Optimal processing of photoreceptor signals is required to maximize behavioural sensitivity

Abstract

The sensitivity of receptor cells places a fundamental limit upon the sensitivity of sensory systems. For example, the signal-to-noise ratio of sensory receptors has been suggested to limit absolute thresholds in the visual and auditory systems. However, the necessity of optimally processing sensory receptor signals for behaviour to approach this limit has received less attention. We investigated the behavioural consequences of increasing the signal-to-noise ratio of the rod photoreceptor single-photon response in a transgenic mouse, the GCAPs-/- knockout. The loss of fast Ca2+ feedback to cGMP synthesis in phototransduction for GCAPs-/- mice increases the magnitude of the rod single-photon response and dark noise, with the increase in size of the single-photon response outweighing the increase in noise. Surprisingly, despite the increased rod signal-to-noise ratio, behavioural performance for GCAPs-/- mice was diminished near absolute visual threshold. We demonstrate in electrophysiological recordings that the diminished performance compared to wild-type mice is explained by poorly tuned postsynaptic processing of the rod single-photon response at the rod bipolar cell. In particular, the level of postsynaptic saturation in GCAPs-/- rod bipolar cells is not sufficient to eliminate rod noise, and degrades the single-photon response signal-to-noise ratio. Thus, it is critical for retinal processing to be optimally tuned near absolute threshold; otherwise the visual system fails to utilize fully the signals present in the rods.

Figure 2. Behavioural estimate of absolute visual threshold in WT and GCAPs^−/− mice

A, escape time from the maze plotted versus light intensity in equivalent photons at the peak absorption wavelength of mouse rhodopsin (501 nm; see Methods), with total darkness plotted at an intensity of 0 ϕμm⁻² s⁻¹. Mean ±s.e.m. at each intensity are plotted for WT mice (n= 8; black symbols) and GCAPs^−/− mice (n= 8; grey symbols). The mean and the standard error are collectively best fit with a Hill equation where the parameters were allowed to vary. In WT mice the Hill parameters were I_1/2= 0.0085 ϕμm⁻² s⁻¹ and n= 2.2, and in GCAPs^−/− mice were I_1/2= 0.013 ϕμm⁻² s⁻¹ and n= 2.9. B, differences in escape behaviour were more pronounced near behavioural threshold which occurs in the transition from high light levels where mice can see the target and escape quickly, to low light levels where mice cannot see the target and search randomly. To facilitate the comparison of escape times at these light levels the normalized time to escape on a mouse-by-mouse basis is plotted at 0.00073 (low), 0.0073 (transition) and 1.45 (high) photons μm⁻² s⁻¹, as indicated by the boxes in A. At the transition background light level of 0.0073 photons μm⁻² s⁻¹ the difference in escape times was statistically significant between WT and GCAPs^−/− mice, P= 0.001 *Student t test P < 0.01. C, the probability that the behavioural data collected from GCAPs^−/− mice could be explained by the WT data was estimated by bootstrapping (see Methods). A histogram plotting I_1/2 values from 100 sets of WT behavioural curves generated by sampling with replacement is shown. The histogram was fitted with a Gaussian distribution that has a mean of 0.0086 photons μm⁻² s⁻¹ and a standard deviation of 0.0013 photons μm⁻² s⁻¹. The best-fit I_1/2 for the GCAPs^−/− behavioural data has a value of 0.013 photons μm⁻² s⁻¹, 3.4 standard deviations above the WT data mean. This reflects a probability that the GCAPs^−/− behavioural data could be explained by the WT data of P < 0.005.

Figure 1. Light-evoked responses of WT and GCAPs^−/− rod photoreceptors

A, response families to 30 ms flashes (arrow) WT and GCAPs^−/− rods. Flash strengths for the WT rod yielded 0.43, 1.4, 4.1, 19, 58 and 180 Rh*, and for the GCAPs^−/− rod yielded 0.34, 0.76, 1.9, 4.0, 8.3 and 15 Rh*. B, derived single-photon responses from WT and GCAPs^−/− rods plotted as a fractional suppression of the dark current versus time. The WT response was derived from 2137 trials across 9 cells. The GCAPs^−/− response was derived from 4280 trials across 23 cells. The effective number of cells was estimated from the total number of trials divided by the number of trials on the cell with the most, and thus can be non-integer. The time-to-peak of the single-photon response was 208 ± 11 ms (n= 4.1) in WT rods and 503 ± 39 ms (n= 18.4) in GCAPs^−/− rods (mean ±s.e.m.). The integration time of the averaged response was 540 ms for WT rods and 900 ms for GCAPs^−/− rods. C, individual trials to flashes of fixed strength in WT and GCAPs^−/− rods. In each epoch a flash was delivered (up triangle) at 0.2 s and on average generated 0.9 Rh* in WT rods, and 1.7 Rh* in GCAPs^−/− rods. The average single-photon response was shown as the right-most epoch (grey) for comparison. D, the SNRs for WT and GCAPs^−/− rods were determined from the fits of histograms to the data in C (see Methods for details). Note the difference in the abscissa for WT and GCAPs^−/− rods.

Figure 8. Comparison of SNRs of rods and rod bipolar cells

SNRs as determined for rods and rod bipolar cells (see Methods). Individual cells are plotted as open symbols, with the mean and s.e.m. included as filled symbols. As shown in Fig. 1 the SNR for WT rods was 1.69 ± 0.15 (mean ±s.e.m.; n= 4) compared to GCAPs^−/− rods where the SNR was 2.51 ± 0.08 (mean ±s.e.m.; n= 6), P= 0.0054. As shown in Fig. 4 the SNR for WT rod bipolar cells was 2.91 ± 0.11 (mean ±s.e.m.; n= 5) compared to GCAPs^−/− rod bipolar cells where the SNR fell to 2.49 ± 0.06 (mean ±s.e.m.; n= 16), P= 0.015. In GCAPs^−/− rod bipolar cells treated with 0.2 μm APB the SNR increased to 2.94 ± 0.17 (mean ±s.e.m.; n= 7), P= 0.041 compared to GCAPs^−/− (Fig. 7). Variability in the SNR may result from variability in individual cells, or from retinal slice to retinal slice. For GCAPs^−/− rod bipolar recordings with and without APB that were done in the same slice, we compared the SNR by plotting a line between the mean values for both conditions. When more than one cell from each condition were derived from the same retinal slice, the line extended to the average value from those cells. Note the upward trend of every line to indicate an improvement in rod bipolar cell SNR in the presence of APB. *P < 0.01, **P < 0.05.

Figure 3. Flash response families in WT and GCAPs^−/− rod bipolar cells

A, light-evoked responses to 10 ms flashes (arrow) for voltage-clamped (V_m=−60 mV) rod bipolar cells. For the WT rod bipolar cell flash strengths were 0.084, 0.25, 0.59, 1.3, 2.6, 5.3 and 11 Rh*/rod, and for the GCAPs^−/− rod bipolar cell flash strengths were 0.084, 0.25, 0.59, 1.3, 2.6, 5.3 Rh*/rod. A vertical dashed line demonstrates the initial time to peak of the rod bipolar responses are similar between WT (145 ± 7.7 ms, n= 8.2) and GCAPs^−/− (169 ± 19 ms, n= 5.8) rod bipolar cells (mean ±s.e.m.) despite the slowed response of GCAPs^−/− rods (see Fig. 1). B, average response–intensity relationship across all WT (n= 22) and GCAPs^−/− (n= 10) rod bipolar cells. Data were fitted will a Hill equation with n= 1.55 and I_1/2= 1.8 Rh*/rod for WT rod bipolar cells, and n= 1.25 and I_1/2= 1.3 Rh*/rod for GCAPs^−/− rod bipolar cells.

Figure 4. Estimation of SNR in WT and GCAPs^−/− rod bipolar cells

A, perforated-patch voltage-clamp (V_m=−60 mV) recordings from WT and GCAPs^−/− rod bipolar cells. Each epoch represents an individual trial where a 10 ms flash (inverted triangle) was delivered at 0.2 s, and the grey records reflect the shape of the average dim flash response and are scaled for comparison with individual trials. Flashes for the WT and GCAPs^−/− rod bipolar cell on average generated 0.064 Rh*/rod. B, SNRs were determined from the fits of histograms generated from the data in A (see Methods). For ease of comparison amplitudes have been normalized to the mean single-photon response.

Figure 5. Dark noise in WT and GCAPs^−/− rod bipolar cells

A, noise variance was measured in voltage-clamped rod bipolar cells immediately after whole-cell break-in (black) and after the dialysis of the cell with 25 μm GTP-γ-S (grey). The average change in holding current following GTP-γ-S dialysis for WT mice was 10.2 ± 0.7 pA (mean ±s.e.m.; n= 10) and in GCAPs^−/− mice was 15.2 ± 2.2 pA (mean ±s.e.m.; n= 11), P= 0.045. B, power spectra of dark noise fluctuations for WT (filled symbols) and GCAPs^−/− (open symbols) rod bipolar cells. Power spectra after the dialysis of GTP-γ-S (grey filled and open symbols) was similar indicating that higher noise in GCAPs^−/− rod bipolar cells was not due to differences in the quality of the recordings.

Figure 6. Model for the non-linear threshold at the rod-to-rod bipolar synapse

A, the probability density of dark noise and single-photon (light) response amplitude in WT is plotted for conditions near absolute visual threshold (∼0.0005 Rh*). Standard deviations for Gaussian distributions were averaged and have been normalized to the mean single-photon response amplitude for WT rods (, ). The dashed vertical lines indicate the position of the non-linear threshold, plotted at 1.3 for WT (Field & Rieke, 2002b). B, the probability density for GCAPs^−/− rods near absolute visual threshold, as for WT rods. Gaussian distributions were averaged and have been normalized to the mean single-photon response amplitude (, ) that were multiplied by 3.4-fold difference in the mean single-photon response amplitude compared to WT rods (see Fig. 1). Non-linear thresholds were plotted at 1.3 (same as WT), 1.6 and 2.7 pA. C and D, rod bipolar response distributions were simulated assuming the convergence of 20 rods per rod bipolar cell and flash strengths of 0.1 Rh*/rod and 0.04 Rh*/rod for WT and GCAPs^−/− distributions, respectively. We assumed a sharp threshold, where responses exceeding the threshold were transmitted and those below the threshold were scaled by 0.15, to reproduce the observed variance in dark noise of the rod bipolar cells in Fig. 4. Simulated histograms resulting from 10,000 repeated flashes were fitted as the sum of Gaussian distributions centred at the absorption of 0, 1, 2 … photons (see Methods), and the response amplitude axis was normalized to the mean single-photon response amplitude.

Figure 7. APB improves SNR in GCAPs^−/− rod bipolar cells

A, perforated-patch voltage-clamp (V_m=−60 mV) recordings from GCAPs^−/− rod bipolar cells in the presence of 0.2 μm APB. Each epoch represents an individual trial where a 10 ms flash (inverted triangle) was delivered at 0.2 s, and the grey records at the right are scaled to the average single-photon response amplitude for comparison. Flashes for these rod bipolar cells on average generated 0.033 Rh*/rod. We assume the APB-dependent increase in synaptic saturation at the rod bipolar cell to be predominantly a postsynaptic action, as the activation of presynaptic metabotropic glutamate receptors at the rod spherule would reduce glutamate release and relieve saturation, the opposite of the effect observed (see also Sampath & Rieke, 2004). B, SNRs were determined from the fits of histograms generated from the data in A (see Methods). For ease of comparison with Fig. 4 amplitudes have been normalized to the mean single-photon response.