signaling modes: the continuous component of the presynaptic calcium signal stopped and spikes at 0.25 Hz took over. As a result, the power spectrum averaged over two minutes of the periodic stimulus contained peaks at 1 Hz and at 0.25 Hz, although with weaker components at 1/2 and 1/3 the fundamental frequency also evident.
Calcium Spikes and Graded Calcium Signals: Switches in Signaling Mode
A second example of a terminal switching between different signaling states is shown in C. The amplitude of the signal at 1 Hz waxed and waned, but the spiking component contained more power and switched between frequencies of 1/2, 1/3, 1/4, and 1/5 Hz. All spiking terminals demonstrated sudden changes in frequency over the period of stimulation, and these were invariably accompanied by modulation of the graded signal. Apart from rapid adaptation of the initial response to the stimulus, these changes in spike frequency occurred randomly: we could not detect time-dependent trends over the whole population of 1769 spiking terminals. In 35% of terminals, the continuous calcium signal disappeared almost completely for some periods.
To assess the relative contributions of graded and spiking responses across the population of bipolar cells we averaged the power spectra of SyGCaMP2 signals from all 1821 terminals, regardless of response (D). The two dominant modes of encoding the stimulus were continuous modulations of calcium at 1 Hz and phase-locking of spikes at four times the fundamental period. To isolate the contribution of spikes, we measured interspike intervals across the whole sample of cells and plotted the distribution of instantaneous spike frequencies (E). In this case, the dominant signal was not at 1 Hz: more spikes fell around the peaks corresponding to frequencies of 1/2, 1/3, and 1/4 Hz.
Three aspects of the behavior shown in are notable. First, graded and spiking signaling modes often occurred simultaneously in an individual terminal. Second, spiking responses could jump up and down in frequency, and even switch on and off. Third, the amplitude of the graded component of the presynaptic calcium signal could be strongly modulated.
A Narrow Voltage Window for Spike Generation
To investigate the mechanisms that might underlie spontaneous changes in the efficiency of spike generation, we used retinal slices to make prolonged voltage recordings in bipolar cells while stimulating with repeated chirp stimuli, as shown by the example in A. In this cell the resting potential at the start of the recording was −43 mV while the spike threshold was −40 mV (indicated with the dashed red line), allowing a generator potential of only a few mV to drive firing. Spikes were suddenly switched off by a spontaneous hyperpolarization of 5 mV that prevented the voltage from reaching threshold (indicated by the blue box). Subsequently, a slow depolarizing drift in membrane potential led to restoration of the spiking response to light but with lower efficiency.
Spiking Is Switched on and off by Fluctuations in Resting Membrane Potential
Spiking could also be blocked by depolarization of the resting potential above the spike threshold. In the example in A, a depolarization to −38 mV was sufficient to block all light-driven spikes, as shown by the later period of recording from this cell boxed in green. Drifts in resting potential of ±5 mV were observed in all 121 cells in which recordings were maintained for periods of minutes, causing the efficiency of spike generation to vary continuously. In 10% of cells, sudden flips in membrane potential were also observed.
To test systematically how membrane voltage affected spike generation we held bipolar cells at different potentials by injection of small currents from the cell body and measured voltage responses to light. In the example in B, spike threshold was −46 mV, and spiking was extinguished at −38 mV. Population data from 238 stimulus presentations in eight cells are shown in C. The transition from purely graded signaling to a combination of spikes and graded signals occurred over a voltage range of less than ~3 mV. The width of this window probably reflects the amplitude of the graded voltage signal, which also averaged 4 mV (lower graph). A second method for determining the voltage window for spike generation was to apply a current ramp. Two examples are shown in D: a cell generating small slow spikes from a threshold of −45 mV (probably driven by L-type calcium channels) is shown in black, and a cell generating fast large spikes from a threshold of −78 mV (probably driven by sodium channels) is shown in red. Collected results from 49 bipolar cells are shown in E, which plots the relation between the spike threshold and the potential at which spikes were extinguished. In none of these cells could spikes be observed at a potential above −35 mV, and in the majority the voltage window for spike generation was less than 12 mV.
Together, the results in demonstrate that the reliability of spike generation in the synaptic terminal of bipolar cells is finely controlled by small changes in membrane potential. This behavior provides an explanation for mode-switching imaged in vivo: spikes can be switched off by small and spontaneous hyperpolarizations that prevent the membrane potential from crossing threshold or by small depolarizations that extinguish the spike-generating mechanism. These results also provide an explanation for changes in the amplitude of the graded calcium signal observed in vivo (B and 6C): the threshold for activation of the L-type calcium current is around −43 mV [34
], and in many bipolar cells the resting membrane potential fluctuated around this value (B and 6C).