Temporal Transformation of the Rod Single-Photon Response in the Retinal Circuitry

Abstract

Purpose: The temporal sensitivity of rod vision is higher than expected based on slow single-photon responses produced by phototransduction in rod outer segments. We sought to establish the retinal basis for the temporal speeding of rod responses and their dependence on phototransduction.

Methods: We made patch-clamp recordings of photocurrent and photovoltage from rods and rod bipolar cells (RBCs), and loose-patch and voltage-clamp recordings from ON α retinal ganglion cells, in three mouse strains: wild-type (WT), rhodopsin hemizygotes (Rh^+/−), and GCAPs^−/−.

Results: Single-photon responses from the three strains differed in the kinetics of their waveforms: Rh^+/− were more rapid than WT and GCAPs^−/− considerably slower. RBCs in all three strains decayed much faster than their respective rod current responses, and much of this transformation was the prior result of conversion of rod photocurrent to photovoltage. Light responses were further accelerated at the bipolar-cell synapse and in the inner plexiform layer, with the remarkable result that ganglion-cell EPSPs and spike output both nearly overlap in waveform from all three strains despite large differences in rod photocurrents. These kinetic alterations were mostly affected by linear filters at each step in integration, with some contribution also from nonlinear interactions especially apparent in the GCAPs^−/− retina.

Conclusions: The retina extracts principally the initial phase of single-photon responses through a series of high-pass, largely linear filters at each integration stage. Much of this transformation occurs in the rods themselves, but important contributions are also made at the rod-to-RBC synapse and within the inner plexiform layer.

As anyone playing catch near dusk has experienced, the temporal sensitivity of vision declines in concert with the light intensity. Part of this reduced temporal sensitivity is the result of a shift from cone-mediated to rod-mediated vision,¹ but even within the scotopic regime temporal sensitivity continues to decline as light intensities approach the threshold for seeing. Rod vision near threshold displays a flicker fusion frequency of approximately 3-4 Hz,² which prima facie is higher than would be predicted based on the slow rod photocurrent observed in primate rods, which can last approximately one second.^3,4 The process of temporal speeding, or high-pass filtering, is known to begin at the synapse with bipolar cells,⁵ and in mammalian rod bipolar cells the photocurrent is essentially complete by the time the rod photocurrent has reached a peak.⁶ Further estimates of temporal resolution near visual threshold from the primate retina suggest that retinal ganglion cells may support temporal frequencies faster than 10 Hz,⁷ of the same order as the timing threshold for distinguishing two single-photon responses in primate rods of ∼50 ms.⁸ These observations all suggest that processes within the retina speed retinal responses to encode higher temporal frequencies than might be predicted based on the slow rod photocurrent, but the retinal location and mechanistic basis for this temporal transformation remains only partially understood. 

To address this deficiency, we have explored the transformation of the time-dependent signals of photoreceptors in downstream retinal cells. Rods in mammals are known to synapse onto a single type of bipolar cell, the rod bipolar cell (RBC).^9–12 The RBCs then synapse onto AII amacrine cells, which make glycinergic synapses with OFF cone bipolar cells,^13,14 as well as electrical junctions with ON cone bipolar cells before terminating onto OFF and ON ganglion cells. To explore the transformation of rod signals in this network, we recorded the dim-light responses of rods, RBCs, and ON α ganglion cells. We took advantage of two transgenic mouse lines that display altered single-photon response kinetics: mice hemizygous for rhodopsin (Rh^+/−) and knockout mice for guanylate cyclase activating proteins 1 and 2 (GCAPs^−/−), which display either accelerated or slowed single-photon responses compared to wild-type (WT) mice.^15–17 

We show in physiological recordings from different levels of the retinal circuitry that rod single-photon responses are high-pass filtered to accentuate the initial phase of the rod photocurrent. This process begins in the rod from the conversion of the outer-segment photocurrent into the inner-segment photovoltage, continues at the synapse between rods and RBCs, and is further achieved within synapses of the inner plexiform layer. Transformations at each stage can be reasonably well explained by linear filters common across different mouse lines, although contributions of nonlinear processing are particularly evident when responses were slowed considerably in the GCAPs^−/− retina. This speeding of the single-photon responses in rods results in a temporally-confined increase in action-potential firing in ON α ganglion cells, which is remarkably similar for all three mouse lines despite large differences in rod photocurrent response kinetics. Our findings reveal that much of the temporal transformation of the slow rod photocurrent occurs initially in its conversion to a photovoltage within the rod, but that further processing downstream in the retina is necessary to produce photoresponses that match the temporal bandwidth of the retinal output. 

Methods

Mouse Lines

WT mice were C57BL6/J obtained from the Jackson Laboratory (Bar Harbor, ME, USA). Mice lacking the genes for guanylyl cyclase activating proteins 1 and 2 (GCAPs^−/−) were initially characterized by Mendez et al.¹⁶ and were generously provided to us by Jeannie Chen of the Keck School of Medicine, the University of Southern California. Rhodopsin hemizygous mice (Rh^+/−) were originally described by Lem et al.¹⁸ and generously provided by Janice Lem and colleagues of the Tufts Medical Center. Although the reason for acceleration of response rise and decay in these mice is controversial and may depend on outer segment dimensions¹⁹ or the strain of the mouse, our only consideration in using these mice was that the response waveform was speeded (Figs. 1 and 2) and not the cause of this effect. 

Figure 1.

 

Representative retinal response families from WT (black), Rh^+/− (red), and GCAPs^−/− mice (blue). (A) WT rod outer-segment photocurrent responses to flashes of 0.43, 1.4, 4.1, 19, 58, and 180 Rh*/rod. (B) WT rod bipolar-cell photocurrent response (V_m = −60 mV) to flashes of 0.084, 0.25, 0.59, 1.3, 2.6, 5.3, and 11 Rh*/rod. (C) WT ON α ganglion cell photocurrent response (V_m = −60 mV) to flashes of 0.0024, 0.0049, 0.0097, 0.020, 0.039, 0.078, and 0.16 Rh*/rod. (D) Rh^+/− rod outer segment photocurrent responses to flashes of 3.3, 6.5, 12, 25, 50, and 95 Rh*/rod. (E) Rh^+/− rod bipolar cell photocurrent response (V_m = −60 mV) to flashes of 0.20, 0.50, 1.2, 2.4, 4.8, 10, and 20 Rh*/rod. (F) Rh^+/− ON α ganglion cell photocurrent response (V_m = −60 mV) to flashes of 0.0011, 0.0022, 0.0045, 0.0090, 0.018, 0.035, and 0.070 Rh*/rod. (G) GCAPs^−/− rod outer segment photocurrent responses to flashes of 0.34, 0.76, 1.9, 4.0, 8.3, and 15 Rh*/rod. (H) GCAPs^−/− rod bipolar cell photocurrent response (V_m = −60 mV) to flashes of 0.084, 0.25, 0.59, 1.3, 2.6, and 5.3 Rh*/rod. (I) GCAPs^−/− ON α ganglion cell photocurrent response (V_m = −60 mV) to flashes of 0.0024, 0.0049, 0.0097, 0.020, 0.039, 0.078, and 0.16 Rh*/rod. (J) Response-intensity relationships of rod photoreceptors for WT (black; fit I_1/2 = 8.8 Rh*/rod; n = 9), Rh^+/− (red; fit I_1/2 = 19 Rh*/rod; n = 9), and GCAPs^−/− mice (blue; fit I_1/2 = 2.9 Rh*/rod; n = 9). (K) Response-intensity relationships of rod bipolar cells for WT (black; fit I_1/2 = 1.8 Rh*/rod; n = 22), Rh^+/− (red; fit I_1/2 = 4.2 Rh*/rod; n = 32), and GCAPs^−/− mice (blue; fit I_1/2 = 1.3 Rh*/rod, n = 10).

Figure 2.

 

Derived elementary response per absorbed photon from rods, rod bipolar cells, and ON α retinal ganglion cells from WT (black), Rh^+/− (red), and GCAPs^−/− (blue) mice. See Material and Methods for procedures for calculating elementary responses across cell types. (A–C) Derived elementary responses from the WT rod outer segment photocurrent, WT rod bipolar cell photocurrent (V_m = −60 mV), and the WT ON α ganglion cell excitatory current (V_m = −60 mV). (D) WT responses from A–C normalized to compare the waveform. Rod photoresponses have been inverted for this comparison; most slowly decaying waveform is rod, dotted is bipolar cell, and bolded is ganglion cell. (E–G) Derived elementary responses from the Rh^+/− rod outer segment photocurrent, Rh^+/− rod bipolar cell photocurrent (V_m = −60 mV), and the Rh^+/− ON α ganglion cell excitatory current (V_m = −60 mV). (H) Rh^+/− responses from E–G normalized to compare waveform as in (D). (I–K) Derived elementary responses from the GCAPs^−/− rod outer segment photocurrent, GCAPs^−/− rod bipolar cell photocurrent (V_m = −60 mV), and the GCAPs^−/− ON α retinal ganglion cell excitatory current (V_m = −60 mV). (H) GCAPs^−/− responses from E–G normalized to compare waveform as in (D). (M) Comparison of the photocurrent between WT (black), Rh^+/− (red), and GCAPs^−/− (blue) rods. (N) Comparison of the photocurrent between WT (black), Rh^+/− (red), and GCAPs^−/− (blue) rod bipolar cells. (O) Comparison of the excitatory input to ON α retinal ganglion cells between WT (black), Rh^+/− (red), and GCAPs^−/− (blue). (P– R) Same as (M–O) but after normalization of peak amplitudes to unity.

Preparation and Physiological Recordings

Mice were dark-adapted overnight and euthanized according to protocols approved by the Institutional Animal Care and Use Committee of the University of Southern California (Protocol 10890), as described previously.^20,21 Briefly, mouse eyes were isolated, hemisected under infrared illumination, and stored in a light-tight container suspended in bicarbonate-buffered Ames’ media (A1420; Sigma-Aldrich Corp., St. Louis, MO, USA) equilibrated with 5% CO₂/95% O₂ at 32°C. At the time of use, the retina was gently removed from the eyecup and prepared for physiological recordings. The recording chamber solution in all experiments was bicarbonate-buffered Ames’ medium (A1420; Sigma-Aldrich Corp.) equilibrated with 5% CO₂/95% O₂ and held at 35°C to 37°C using a temperature controller (Warner Instruments, Hamden, CT, USA). 

Rod photocurrents were measured with suction electrodes from finely chopped pieces of retinal tissue. Clusters of cells with the outer segments protruding were targeted, and individual rod outer segments were drawn gently into a suction electrode containing Ames’ media buffered with 10 mM HEPES adjusted to pH 7.4. Light-evoked currents were measured following 10-ms flashes from an LED (λ_max ∼470 nm with full width at half maximum [FWHM] ∼30 nm), or 30 ms flashes from tungsten-halogen source passed through an interference filter (λ_max ∼500 nm, FWHM ∼15 nm). Responses were low-pass filtered at 20 Hz with an eight-pole Bessel filter and digitized at 1 kHz. 

The rod photovoltage and light-evoked currents from RBCs and retinal ganglion cells were recorded with patch electrodes from dark-adapted retinal slices or whole mounts as previously described.²⁰ ON α retinal ganglion cells were identified from their large size and response characteristics see.¹³ The internal solution for these experiments consisted of (in mM): 125 K-Aspartate, 10 KCl, 10 HEPES, 5 NMG-HEDTA, 0.5 CaCl₂, 1 ATP-Mg, and 0.2 GTP-Mg; pH was adjusted to 7.2 with N-methyl glucamine hydroxide. For the perforated patch recordings of Figure 3, 0.5 mg/mL of the ionophore amphotericin B (A9528; Sigma-Aldrich Corp.) was solubilized and added to the internal solution. Recordings were established by filling the tip of the electrode with plain internal solution and backfilling with internal solution containing Amphotericin B. Response families were measured following 10 ms flashes from a blue LED (λ_max ∼470 nm, FWHM ∼30 nm); the strength of the dimmest flash was set to the intensity required to produce a just-measurable response, and flash strengths were then increased from that value by factors of 2. Membrane currents were low-pass filtered at 300 Hz with an eight-pole Bessel filter and digitized at 1 kHz. 

Figure 3.

 

Rod photovoltages from WT (black), Rh^+/− (red), and GCAPs^−/− (blue) mice. (A) WT rod photovoltage response families for flashes of 2.0, 4.4, 9.3, 19, 39, 80, and 160 Rh*/rod. (B) Rh^+/− rod photovoltage response families for flashes of 2.3, 5.6, 12, 25, 51, 110, and 220 Rh*/rod. (C) GCAPs^−/− rod photovoltage response families for flashes of 0.72, 2.4, 7.3, 31, and 60 Rh*/rod. (D) Normalized response-intensity relationships of rod photovoltages for WT (black; fit I_1/2 = 9.5 Rh*/rod; n = 29), Rh^+/− (red; fit I_1/2 = 17 Rh*/rod; n = 16), and GCAPs^−/− mice (blue; fit I_1/2 = 3.9 Rh*/rod; n = 18). (E) WT rod voltage per photon derived from 555 dim flashes averaging 2.6 Rh*/rod across 29 rods. (F) Rh^+/− rod voltage per photon derived from 310 dim flashes averaging 2.0 Rh*/rod across 16 rods. (G) GCAPs^−/− rod voltage per photon derived from 250 dim flashes averaging 1.1 Rh*/rod across 18 rods. (H) Comparison of the rod photovoltage between WT (black), Rh^+/− (red), and GCAPs^−/− (blue) rods. (I) Elementary rod photovoltages normalized to unity peak amplitude. (J) Superimposed rod photocurrent from Figure 2 (solid), rod photovoltage (dashed), and rod bipolar cell photocurrent (bold) from the WT retina. Note similarity in the time-to-peak of the rod photovoltage and rod bipolar cell photocurrent. (K) As for J but from Rh^+/− retina. (L) As for J but from GCAPs^−/− retina.

Calculation of Rod, RBC, and Retinal Ganglion Cell Elementary Responses

Comparison of the timing of the dim flash response as it progresses through the retinal circuitry requires an estimation of the elementary response per photon at each stage. We estimated the elementary response per photon using procedures described in Sampath et al.⁶ Briefly, dim flash responses in a fixed amplitude range (i.e., larger than noise fluctuations but within the linear range) were averaged and then divided by the average flash strength used to evoke those responses. The elementary rod photocurrent was estimated from responses ranging from 5% to 25% of the maximum amplitude, the elementary rod photovoltage from responses ranging from 5% to 30% of the maximum amplitude, the elementary RBC photocurrent from responses ranging from 5% to 25% of the maximum amplitude, and the elementary ON α ganglion cell excitatory current from responses ranging from 3% to 10% of the maximum amplitude. The effective number of cells generating the average (see Table) was then determined by dividing the total trials across all cells by the number of trials for the cell from which the greatest number was recorded. This process effectively normalizes the contribution of each cell to the cell with the most trials. 

Table.

 

The Response Properties Of Each Cell Type In Three Different Mouse Lines

 

Extraction of Linear Filter

We studied signal transmission across the retina analytically by estimating the parameters of a linear filter required to transform a response from one stage to another in WT retina. We then applied this filter to the responses of Rh^+/− and GCAPs^−/− mice to establish the extent to which a simple linear transformation could account for response propagation in these circuits. Because current recordings were inherently noisy, we first estimated linear response parameters from fits to WT responses for a cascade of Poisson processes with four stages,²² 

\begin{eqnarray}{{r}^{\rm{*}}}\left( t \right) = i{{S}_f}{{T}^3}{{e}^{3\left( {1 - T} \right)}}\end{eqnarray}

(1)

where r*(t) is the linear response before allowing for compression by saturation, i is the flash intensity (photons µm⁻²), S_f is the flash sensitivity (pA photons⁻¹ µm²), and T is normalized time (T = t/t_peak where t_peak is the time at which the response reaches its peak amplitude). We then computed (for example between rod photocurrent and rod photovoltage) the linear (time-invariant) filter function H(x) satisfying  

\begin{eqnarray}Ro{{d}_V} = H{{\left( t \right)}_1}\,{\rm{*}}\, Ro{{d}_I}\end{eqnarray}

(2)

 

We did this by taking the Fourier transform of the fits to (Equation 1) of both Rod_V and Rod_I and then dividing their transforms. The result was then converted into the time domain and is shown as inserts next to the WT responses in Figure 6A. Similar filter functions H(t)₂ and H(t)₃ were calculated for the transformations from rod voltage to rod bipolar-cell current (Fig. 6D), and from rod bipolar-cell current to ganglion cell response (Fig. 6G). 

For the Rh^+/− and GCAPs^−/− photoreceptors, we first fit the waveform (for example that of the rod photovoltage) to (Equation 1), took its Fourier transform, multiplied by the appropriate filter function derived from WT responses (for example H(t)₁), and then took the inverse transform of the result. In doing so, we effectively applied the linear filter transformation occurring in WT retina for the WT response and applied it to the mutant responses. 

Results

Responses of Photoreceptors

To establish how single-photon responses are transformed by the retinal circuitry, we studied three lines of mice that displayed variations in the magnitude and time course of rod photocurrents (see Methods). We used suction-pipette recordings to measure the photocurrent from rod outer segments in response to a series of flashes ranging in strength from those barely evoking a response to those that saturate the photocurrent. Figures 1A, 1D, and 1G show families of rod photocurrents in WT, Rh^+/−, and GCAPs^−/− mice. As previously reported, response kinetics were accelerated in Rh^+/−15 and slowed in GCAPs^−/−.^16,17 Response-intensity relationships shown in Figure 1J reveal that sensitivity of Rh^+/− rods was roughly twofold lower than WT, as expected for rods expressing half the Rh compared to WT; and the response sensitivity of GCAPs^−/− was about threefold higher than WT (I_1/2, Table). 

The population mean single-photon response estimated from dim-flash responses in these families also agreed with previous studies (Figs. 2A, 2E, and 2I). The WT single-photon response reached its peak at ∼210 ms and recovered by ∼1 second. The Rh^+/− single-photon response (computed by assuming a collecting area half that the collecting area of WT, see reference 23) displayed a faster rising phase peaking at ∼160 ms and recovered quickly with a peak amplitude comparable to that of WT rods. The GCAPs^−/− photoresponses displayed an initial trajectory similar to WT but instead peaked near 500 ms owing to the lack of GCAPs-mediated Ca²⁺-dependent feedback to cGMP synthesis. This prolonged rising phase was accompanied by a 3.2-fold increase in the single-photon response peak amplitude compared to WT.^17,24 Kinetic differences are particularly evident when the responses have been normalized and superimposed (Fig. 2P). Kinetic parameters for the single-photon photocurrent responses are given in the Table (τ_int and T_peak). 

Responses of Bipolar and Ganglion Cells

To determine how these differences in rod photocurrent kinetics are reflected downstream in the retina, we recorded photocurrent responses in voltage clamp RBCs, and light-evoked excitatory currents of ON α ganglion cells in retinal slices and whole mounts. For both cell types we kept the holding potential at −60 mV, which for the ganglion cells would likely capture mostly excitatory synaptic input. Because both cell types pool the signals of many rods, we could not record single-photon responses but instead recorded responses at the dimmest intensity we could use for which responses were sufficiently above noise to be usable. The amplitude range of these responses used for calculating the elementary response are provided in the Material and Methods, and the mean amplitudes as a percentage of the maximum are provided in the Table. In addition, response-intensity relationships for RBC photocurrents for all three mouse lines are shown in Figure 1K, mirroring the sensitivity changes observed in the rod photocurrent. 

In WT mice, the RBC photocurrent at all intensities was accelerated from that of the rods (Fig. 1B), including at the level of the single-photon response (Fig. 2B). Consistent with previous studies,^6,24,25 the RBC dim flash response (Fig. 2B) peaked at ∼150 ms and returned to the baseline by ∼200 ms, and the integration time was markedly decreased from 540 ms in rods to 130 ms in RBCs (Table). In WT ON α ganglion cells, the family of photocurrents (Fig. 1C) and the dim-flash response kinetics (Fig. 2C) were further speeded compared to RBC photocurrents, but the difference was not as prominent as the difference between rods and RBCs. 

A similar effect was observed for RBCs in the Rh^+/− and GCAPs^−/− retinas at all flash intensities (Figs. 1E, 1H), as well as at the level of the single-photon response (Figs. 2F, 2J, Table). This speeding process was especially evident for the slow GCAPs^−/− responses. At dim intensities comparable to those producing single-photon responses in the rods, the time-to-peak of the GCAPs^−/− RBC response was close to the WT (150 ± 7.0 ms in WT vs. 190 ± 13 ms in GCAPs^−/−) despite a 2.4-times longer time-to-peak in the GCAPs^−/− rod photocurrents (Table). For the ON α ganglion cells, the effects in Rh^+/− and GCAPs^−/− retinas were similar to those in WT for all flash intensities (Figs. 1F, 1I), with dim flash response even closer to the WT (Figs. 2G, 2K). Compared to the RBC photocurrent, the integration time of ON α ganglion cell dim flash responses were shortened by ∼60% in GCAPs^−/− as opposed to ∼50% in WT and Rh^+/−. 

To facilitate comparison of waveforms, we have superimposed responses from the three genotypes for rods (Fig. 2M), RBCs (Fig. 2N), and ON α ganglion cells (Fig. 2O). Comparison of these three figures shows that the differences in waveform are diminished from the rods to the RBCs, and then again between the RBCs and ON α ganglion cells. The result is that the waveforms of the ON α ganglion cell excitatory currents in the three different lines become largely overlapping (Fig. 2O), despite the large difference in the responses of the rods. This effect is mostly the result of a quickening of the decay of the response, caused by an apparent high-pass filtering as the signal moves through the different retinal layers. The progression of the signal for the different animal lines can be seen in Figures 2D, 2H, and 2L, where we have superimposed normalized response waveforms of rods (solid), RBCs (dashed), and ganglion cells (bold). For all three lines, the major quickening of the response occurs largely between the rods and RBCs, although for the GCAPs^−/− retina there is a pronounced further acceleration between the RBCs and the ON α ganglion cells. 

Speeding of Rod Photovoltage

Because rods are known to express a variety of voltage-dependent and Ca²⁺-activated conductances in their inner segments, which are known to alter the waveform of the rod photovoltage,²⁶ we thought it possible that some part of the speeding of the response between the rod and RBC photocurrents might occur at the conversion of the rod photocurrent to photovoltage. We therefore repeated the experiments of Figures 1 and 2 for the rods but recorded the photovoltage response in current clamp. These experiments were done with the perforated patch method²⁷ to preserve the voltage waveform for as long as possible, but even with perforated patch the voltage response tended to become slower as the recording progressed. We therefore monitored the dim-flash and saturating response shape during the recordings, and we used data only up to the point at which response kinetics began to slow. This procedure typically let us to collect five to 10 photovoltage families within two to three minutes of recording. 

In Figures 3A–C, we show photovoltage families from the three mouse lines. Response-intensity relationships for the three lines are shown in Figure 3D, which also mirror the sensitivity changes observed in the rod photocurrent (Fig. 1J). Because we had difficulty maintaining stable rod recordings, we made no attempt to record single-photon responses but instead show responses to dim flashes. Although the rod photovoltage at brighter intensities displayed an initial peak or nose and became highly nonlinear, we found for dimmer flashes that waveforms normalized to flash intensity are invariant for responses up to about 35% of maximum amplitude. We could therefore use these dim-intensity responses to estimate the waveform of the response to a single photon, which we show for the three lines in Figures 3E–G. To visualize the effect of the current-to-voltage conversion in the rods, we superimpose in Figures 3J–L the normalized rod photocurrents (solid lines), rod photovoltages (dashed lines), and RBC photocurrents (bold lines). The rod photovoltage in WT mice displayed a strikingly similar time course to the RBC photocurrent. The response peaked at ∼140 ms, close to the time-to-peak of WT RBC photocurrent, and the integration time was ∼210 ms, which accounted for ∼80% of the reduction of integration time between rod photocurrent and RBC photocurrent, as well as ∼70% of the total reduction of integration time between rod and ON α ganglion cell photocurrents (Table). A similar effect was observed for the Rh^+/− retina. For the GCAPs^−/− retina, however, the rod photovoltage was intermediate between the rod photocurrent and the RBC photocurrent. The time-to-peak was 300 ms, still much longer than the time-to-peak of the RBC photocurrent (190 ms), and the integration time was 700 ms, a value much larger than the integration time of the RBC photocurrent (290 ms). This finding reveals that some further process must exist that has a larger role in the GCAPs^−/− retina than for the other two genotypes. We return below in the Results (Fig. 6) and in the Discussion to the nature of the mechanisms effecting signal conversion at different stages of retinal integration. 

Ganglion-Cell Spike Output

The experiments described so far have shown that the responses of rods are high-pass filtered in their passage to RBCs predominantly by the conversion of outer-segment photocurrent to inner-segment photovoltage. A further stage of filtering occurs at the synapse between rods and RBCs, and then in the inner plexiform layer between the RBCs and ON α ganglion cells, especially evident for the GCAPs^−/− retina. As a consequence of these transformations, the ganglion-cell photocurrents in the three genotypes have remarkably similar waveforms (Fig. 2O) despite the very different photocurrents that generate them (Fig. 2M). 

One concern with this conclusion is that we have recorded the waveform of voltage-clamped ganglion-cell currents at a holding potential of −60 mV, likely to reflect mostly excitatory input. This current may not have a clear relationship to the spiking output of the ganglion cell. It is however this spiking output that is transmitted to the visual centers of the brain. We therefore made a more careful examination of the responses of the ON α ganglion cells to dim intensities. ON α ganglion cell spike responses were recorded with cell-attached loose-patch recording. The ON α ganglion cells could be identified by their large cell body size, high sensitivity to light, and sustained spikes in response to a light step (Fig. 4A).^28,29 After collecting a sufficient number of responses to dim flashes generating only a few spikes per trial (Fig. 4B), we replaced the pipette with a fresh patch pipette to record synaptic currents. 

Figure 4.

 

Recordings from WT ON α retinal ganglion cells. (A) Cell-attached recording from ON α ganglion cell showing sustained firing to a step of background light of 1 Rh* rod⁻¹ s⁻¹. (B) Raster plots showing the timing of action potentials fired after a brief flash of 0.044 Rh*/rod. Fifty sequential trials in cell-attached mode for flash delivered at t = 0. (C) Inhibitory currents (V_hold = 0 mV) were isolated for flashes of light of 0.0044, 0.0089, 0.018, 0.035, 0.071, and 0.14 Rh*/rod. Note that the dimmest flash (red) did not produce an outward current. (D) Excitatory currents (V_hold = −70 mV) for the same flash intensities; the dimmest flash (red) of 0.0044 Rh*/rod produced a small inward current. (E) Excitatory currents (V_hold = −70 mV) for a flash of 0.0044 Rh*/rod. Top trace (red) is the average excitatory current.

In Figures 4C and 4D, we show responses to increasing flash intensities at a holding potential of 0 mV to reveal inhibitory inputs (IPSCs), and at −70 mV to isolate excitatory EPSCs. When flash intensity was gradually increased from very dim flashes causing almost no response to brighter flashes, EPSCs began to emerge before IPSCs, and EPSC amplitudes were much larger than IPSC amplitudes at any given flash intensity, consistent with previous studies.^28,29 We then gave a series of dim flashes in the same cell (Fig. 4E), which the recordings of Figure 4B had shown to evoke about one or two spikes on average. Flashes at this intensity typically produced an EPSC with an amplitude less than 10% of the maximal EPSC. No IPSC was ever observed at this flash intensity. These results show that EPSCs are the dominant synaptic input influencing ON α ganglion cell spike outputs near absolute visual threshold. 

We collected dim flash EPSCs and spikes this way from multiple ganglion cells in WT, Rh^+/− and GCAPs^−/− mice, and we analyzed how different EPSC kinetics are translated into changes in spike rates. These results are shown in Figure 5. Normalized EPSC current waveforms (like those in Fig. 4E) are given for the three genotypes in Figures 5A, 5D, and 5G; and normalized ganglion-cell spiking rates are given in Figures 5B, 5E, and 5H. These waveforms are then superimposed in Figures 5J and 5K. Like the ganglion-cell currents previously described (Fig. 2O), the waveforms are very similar and nearly overlapping despite larger differences in the waveforms of the rod photoresponses. These recordings show moreover that the waveforms of the EPSCs and those of the averaged spike output are very similar for all three genotypes—see Figures 5C, 5F, and 5I. In these cases, ON α ganglion cells spikes fell within the envelope of their respective EPSCs. These results indicate that the high-pass filtering of single-photon responses through the primary rod pathway is critical for generating the fast and brief spike outputs of ganglion cells to improve the temporal resolution of vision even in very low light levels. 

Figure 5.

 

EPSCs and action potential histograms for WT (black), Rh^+/− (red), and GCAPs^−/− (blue) ON α ganglion cells. (A) Average normalized EPSC produced by WT ON α retinal ganglion cells (n = 8 cells) to a flash of on average 0.078 Rh*/rod at time = 0. The flash generated an excitatory input that was 7.0% of the maximum amplitude. (B) Average action potential histograms from the same set of WT ON α ganglion cells for the same flash strength. Histograms have been normalized to the peak amplitude. (C) Overlay of the inverse of the WT average EPSC and the WT average action-potential histogram. (D) Average normalized EPSC produced by Rh^+/− ON α ganglion cells (n = 7 cells) to a flash of on average 0.074 Rh*/rod at time = 0. The flash generated an excitatory input that was 4.6% of the maximum amplitude. (E) Average action-potential histograms from the same set of Rh^+/− ON α ganglion cells for the same flash strength. Histograms have been normalized to the peak amplitude. (F) Overlay of the inverse of the Rh^+/− average EPSC and the Rh^+/− average action-potential histogram. (G) Average normalized EPSC produced by GCAPs^−/− ON α ganglion cells (n = 6 cells) to a flash of on average 0.056 Rh*/rod at time = 0. The flash generated an excitatory input that was 8.4% of the maximum amplitude. (H) Average action-potential histograms from the same set of GCAPs^−/− ON α ganglion cells for the same flash strength. Histograms have been normalized to the peak amplitude. (I) Overlay of the inverse of the GCAPs^−/− average EPSC and the GCAPs^−/− average action-potential histogram. (J) Overlay of normalized EPSC for WT (black), Rh^+/− (red), and GCAPs^−/− (blue) ON α ganglion cells. (K) Overlay of action potential histograms for WT (black), Rh^+/− (red), and GCAPs^−/− (blue) ON α ganglion cells.

Linear and NonLinear Processing of Rod Signals Near Threshold

To gain further insight into the nature of the processing of the rod signals, we asked whether the filtering at successive stages in integration could be described by simple linear filters, or whether nonlinear processes were also essential.³⁰ We used Fourier methods to calculate a filter between each of the stages in a WT retina (see Materials and Methods) and then attempted to apply this same filtering for the two mutant lines. 

The method we used can be seen from the data in Figures 6A–C for the conversion of rod photocurrent to photovoltage. We first fit the photocurrent waveform (Fig. 6A, slowly decaying bold black line) to a cascade of Poisson processes with four stages (Equation 1).²² An attempt to use the waveform itself in our calculations was unsuccessful because of the intrinsic noise of the response. This fit is indicated in Figure 6A by the dotted black line superimposed on the current response and is difficult to see because of the goodness of fit. We then fit the photovoltage with (Equation 1) though with different parameters. The photovoltage and its fit are shown as the more rapidly decaying solid and dashed black lines in Figure 1A. From these two fits we used Fourier routines in MATLAB and (Equation 2) to calculate a linear filter that would convert the photocurrent to photovoltage. The waveform of this filter is shown as the inset to Figure 1A in the time domain. 

Figure 6.

 

Predictions of linear transformations of single-photon and dim-light responses for WT, Rh^+/−, and GCAPs^−/− mice at three stages in retinal processing. See Material and Methods for description of calculations. (A–C). Conversion of rod photocurrent to rod photovoltage. The photoreceptor single-photon currents (bold) were first fit with (Equation 1) of the Methods, which is a Poisson processes with four stages.²² These fits are shown as the dashed lines superimposed upon the photocurrent. The photovoltages (solid line) were then also fit to (Equation 1), and the fits for photovoltage and photocurrent of the WT rod were used to calculate a linear filter that would best describe the transformation of photocurrent to photovoltage for the WT rod (see Material and Methods, Equation 2). The waveform of this filter in the time domain is given in the inset to A. This filter was then convolved with the photocurrent responses in B and C to predict the rod photovoltage for Rh^+/− (red), and GCAPs^−/− (blue) rods, given as the more rapidly decaying dotted lines. (D– F). Same procedure as for A–C except that the calculation of the linear filter was based on the WT rod photovoltage and RBC photocurrent. (G–I). Same procedure as for A–C except that the calculation of the linear filter was based on the WT RBC and ON α ganglion photocurrents.

Having calculated this filter for WT rods, we then applied it to Rh^+/− and GCAPs^−/− rods. We fit both the photocurrent and photovoltage to (Equation 1) as before and show the fits only for currents in Figures 6B and 6C as the more slowly decaying dotted red and blue waveforms. Instead of calculating filters for the Rh^+/− and GCAPs^−/− rods, we used the WT filter from the inset of Figure 1A to attempt to convert the Rh^+/− or GCAPs^−/− currents into their photovoltages. The results of these calculations are shown as the more rapidly decaying dotted red and blue lines. The fits are good though not perfect, indicating that a simple linear filter is a reasonable approximation to the stage of integration occurring between the outer-segment current and the inner-segment voltage. Deviations from the fit must then be the result of additional non-linear processes, which appear to be more prominent for GCAPs^−/− than for Rh^+/−. 

We repeated this same procedure for the conversions of rod photovoltage to RBC photocurrent (Figs. 6D–F), and then RBC photocurrent to ON α ganglion cell photocurrent (Figs. 6G–I). Each stage required further high-pass filtering (insets to Figs. 6D, 6G), with the result that the signals became progressively more transient. Both the WT and Rh^+/− rods were surprisingly well fit by the same linear filter, although these fits were again not perfect. However, GCAPs^−/− retinas were again markedly different. Here the fits using the WT linear filter again did not capture the full extent of high-pass filtering observed in RBC (Fig. 6F) or ON α ganglion cell photocurrents (Fig. 6I), indicating that nonlinear contributions were also needed to account for the speeding of light responses. One possible explanation might be the much larger and slower GCAPs^−/− rod photocurrents (Fig. 2M); but when we repeated our WT fitting with brighter flashes to give responses comparable in amplitude to the single-photon responses of GCAPs^−/− rods, we were still not able to use the same WT linear filter to fit GCAPs^−/− RBC responses. It should be noted, however, that brighter WT responses of comparable amplitude to GCAPs^−/− responses would also display a speeded rising phase, which may engage rod conductances differently despite the controlling for response size (see Discussion). Thus some other sources of filtering probably with significant nonlinear contributions seem likely to be contributing to integration of GCAPs^−/− responses at all three stages in processing. These mechanisms are surprisingly effective at producing transient GCAPs^−/− ON α ganglion cell EPSCs that are quite similar to WT (Fig. 2C vs. Fig. 2K), and for spike output in Figure 5K. 

Discussion

We have made whole-cell, perforated-patch, and loose-patch recordings from the mouse retina to study the kinetic transformation of rod signals in dim light as they progress to the retinal output. We have recorded from rods, RBCs, and ON α ganglion cells from retinas of three lines of mice (Figs. 1, 2): WT, rhodopsin hemizygotes (Rh^+/−), and GCAPs knockouts (GCAPs^−/−). We show that these three lines have rod single-photon responses of different kinetic waveforms: Rh^+/− faster than WT and GCAPs^−/− much slower. We show that in all three lines, the rod responses become progressively more transient with more accelerated decay as signals propagate through the retina (Figs. 2D, 2H, 2L). Much of this acceleration occurs in the conversion of the rod outer-segment photocurrent to the inner-segment photovoltage. Our previous work indicates that the speeding is mediated by a K⁺ conductance, because it is prevented when using pipette internal solutions where K⁺ is substituted with Cs⁺ (see Fig. 5 of reference 31). Further evidence is required to establish whether this effect is generated by i_h,³² i_kx,³³ or another leak conductance. Beyond the rod photovoltage, response waveforms are also shaped at synapses between rods and RBCs and by integration within the inner plexiform layer. At all three stages we investigated, we found that for WT and Rh^+/− retinas, the transformations at each stage could be modeled with reasonable accuracy by linear high-pass filters (Fig. 6). For GCAPs^−/− retinas, however, these filters gave poorer fits at all three stages, indicating that components of nonlinear processing must also contribute even in dim light. 

The result of this high-pass processing is that action-potential firing of ON a ganglion cells is remarkably similar in all three retinas despite large differences in rod single-photon waveforms. This result is of particular significance, since recent evidence indicates that the ON α ganglion cells in mouse are the mammalian orthologs of midget ganglion cells in primates.³⁴ Thus the mechanisms of integration we are describing are likely of universal relevance among mammals including man. Moreover, much of the acceleration of rod responses occurs before the signals enter the inner plexiform layer where AII amacrine cells diverge between ON and OFF pathways (see Figs. 2, 3), suggesting that most if not all ganglion cells might display transient responses like the ON α ganglion cells.⁷ It would nevertheless be interesting to extend our work in future studies to other ganglion-cell types, and in particular to the OFF sustained cells which Westö and colleagues³⁵ have shown to be more sensitive than ON cells to dim stimuli near the absolute threshold of behavior. 

Comparison of the waveform of rod single-photon responses with the dim-flash responses of RBCs and ON α ganglion cells (Fig. 2) shows that the waveform of rod outer-segment photocurrents is accelerated with decreasing time-to-peak and accelerated time courses of decay see also.⁶ Our observations confirm previous work, which has shown that bipolar-cell responses decay more rapidly than rod responses,^5,6,36 and that further acceleration of the response can occur in the inner plexiform layer.^6,37–39 Further acceleration across the rod-to-RBC synapse may result from the action of the RGS proteins RGS7 and RGS11 in mGluR6 transduction^40,41; and in part from voltage-gated HCN channels in the RBCs,⁴² a depolarizing conductance which will be more engaged in GCAPs^−/− rods due to the larger and slower hyperpolarization per photon (Fig. 3H). In the inner plexiform layer, further acceleration could occur at synapses between RBCs and AII amacrine cells³⁹ but also at synapses with ON and OFF bipolar cells and their respective ganglion cells. Little is known at present about the relative roles of these different areas of synaptic contact in setting the dynamic parameters of the scotopic response in dim light. 

We have made the surprising discovery that retinas whose rods have very different waveforms of single-photon responses show a similar operation of filtering within the retina that is sufficiently strong ultimately to produce ganglion-cell responses with strikingly similar waveforms. This observation is important, because it demonstrates that even in very dim light, the retina and visual system prioritize transient responses to changing illumination. We are accustomed to the importance of sensitivity to movement and change in brighter light for rods and for cones; our observations show that this sensitivity extends to even the dimmest illumination where we were able to make measurements. This sharpening of the responses has been suggested to improve the temporal fidelity of rod vision in two ways. First, the high-pass filtered photoresponses provide more accurate information about the timing of photon absorptions.⁵ Second, speeded responses increase temporal resolution at light levels where single-photon responses in individual rods begin to overlap in time.^43,44 

Our results confirm and extend those of Umino and colleagues,⁴³ who studied behavior in mice with mutations resulting in overexpressed or no expression of the R9AP GAP proteins. The photoreceptors of these animals have responses with slower or more rapid response decay, which nevertheless have no significant effect on the behavioral temporal-contrast sensitivity of rod vision.⁴⁴ Umino and colleagues inferred from their observations that only the initial phase of the rod response is communicated to the rest of the retina and nervous system to influence detection. We now show this result explicitly for the transformation of the dim-light rod signal from the single-photon outer-segment current to the ganglion-cell spike output. 

Our findings raise the question of why there is so little variability in the waveform of decay of single-photon rod responses. This decay has been shown by numerous studies to be reduced by multiple stages in rhodopsin phosphorylation⁴⁵ and other mechanisms including Ca²⁺ regulation of cGMP synthesis.^46,47 It is of course possible, as Umino and colleagues⁴³ have suggested, that this suppression of variability of decay is only important for the accumulation of several single-photon responses when single rods are receiving multiple photon absorptions. This part of scotopic vision is, however, rather small by comparison to the larger range of dim illuminations in which rod signals are pooled, and the probability of multiple hits within a single rod is low. Our results show that even at the very first synapse, the decay phase of the rod single-photon response is no longer transmitted to the rest of the retina. We are left wondering about the nature of the selection pressure that has resulted in the decrease of the variability of response decay, when this part of the rod response seems to have so little effect on the retinal output over most of the scotopic range. 

Acknowledgments

The authors thank Gordon Fain for editorial assistance and critical review of the manuscript. GCAPs^−/− mice were provided by Jeannie Chen (University of Southern California), and Rh^+/− mice were provided by Janice Lem (Tufts Medical Center). 

Supported by NIH grant EY17606 (APS), the McKnight Endowment Fund for Neuroscience (APS), an unrestricted grant from Research to Prevent Blindness to the UCLA Department of Ophthalmology, and the Stein Eye Core Grant (P30 EY00331). 

Disclosure: H. Okawa, None; A.P. Sampath, None