Functional-pathway-dominant contrast adaptation and sensitization in mouse retinal ganglion cells

Abstract

Retinal ganglion cells (RGCs) reduce their light sensitivity during persistent high-contrast stimulation to prevent saturation to strong inputs and improve coding efficiency. This process is known as contrast adaptation. However, contrast adaptation also reduces RGCs’ light response to weak inputs. On the other hand, some RGCs undergo contrast sensitization, and these RGCs respond to weak inputs following high contrast stimulation. In the present study, multi-electrode recordings were conducted on isolated mouse retinas under full-field visual stimulation with different contrast levels. Adaptation and sensitization were mainly observed in OFF and ON pathways, respectively. The results of linear–nonlinear analysis and stimulus reconstruction revealed that both the light sensitivity and encoded information were changed in opposite directions in adaptation and sensitization processes. Our work suggests that contrast adaptation and sensitization are two opposite dynamic processes. In mouse retina, OFF RGCs utilize adaptation to increase the discrimination of strong OFF inputs. On the other hand, ON RGCs use sensitization to increase the sensitivity to weak ON inputs. This functional differentiation might be meaningful for the mouse’s survival as it lives in environments in which strong OFF stimuli often indicate potential predators while weak ON stimuli are usually related to movement and might be important for predation.

Introduction

Luminance in the natural environment changes within a 10¹¹-fold range, from about 10⁻⁶ cd/m² in darkness to 10⁵ cd/m² in full sunlight (Reinagel 2001). However, neuronal activity only covers a narrow operating range of no more than 10²-fold (Demb and Singer 2015). Therefore, the visual neural system must continuously adjust its responsiveness to match the dynamic input. To achieve this, a number of strategies are involved, which includes visual adaptation to ambient luminance and contrast. It has been well observed that neurons fire with relatively high rates at the onset of high-contrast stimulation and activity gradually adapts to persistent stimulation, preventing saturation to strong inputs (Demb 2008) thus increasing coding efficiency (Durant et al. 2007; Jin et al. 2005). However, this adaptation also makes neurons less sensitive to weak inputs (Baccus and Meister 2002; Brown and Masland 2001; Manookin and Demb 2006). Indeed it has been reported that in some retinal neurons, another dynamic process characterized by sensitization maintains the neurons’ ability to respond to weak inputs (Kastner and Baccus 2011). In contrast to adaptation, sensitizing neurons fire with relatively low rates at the onset of high-contrast stimulation and the firing activities are gradually increased during persistent stimulation. Correspondingly, the firing rates are relatively high at the onset of low-contrast stimulation and decrease gradually during persistent low-contrast stimulation.

Till now, there are only a few studies about sensitization in retinal neurons (Appleby and Manookin 2019; Kastner and Baccus 2011, 2013; Kastner et al. 2019; Nikolaev et al. 2013). The basic properties of sensitization and its functional contribution to retinal information coding are poorly understood. In the present study, mouse retinal ganglion cells (RGCs) were classified into adaptation and sensitization groups, based on the modulation of firing rate under sustained full-field Gaussian contrast stimulation. It was found that adaptation and sensitization were basically segregated respectively to OFF- and ON-center RGCs, while in ON–OFF RGCs, adaptation and sensitization were restricted mostly to the OFF- and ON-response phases, respectively. The results of linear–nonlinear analysis and stimulus reconstruction revealed that, during persistent high-contrast stimulation, both the average light sensitivity of the retinal network and stimulus information encoded in RGC spike trains were reduced during adaptation but enhanced during sensitization; and the opposite was true for the persistent low-contrast stimulation. Our work suggests that sensitization is another dynamic process in mouse retina which is opposite to adaptation, making the retina more efficient when exposed to fast-changing inputs.

Methods

Electrophysiology

Multi-electrode recordings (Puyang et al. 2017; Yan et al. 2016) were performed on isolated mouse retina. C57BL/6 male mice of 6–8 weeks old were dark adapted for 30 min before the experiment and sacrificed by cervical dislocation under dim red light. Eyes were enucleated and corneas were cut. Lens and vitreous were removed with tweezers, then each eye-cup was cut into two parts and the neural retina was carefully separated from the pigment epithelium. The isolated retina was placed on a piece of filter paper (0.22 µm pore size, White GSWP, Millipore Corporation, Bedford, MA, USA), with the photoreceptor side contacting the filter paper. All procedures were conducted in oxygenated (95% O₂ and 5% CO₂) Ringer’s solution (124.0 mM NaCl, 2.5 mM KCl, 1.3 mM NaH₂PO₄, 2.0 mM CaCl₂, 2.0 mM MgCl₂, 22.0 mM glucose, 26.0 mM NaHCO₃ and pH = 7.4). The mounted retina was immediately transferred onto a 60-channel multi-electrode array (MEA, electrode diameter 10 µm, spacing 100 µm; Multi Channel Systems MCS GmbH, Reutlingen, Germany), with the ganglion cell layer facing the electrodes and continuously perfused with oxygenated Ringer’s solution at 37 °C. The Ringer’s solution was pumped into a perfusion chamber by a tubing pump (REGLO Digital MS-4/8, Ismatec, Switzerland), with a flow rate of 0.49 ml/min. The volume of the perfusion chamber was 0.6 ml. Animal procedures were approved by Shanghai Jiao Tong University Institutional Animal Care and Use Committee (Animal Protocol: A2016007).

Raw electrode data were amplified through a 60-channel amplifier (single-ended, amplification 1200$\times$, amplifier input impedance > 10¹⁰ Ω, output impedance 330 Ω). Signals were sampled at a rate of 20 kHz (MC_Rack, Multi Channel Systems MCS GmbH, Reutlingen, Germany) for further analyses.

Visual stimulation

The light stimulus was generated by an organic light emitting display (OLED-XL, eMagin Corporation, California, USA), and focused on the isolated retina via the lens system of an inverted microscope. The maximal luminance incident on the retina was 49.2 cd/m², measured using a luminance meter (SM208, Sanpometer, Shenzhen, China).

Full-field Gaussian contrast stimulation was used to estimate temporal kinetics and light sensitivity of the retinal network. The luminance of each step of the stimulus sequence was drawn from a Gaussian white noise distribution with a constant mean luminance ($M=$ 24.6 cd/m², equivalent to $5.46\times {10}^3$ R*/rod/s, mesopic condition) and standard deviation ($W$) being either high ($W_{\text{high}}=$ 8.6 cd/m²) or low ($W_{\text{low}}=$ 3.3 cd/m²), thus providing high (35.0%) and low contrast (13.4%) conditions, with contrast being defined as $C=W/M\times 100\%$. In every contrast flickering sequence, the duration of each luminance frame was set as 50 ms (with frame refresh rate being 20 Hz). The stimulation started with a 20-s high-contrast flickering sequence, which was followed by a 20-s low-contrast sequence in one trial, and 20 trials were presented sequentially. Before the contrast stimulus protocol, a sustained full-field white light (24.6 cd/m²) was applied for 30 s to adjust the light sensitivity of the RGCs to the mean luminance (Liu et al. 2007).

Response property identification

In the present study, ON-center, OFF-center and ON–OFF RGC were first identified according to their responses elicited by full-field light flashes, which consisted of 5-s ON duration (49.2 cd/m²) and 5-s dark interval (0 cd/m²), and were repeated for 30 times. The RGC subtype was further confirmed by analyzing the cell’s response properties under the sustained full-field Gaussian contrast stimulation, using spike-triggered average (STA) or spike-triggered covariance (STC) methods (Liu and Gollisch 2015; Puyang et al. 2017).

The STA analysis was adopted to confirm the ON-center and OFF-center RGCs. In our study, the spike-triggered stimulus was defined as the stimulus sequence 0–500 ms preceding a spike. An ON (OFF) cell was identified when its averaged spike-triggered stimulus has a positive (negative) first peak.

The STC analysis was adopted to separate the ON- and OFF-response phases of an ON–OFF RGC. In our study, all the spike-triggered stimuli were first reduced into 2-dimensional data by principal component analysis (PCA) (Pang et al. 2016). If these data formed two identifiable clusters (corresponding to ON and OFF stimuli) in the 2-dimensional plane, this cell was confirmed to be an ON–OFF RGC. A K-means clustering algorithm was then adopted to separate the two clusters, and the relevant responses of this ON–OFF RGC were divided into ON- and OFF-response accordingly.

Adaptation/sensitization identification

To classify the dynamic modulation type of each RGC (adaptation or sensitization), an adaptation index ($\mathit{AI}$) was adopted:

$$AI=\frac{r_{\mathit{early}}-r_{\mathit{late}}}{r_{\mathit{early}}+r_{\mathit{late}}},$$

where $r_{\mathit{early}}$ and $r_{\mathit{late}}$ refer to the mean firing rate during the early (0.5–5.5 s) and late (14.5–19.5 s) periods of a certain stimulus. Thus, $\mathit{AI}$ values could be calculated for both high-contrast and low-contrast periods (${\mathit{AI}}_{\mathit{high}}$ and ${\mathit{AI}}_{\mathit{low}}$). A cell whose firing rate decreased during high-contrast stimulation but increased during low-contrast period was classified as an adapting cell, while a cell with negative ${\mathit{AI}}_{\mathit{high}}$ and positive ${\mathit{AI}}_{\mathit{low}}$ was classified as a sensitizing cell. Those cells whose firing rates changed in the same direction during high- and low-contrast periods were identified as non-adapting-non-sensitizing cells.

Linear–nonlinear analysis

In the present study, a linear–nonlinear (LN) model was applied to describe temporal kinetics and light sensitivity of the retinal network which transforms light stimulation to spikes in an RGC (Chichilnisky 2001). The input signal of the LN model, $s\left(t\right)$, indicates the intensity of light striking at the surface of the photoreceptor layer and the output signal $r\left(t\right)$ represents the instantaneous firing rate of an RGC. A well-fitted model therefore mimics the overall function of retinal neural network from photoreceptors receiving light stimulation to RGC bursting spikes.

Functionally, the linear part of the model, $\mathit{L}\left(t\right)$, characterizes how the retinal network integrates the spatial, temporal and spectral components of the stimulus and was set as a normalized linear kernel estimated from the STA algorithm (Chichilnisky 2001; Schwartz et al. 2006). The intermediate variable $g\left(t\right)$, which indicates the integrated outputs of the linear part, was then computed by convolution of the linear part $\mathit{L}\left(t\right)$ and the input signal $s\left(t\right)$.

The nonlinear part of the model, $\mathit{N}\left(g\right)$, characterizes the function by which spikes are generated at the RGC. In the present study, using the pre-smoothed data processed by sliding window averaging ($\text{window length}=\left(g_{\text{max}},-,g_{\text{min}}\right)/20$, $\text{steps}=\text{window length}/5$, from $g_{\text{min}}$ to $g_{\text{max}}$), the nonlinear part was fitted by a sigmoidal function, based on least-square criteria. The sigmoidal function was:

$$r=\frac{m}{1+{\text{e}}^{-k\left(g,-,g_{1/2}\right)}},$$

where $m$ indicates its maximal value, $g_{1/2}$ indicates the $g$ value where $r=\frac{m}{2}$, and $k$ controls the maximal slope of the sigmoidal function.

In this study, parameters of the LN model were fitted based on the RGC’s response to the sustained full-field Gaussian contrast stimulation, which involves neither spatial nor spectral properties. Therefore, the linear part characterizes only the temporal kinetics of the retinal network, which could be represented by the peak-time ($T_{\mathbf{L}}$) of the normalized linear kernel. A smaller $T_{\mathbf{L}}$ indicates a faster temporal kinetics of the retinal network (Baccus and Meister 2002). Since the linear part was normalized, all magnifying effects were transferred to the nonlinear part. For the nonlinear part, the averaged slope ($P_{\mathbf{N}}$) of the sigmoidal function was adopted to represent average light sensitivity of the retinal network, which was calculated by dividing the averaged height by the width:

$$P_{\mathbf{N}}=\frac{{\int }_{g1}^{g_2}r\left(g\right)\text{d}g}{g_2-g_1}/(g_2-g_1)=\frac{{\int }_{g_1}^{g_2}r\left(g\right)\text{d}g}{{\left(g_2,-,g_1\right)}^2},$$

where $r\left(g\right)$ is the fitted sigmoidal function, $g_1$ and $g_2$ are related to 5% and 100% $r_{\text{max}}$, respectively.

Stimulus reconstruction and information estimation

According to the LN analysis, the linear kernel estimated by the STA algorithm encompasses the stimulus feature that the spikes represent. Further, precise onset of a spike encodes the time when the feature appears (Gaudry and Reinagel 2007; Keat et al. 2001).

In our study, we tried to reconstruct the light stimuli based on an individual RGC’s spike train and then evaluate the similarity between the reconstructed stimuli and real light stimuli.

To perform the reconstruction, the linear kernel, $\text{k}$, was derived by the STA algorithm based on the spike train recorded during a specific time period. Then the kernel was aligned with each spike, with ${\text{k}}_{\text{i}}$ corresponding to the ith spike (Fig. 1a, b). Finally, all the $\text{k}$ were summed to get continuous reconstructed stimulus, ${s_{\text{r}}}^{\prime }\left(t\right)$. Then ${s_{\text{r}}}^{\prime }\left(t\right)$ was discretized by its mean value in each 50-ms time window to generate the discrete reconstructed stimulus, $s_{\text{r}}\left(t\right)$, (Fig. 1c).

Fig. 1.

Reconstruction of a stimulus based on a single RGC’s spike train. a A piece of spike train (10 spikes) of an RGC. b The linear kernel $\text{k}$ is aligned with each of the spikes. ${\text{k}}_1$–${\text{k}}_{10}$ corresponding to the 1st–10th spike, they have identical shape but occur at different times. The dotted lines are one spike-linear kernel pair. c The thin line indicates the continuous reconstructed stimulus ${s_{\text{r}}}^{\prime }\left(t\right)$, computed by summing all the $\text{k}$. The thick line indicates the discrete reconstructed stimulus $s_{\text{r}}\left(t\right)$, determined by the mean value of ${s_{\text{r}}}^{\prime }\left(t\right)$ in each 50-ms time window

After the reconstruction, the Pearson correlation coefficient (PCC) was used to evaluate the similarity between the reconstructed and the real light stimuli (Appleby and Manookin 2019):

$$\text{PCC}=\frac{\text{cov}\left(s,,,s_{\text{r}}\right)}{{\mathrm{\sigma }}_s{\mathrm{\sigma }}_{s_{\text{r}}}},$$

where $\text{cov}$ indicates covariance, $\mathrm{\sigma }$ is standard deviation, $s$ and $s_{\text{r}}$ refer to real and reconstructed light stimuli respectively. The maximum value of PCC is 1, which occurs when the reconstructed and the real stimuli are identical. Thus, the larger is PCC, the more information carried by the spike train in its temporal structure about the light stimulation.

Statistics

All analyses were carried out with Prism software version 8 (GraphPad). All statistical comparisons were performed using the Wilcoxon signed-rank test unless otherwise specified. A $p$ value $<$ 0.05 was considered to be statistically significant. All data in the main text are presented as $\text{mean}\pm \text{SD}$ and error bars in figures represent SEM.

In the statistic used to characterize the relationship between adaptation/sensitization and ON/OFF cells, cells were excluded if: (1) The mean firing rate during the low-contrast period was higher than that during the high-contrast period (the defining property of suppressed-by-contrast RGC by Rodieck (1967)); (2) The total number of spikes during the early (0.5–5.5 s) and late (14.5–19.5 s) periods of the low contrast stimulation was less than 100; (3) Cells showed neither adaptation nor sensitization.

In the statistic for the results of LN analysis, cells were excluded if: (1) The spike number in any trial was greater than 20% of the spike number in all trials; (2) The linear or nonlinear part of LN model was not smooth.

In the statistic for the results of stimulus reconstruction, cells were recruited only if the cluster of spikes and the cluster of noises were clearly separated during the spike sorting process.

Results

Contrast adaptation and sensitization in mouse retinal ganglion cells

RGCs respond to temporal contrast stimulation and swiftly adjust their response when contrast level changes. Typically, when the contrast level is switched from low to high, RGCs increase their firing rate, and vice versa. Meanwhile, RGCs’ responsiveness keeps on changing, although slowly, when the contrast is kept at a constant level (Baccus and Meister 2002; Kim and Rieke 2001).

Among RGCs recorded in our study, a group of them fired with an activity pattern similar to previously reported adaptation. The example given in Fig. 2a shows that, the cell’s firing rate was relatively high at the onset of high-contrast stimulus and was gradually decreasing during the following 20-s prolonged high-contrast stimulus (${\mathit{AI}}_{\mathit{high}}=0.0655$). On the other hand, the cell’s firing rate was relatively low at the onset of low-contrast stimulus and was gradually increasing when such stimulation was persisted (${\mathit{AI}}_{\mathit{low}}=-0.1834$).

Fig. 2.

Adaptation and sensitization in mouse RGCs. a, b Top, schematic diagrams of light intensity (I) of sustained full-field Gaussian contrast stimulus (see Methods) in one trial. Four time periods are defined: H_early and L_early, 0.5–5.5 s after the onset of high- and low-contrast stimulus, respectively; H_late and L_late, 14.5–19.5 s after the onset of high- and low-contrast stimulus, respectively. Middle, raster plots (20 trials) of the firing activities of an adapting RGC and a sensitizing RGC in response to the contrast stimuli, respectively. Bottom, corresponding peri-stimulus time histograms (PSTH) of the cells. Time bin $=$ 500 ms

Meanwhile, another group of RGCs exhibited a different response pattern in exposure to the same stimulation. For an RGC illustrated in Fig. 2b, its firing rate was relatively low at the onset of high-contrast stimulus, which was followed by a further sluggish increase during the persistent high-contrast stimulus (${\mathit{AI}}_{\mathit{high}}=-0.0635$). Accordingly, at the onset of low-contrast stimulus, the firing rate was relatively high, which was followed by a gradual decrease in exposure to the prolonged low-contrast stimulus (${\mathit{AI}}_{\mathit{low}}=0.0702$). This type of RGCs changed the firing rate in an opposite way as compare to the adapting RGCs during the persistent high- and low-contrast stimulations. According to Kastner and Baccus (2011), this kind of dynamic process in RGCs was characterized as sensitization.

Functional-pathway-dominance of adaptation and sensitization

In our study, many RGCs could be clearly classified into adaptation and sensitization groups, and in the meantime, they could also be classified into ON-center and OFF-center groups. We investigated whether the response modulation effects were related to the different functional subtypes based on data obtained from 465 cells (Fig. 3a). The result shows that, the 283 adapting cells included 218 OFF cells (77.0%) and 65 ON cells (23.0%), while the 182 sensitizing cells included 169 ON cells (92.9%) and 13 OFF cells (7.1%). Meanwhile, this also reflects that OFF cells were mostly adapting cells and ON cells were mostly sensitizing cells.

Fig. 3.

Functional-pathway-dominance of adaptation and sensitization. a The distributions of adaptation and sensitization in OFF- and ON-center RGCs. b Upper, 2D-projection of the spike-triggered stimuli of an example ON–OFF RGC (PC1 and PC2). Bottom, two averaged stimuli based on the data in the clusters in the upper panel. Black: OFF-stimuli; Grey: ON-stimuli. c PSTHs of the separated OFF- (upper) and ON-response (bottom) of the example ON–OFF RGC. d The distributions of adaptation and sensitization in OFF- and ON-components of ON–OFF RGCs

Given the strong relationship between adaptation/sensitization and OFF/ON-center RGCs, one may consider about whether the segregation is originated from the RGCs or from the functional pathways (ON- and OFF-pathway). ON–OFF RGC in this way serves as a good model to make a primary judgement since it receives and integrates inputs from both ON- and OFF-pathway. We then further investigated the segregation of adaptation and sensitization in different functional pathways in ON–OFF RGCs. The separation of ON- and OFF-component for an example ON–OFF RGC is illustrated in Fig. 3b, c, in which the OFF-component (Fig. 3b, black; Fig. 3c, upper panel) showed adaptation (${\mathit{AI}}_{\mathit{high}}=0.0828$, ${\mathit{AI}}_{\mathit{low}}=-0.3171$), while the ON-component (Fig. 3b, grey; Fig. 3c, bottom panel) showed sensitization (${\mathit{AI}}_{\mathit{high}}=-0.0223$, ${\mathit{AI}}_{\mathit{low}}=0.0079$).

In our study, for 13 ON–OFF RGCs whose two components were both available for the analysis, adaptation was found in 13 OFF-components (68.4%) and 6 ON-component (31.6%) while sensitization was exclusively found in 7 ON-components (100.0%). In addition to these 13 ON–OFF cells, there were also 22 ON–OFF cells in which only one component was available (12 OFF and 10 ON) for quantitative analysis, following the criteria explained in Methods part. While taking all these components into consideration (Fig. 3d), the statistical result shows that adaptation occurred in 25 OFF-components (73.5%) and 9 ON-components (26.5%), while sensitization exclusively occurred in 14 ON-components (100.0%). These results matched the observations from OFF- and ON-center cells that adaptation and sensitization were mostly attributed to OFF- and ON-cells, respectively.

These results suggest that, instead of cell-type-dominant, adaptation and sensitization are functional-pathway-dominant.

Modulations of network responsiveness during adaptation and sensitization

The changes of RGC’s firing rate in exposure to sustained contrast stimulations were considered as the result of responsiveness changes in the retinal network (Kastner and Baccus 2011). In our study, LN model (Fig. 4a) analysis (see Methods) was applied to quantify the changes of temporal kinetics ($T_{\mathbf{L}}$) and average light sensitivity ($P_{\mathbf{N}}$) for both adaptation and sensitization. $P_{\mathbf{N}}$ values calculated for the four time periods were normalized against the value calculated for H_early for each RGC. Thus, the normalized $P_{\mathbf{N}}$ has an arbitrary unit (AU) and a smaller $P_{\mathbf{N}}$ reflects a relatively lower light sensitivity. Statistical results were based on data collected from 334 RGCs (187 adapting cells and 147 sensitizing cells).

Fig. 4.

Linear–nonlinear analysis of adaptation and sensitization. a Schematic diagram of the LN model (see Methods). $s\left(t\right)$: light intensity; $\mathit{L}\left(t\right)$: the linear part, with $T_{\mathbf{L}}$ representing the peak-time; $g\left(t\right)$: integrated output of the linear part; $\mathit{N}\left(g\right)$: the nonlinear part, in which $g_1$ and $g_2$ are related to 5% and 100% $r_{\text{max}}$, respectively; $r\left(t\right)$: instantaneous firing rate. b The linear part (left) and nonlinear part (right) calculated for the four time periods for an adapting RGC. c Comparison of $T_{\mathbf{L}}$ (left) and normalized $P_{\mathbf{N}}$ (right) between the early and late periods (Wilcoxon signed-rank test, ** means $p<0.05$) for adaptation. Error bars: $\text{mean}\pm \text{SEM}$, $\text{N}=$ 187. d Same as b but for a sensitizing RGC. e Same as c but for sensitization. Error bars: $\text{mean}\pm \text{SEM}$, $\text{N}=$ 147

For an example adapting RGC presented in Fig. 4b, the $T_{\mathbf{L}}$ value was barely changed during the persistent high- (H_early: 76 ms, H_late: 77 ms) and low-contrast (L_early: 88 ms, L_late: 85 ms) stimuli. Meanwhile, the normalized $P_{\mathbf{N}}$ value was decreased during high-contrast (H_early: 1.00, H_late: 0.92, AU) period and increased during low-contrast (L_early: 1.25, L_late: 1.87, AU) period. Statistics for all the 187 adapting RGCs showed that $T_{\mathbf{L}}$ value (Fig. 4c, left) was not significantly changed during prolonged high- (H_early: 75.81 $\pm$ 9.19 ms, H_late: 75.98 $\pm$ 9.10 ms, $\text{mean}\pm \text{SD}$, $p>0.05$) or low-contrast (L_early: 84.97 $\pm$ 10.93 ms, L_late: 85.32 $\pm$ 10.34 ms, $\text{mean}\pm \text{SD}$, $p>0.05$) stimulation but the normalized $P_{\mathbf{N}}$ value (Fig. 4c, right) was significantly decreased during the persistent high-contrast (H_early: 1.00, H_late: 0.94 $\pm$ 0.08, AU, $\text{mean}\pm \text{SD}$, $p<0.05$) stimulation and increased during the persistent low-contrast (L_early: 1.36 $\pm$ 0.71, L_late: 1.74 $\pm$ 0.77, AU, $\text{mean}\pm \text{SD}$, $p<0.05$) stimulation.

The same analysis was applied on sensitizing cells. For an example sensitizing RGC presented in Fig. 4d, $T_{\mathbf{L}}$ value was slightly changed during the persistent high- (H_early: 76 ms, H_late: 75 ms) and low-contrast (L_early: 78 ms, L_late: 83 ms) stimuli while the normalized $P_{\mathbf{N}}$ value was increased during the high-contrast (H_early: 1.00, H_late: 1.02, AU) period and decreased during the low-contrast (L_early: 1.73, L_late: 1.65, AU) period. Statistics of all 147 sensitizing RGCs showed that $T_{\mathbf{L}}$value (Fig. 4e, left) was not significantly changed during the persistent high-contrast (H_early: 67.93 $\pm$ 9.98 ms, H_late: 67.78 $\pm$ 9.44 ms, $\text{mean}\pm \text{SD}$, $p>0.05$) stimulation but was significantly increased during the persistent low-contrast (L_early: 72.67 $\pm$ 12.19 ms, L_late: 74.16 $\pm$ 11.65 ms, $\text{mean}\pm \text{SD}$, $p<0.05$) stimulation. The normalized $P_{\mathbf{N}}$ value of the 147 sensitizing RGCs (Fig. 4e, right) was significantly increased during the persistent high-contrast (H_early: 1.00, H_late: 1.03 $\pm$ 0.06, AU, $\text{mean}\pm \text{SD}$, $p<0.05$) stimulation and decreased during the persistent low-contrast (L_early: 1.76 $\pm$ 0.62, L_late: 1.66 $\pm$ 0.61, AU, $\text{mean}\pm \text{SD}$, $p<0.05$) stimulation.

Information encoded in spike trains of adapting and sensitizing cells

In our study, adaptation and sensitization were simultaneously observed in RGCs in exposure to sustained full-field Gaussian contrast stimulation (Fig. 2). Meanwhile, the majority of adaptation was observed in OFF-center cells (and OFF-component of ON–OFF RGCs) and the majority of sensitization was observed in ON-center cells (and ON-component of ON–OFF RGCs) (Fig. 3) and the average sensitivity of retinal network was oppositely changed for adaptation and sensitization (Fig. 4). While the functional advantages of adaptation in retinas have been well-understood, we wanted to know whether sensitization contributes to information coding.

Since temporal structure of single RGC’s spike train was thought to encode information about light stimulation (Berry et al. 1997; Van Rullen and Thorpe 2001), we tried to evaluate the amount of information carried by the temporal structure of adapting and sensitizing RGCs’ spike train, respectively. In the present study, the stimuli were reconstructed according to single RGC’s spike train, and the similarity between the reconstructed stimuli and the real light stimuli was evaluated by the PCC index (see Methods). The temporal structure of spike train is considered to carry more information when the PCC value is bigger.

Data from 246 cells (130 adapting cells and 116 sensitizing cells) were included in this stimulus reconstruction stage.

The light stimuli $s\left(\text{t}\right)$ and the reconstructed stimuli $s_r\left(\text{t}\right)$ calculated based on the spike train of an adapting RGC were compared (Fig. 5a). For this adapting RGC, PCC value was slightly decreased during the persistent high-contrast (H_early: 0.6210, H_late: 0.6146) stimulus, and was increased during the persistent low-contrast (L_early: 0.4542, L_late: 0.4934) stimulus. The average PCC value for the 130 adapting RGCs (Fig. 5c) also showed significant changes during the persistent high-(H_early: 0.6301 $\pm$ 0.0595, H_late: 0.6229 $\pm$ 0.0620, $\text{mean}\pm \text{SD}$, $p<0.05$) and low-contrast (L_early: 0.4340 $\pm$ 0.1072, L_late: 0.4585 $\pm$ 0.0997, $\text{mean}\pm \text{SD}$, $p<0.05$) stimulations.

Fig. 5.

Reconstructed stimuli and the PCC indices. a, b The real (grey) and reconstructed (black) light stimuli in one trial during the four selected time periods for an adapting RGC and a sensitizing RGC, respectively. c Comparison of PCC between the early and late periods (Wilcoxon signed-rank test, ** means $p<0.05$) for adaptation. Error bars: $\text{mean}\pm \text{SEM}$, $\text{N}=$ 130. d Same as c but for sensitization. Error bars: $\text{mean}\pm \text{SEM}$, $\text{N}=$ 116

For the example sensitizing RGC (Fig. 5b), its PCC value was increased during the persistent high-contrast (H_early: 0.6852, H_late: 0.6952) stimulus and was decreased during the persistent low-contrast (L_early: 0.6387, L_late: 0.6189) stimulus. The average PCC value for the 116 sensitizing RGCs (Fig. 5d) showed significant changes during the persistent high- (H_early: 0.6552 $\pm$ 0.0697, H_late: 0.6647 $\pm$ 0.0709, $\text{mean}\pm \text{SD}$, $p<0.05$) and low-contrast (L_early: 0.5899 $\pm$ 0.1164, L_late: 0.5775 $\pm$ 0.1209, $\text{mean}\pm \text{SD}$, $p<0.05$) stimulations.

Discussion

Contrast adaptation and sensitization were simultaneously observed in mouse RGCs under persistent full-field Gaussian contrast stimulation. The main findings of the present study are: (1) the majority of adaptation is observed in OFF-center cells (and OFF-component of ON–OFF RGCs) and the majority of sensitization is observed in ON-center cells (and ON-component of ON–OFF RGCs); (2) the average light sensitivity is oppositely modulated for adaptation and sensitization during persistent stimulations, the temporal kinetics is slowed down exclusively during sensitization in response to persistent low-contrast stimulation; and (3) the encoded information in RGC’s spike train is changed in opposite directions for adaptation and sensitization during the persistent stimulations.

Functional-pathway-dominance of adaptation and sensitization

In the present study, two types of dynamic processes were observed, namely adaptation and sensitization, which are consistent with the previously reported contrast adaptation (Baccus and Meister 2002; Kim and Rieke 2001) and contrast sensitization (Appleby and Manookin 2019; Kastner and Baccus 2011) in vertebrate retinas. On the other hand, RGCs are usually classified into OFF-center and ON-center subtypes according to their receptive field properties. Our results revealed that adaptation and sensitization are mainly attributed to OFF- and ON-signal pathways, respectively.

The functional-pathway-dominance of adaptation and sensitization in our study is consistent with previous studies. A study focused on salamander fast OFF RGCs reported that more than 75$\%$ of them showed adaptation under sustained contrast stimulation (Kastner and Baccus 2011). Furthermore, the afterhyperpolarization, which arises from the presynaptic bipolar cells and leads to contrast adaptation in RGCs, was observed in nearly all OFF-center RGCs but was weak or absent in most ON-center RGCs in experiments conducted on guinea pig Y-type RGC (Manookin and Demb 2006). In macaque monkey retina, broad thorny and parasol cells showed adaptation and midget cells showed sensitization (Appleby and Manookin 2019).

Functional benefits of adaptation and sensitization

The responsiveness of retinal network keeps changing during persistent stimulations. In this study, it was found that for adapting RGCs, the light sensitivity was gradually decreased during the persistent high-contrast stimulation while was gradually increased during the persistent low-contrast stimulation. The opposites were observed in sensitizing RGCs. These properties of adaptation and sensitization are consistent with previous studies about contrast adaptation (Baccus and Meister 2002; Manookin and Demb 2006) and sensitization (Kastner and Baccus 2011; Nikolaev et al. 2013), respectively.

One functional benefit of visual adaptation is that it improves the discriminability of rare stimulation by suppressing responses to frequent stimulation (Dragoi et al. 2002; Hosoya et al. 2005; Sharpee et al. 2006). Under the full-filed sustained Gaussian contrast stimulation in this study, the rare and frequent stimuli are related to strong and weak ones, respectively (Fig. 6). In this way, contrast adaptation is suggested to improve the discriminability of strong stimulation, i.e. contrast adaptation prevents saturation to strong inputs.

Fig. 6.

Distribution of light intensity in one full-field Gaussian contrast stimulation

On the other hand, sensitization enhanced the light sensitivity to weak stimuli during the persistent high-contrast stimulation (Fig. 4d, e). In addition, during the low-contrast period which is dominated by the weak stimuli, sensitizing RGCs were found to encode more information than adapting RGCs (Mann-Whitney test, adaptation vs. sensitization, for L_early and L_late, respectively. $p<0.05$, $\text{N}=$ 130 for adaptation and $\text{N}=$ 116 for sensitization). In this way, adaptation and sensitization work oppositely to encode the visual inputs with adapting cells encoding strong signals and sensitizing cells encoding weak signals. Although the coding capacity was estimated based on a simple linear–nonlinear model, and the parameters calculated might be somewhat biased, they are qualitatively reliable.

As suggested by Barlow (1961), the primary effect of the sensory messages an animal receives is not to enrich its subjective experience of the world but to modify its behavior so as to have a greater chance of survival. In the visual world of mouse, a strong OFF stimulus from its back is rare but often indicates the approaching of predators and a delayed recognition of such kind of stimulation is fatal to this species. In this way, adaptation in OFF RGCs prevents saturation to strong OFF inputs which enables mouse immediately respond to potential dangers. On the other hand, weak ON and weak OFF inputs are frequent and important for the movement and predation of mouse on the ground. In the natural environment in which mouse lives, the ON and OFF inputs are evenly distributed along time and an ON-signal is often temporally accompanied by an OFF-signal (Baden et al. 2013; Ratliff et al. 2010). Thus, with the dominance of sensitization in ON RGCs, mice are able to catch the faint light.

Ubiquity of adaptation and sensitization

In the present study, adaptation and sensitization showed opposite firing profiles and modulations of light sensitivity during the persistent contrast stimulations. Furthermore, these two dynamic processes oppositely worked with each other with adapting and sensitizing cells encoding strong OFF and weak ON stimulations, respectively.

Such opposite dynamic processes were also observed for other aspects. In mouse retina, the direction-selective RGCs changed the directional preference after drifting grating stimuli (Rivlin-Etzion et al. 2012), by adapting to the original prefer-direction stimulation but sensitizing to the original null-direction stimulation at the meantime. In macaque monkey retina, the parasol ganglion cells strongly adapted during high-contrast stimulation while the midget ganglion cells sensitized (Appleby and Manookin 2019).

In addition, adaptation and sensitization were also observed in the primary visual cortex of cat. During the drifting grating stimulation, ~ 72$\%$ of the neurons gradually decreased the firing rate while ~ 28$\%$ of the neurons gradually increased the firing rate (Albrecht et al. 1984). On the other hand, for neurons in cat striate cortex, after a 60-s unidirectional stimulus, it was observed that some neurons adapted to this type of stimulus while some neurons sensitized (Marlin et al. 1988).

Taken together, adaptation occurs in a variety of forms in visual systems. Given that adapted neurons become less sensitive to weak inputs (Baccus and Meister 2002; Brown and Masland 2001; Manookin and Demb 2006; Solomon et al. 2004), another strategy should be processed to maintain the information transmission. In our study, sensitization was found to work coordinately (although in an opposite manner) with adaptation under the full-field contrast stimulation. The coexistence of adaptation and sensitization in visual systems and their functional advantages deserve more attention in the future.

Author contributions

MD and PL designed the study, MD performed the experiments, MD and PL analyzed the data, MD and PL wrote the manuscript.

Funding

This work was supported by grants from National Natural Science Foundation of China (No. 31471054, to PL).

Availability of data and material

The datasets generated for this study are available on request to the corresponding author.

Compliance with ethical standards

Conflict of interest

The authors declare that they have no conflict of interest.

Consent to participate

Not applicable.

Code availability

The codes used in this study are available on request to the corresponding author.

Ethical approval

The animal study was reviewed and approved by Shanghai Jiao Tong University Institutional Animal Care and Use Committee (SJTU IACUC).