Light adaptation largely preserves the relative organization of parallel retinal channels: implications for efficient coding theory

Curation statements for this article:
  • Curated by eLife

    eLife logo

    eLife Assessment

    This paper describes careful and convincing work on the effects of light adaptation on the spatio-temporal receptive fields of retinal ganglion cells assessed using an ex vivo preparation on a multielectrode array. In general, this valuable quantitative analysis shows that cells become tuned to faster and smaller features in bright light, but with the same basic receptive field shapes, and most cell types adapt similarly and in parallel, with some exceptions. It remains unclear why the results presented here differ from some prior studies that painted a different picture of adaptation to light intensity.

This article has been Reviewed by the following groups

Read the full article See related articles

Abstract

Two major functions performed by the retina are to establish the parallel processing of visual information and to adapt visual encoding to the trillion-fold range of light intensities encountered in the environment. Previous work has highlighted many specialized cell types and circuits that instantiate parallel processing and light adaptation. However, fully understanding either process requires identifying how light adaptation and parallel processing interact. One possibility is that light adaptation causes uniform or proportional scaling to the receptive fields (RFs) of different retinal ganglion cell (RGC) types, the output neurons of the retina. Alternatively, light adaptation could cause a reorganization of RF structures across RGC types. To resolve these possibilities, we examined how the spatiotemporal RF structure of six simultaneously measured RGC types in the rat retina change from rod- to cone-mediated light levels. While light adaptation altered the RF properties of all six RGC types, we found that the relative structure across different RGC types was largely preserved across light levels. Surprisingly, most RGC types retained their center-surround RF structure even at low light levels, an observation that is at odds with prior efficient coding predictions. However, we show these predictions are incomplete and when RFs interact over a finite viewing area, efficient coding predicts the retention of surrounds under low signal- to-noise conditions.

Article activity feed

  1. eLife Assessment

    This paper describes careful and convincing work on the effects of light adaptation on the spatio-temporal receptive fields of retinal ganglion cells assessed using an ex vivo preparation on a multielectrode array. In general, this valuable quantitative analysis shows that cells become tuned to faster and smaller features in bright light, but with the same basic receptive field shapes, and most cell types adapt similarly and in parallel, with some exceptions. It remains unclear why the results presented here differ from some prior studies that painted a different picture of adaptation to light intensity.

  2. Reviewer #1 (Public Review):

    The retina adapts to statistics of the scenes it views, and in particular can adapt to million-fold changes in light intensity. The adaptation alters how the population of retinal ganglion cells encodes the scenes, and it is unclear how downstream circuits could handle different types of changes in scene encoding. This careful study investigates how a population of six different retinal ganglion cells changes their encoding response properties as a function of mean luminance level. It does this by mapping receptive fields and response nonlinearities over a million-fold change in light intensity, from the scotopic to photopic regime.

    This paper has several strengths. (1) It is clearly written and lays out clearly the alternative hypotheses it examines. (2) Its experiments are carefully done and rigorously analyzed. (3) It finds interesting results, pointing especially to two cell types that seem to stand out for their different responses to increasing light levels in these experiments. It also finds less spatial and temporal integration at low light levels than theory predicts (or than others have found).

    This paper has relatively minor weaknesses. (1) It does not particularly resolve why its results are different from prior, similar experiments, though it carefully discusses and highlights those differences. (2) The functional implications of these results, and especially the two cell types that stand out from the crowd, are not discussed with much depth. This likely reflects an issue beyond this paper: that the functions of different RGC types for visual processing in the brain remain poorly understood, which in turn makes it difficult to understand functional implications of these results.

  3. Reviewer #2 (Public Review):

    This paper describes careful work on the effects of light adaptation on the spatio-temporal receptive fields of retinal ganglion cells assessed using an ex vivo preparation on a multielectrode array. The results are analyzed systematically and clearly but do not provide a clear insight into this important aspect of retinal function, in part because of doubts about how they might extrapolate to the situation in vivo.

    Strengths:

    - A potentially interesting question: Do changes in average luminance alter spatio-temporal receptive fields of different types of retinal ganglion cell (RGC) in different ways? Or are different types of RGC affected in similar ways and to similar extents? Previous work going back decades has investigated the effects of luminance adaptation in a piece-meal way but this study makes a more systematic comparison across six types of RGC.

    - Experiments and analyses have been carried to a very good level.

    - Thorough description of the results with clear figures.

    - A scholarly presentation that includes important literature.

    Weaknesses:

    - Some of the effects observed in the six types of RGC, such as shortening of integration times at higher luminance, have been recognized for some time (as is clearly discussed).

    - The most novel funding is probably that one of the six types of RGC, the ON brisk-sustained, exhibited a proportionately larger change in integration time than other types, but this has not been placed in a functional context. The impact of the paper in terms of understanding retinal processing is unclear.

    - A major drawback of all work on an MEA, usually ignored by users of this technique, is that diffusible neuromodulators are removed by perfusion. Neuromodulators can profoundly adjust processing through the retina and their levels can change in response to changes in luminance. The retinal network on an MEA is therefore likely to be far from its normal state. It is hard to know how far the ability of a retina on MEA to adapt reflects adaptation in vivo.

    - An important control, reversing the direction of the luminance change, has not been carried out. If stepping back from a higher light level to a lower level does not (at least partially) reverse the changes caused by an increase in luminance, it may be that the lack of a pigment epithelium is affecting the results more severely than estimated by calculation.

  4. Reviewer #3 (Public Review):

    There have been variable reports on the degree to which light adaptation can modify a retinal ganglion cell's receptive field properties. The classic description suggested that the receptive field would become more low-pass in both space and time in dim light, better to integrate when signals are weak. More recent studies, particularly Tikidji-Hamburyan et al. (2015), suggested more radical changes in receptive fields across light levels, raising questions about how downstream circuits could keep track of highly variable output of the same ganglion cell, depending on the mean luminance.

    The current study revisits this problem with a straightforward set of experiments: measuring spikes in rat retinal ganglion cells, using a multi-electrode array, and presenting either spatiotemporal white noise or drifting gratings while varying the mean intensity over a one-million fold range. The technical achievement of this work is the careful and quantitative characterization of receptive field properties of specific cell types and the careful tracking of individual cells across light levels using each cell's electrophysiological image.

    The current study doesn't exactly agree with either of the extreme views from the literature. That is, the six ganglion cell types studied here mostly adapted in a similar way - becoming tuned to higher temporal and spatial frequencies in bright light. However, the receptive field did not become extremely low-pass at the dimmest light level, and there were exceptions to how individual cell types adapted, with some types showing a unique pattern, e.g., the ON brisk-sustained cell shows a particularly large shift in the peak temporal frequency in bright light. It's not exactly clear what to make of these unique patterns, which were rare, and it suggests that downstream circuits must somehow be attuned to these patterns or at least there is some benefit to these patterns that we do not currently understand.

    The study offers many technical strengths and is written clearly. If there is a weakness, it is that the new analysis does not clarify some of the discrepancies from previous studies, which had used various species, cell types and light levels that do not completely overlap with the current study. Still, if future studies followed the standard set in the current one, the field would certainly benefit.

  5. Author response:

    Reviewer #1 (Public Review):

    The retina adapts to statistics of the scenes it views, and in particular can adapt to million-fold changes in light intensity. The adaptation alters how the population of retinal ganglion cells encodes the scenes, and it is unclear how downstream circuits could handle different types of changes in scene encoding. This careful study investigates how a population of six different retinal ganglion cells changes their encoding response properties as a function of mean luminance level. It does this by mapping receptive fields and response nonlinearities over a million-fold change in light intensity, from the scotopic to photopic regime.

    This paper has several strengths. (1) It is clearly written and lays out clearly the alternative hypotheses it examines. (2) Its experiments are carefully done and rigorously analyzed. (3) It finds interesting results, pointing especially to two cell types that seem to stand out for their different responses to increasing light levels in these experiments. It also finds less spatial and temporal integration at low light levels than theory predicts (or than others have found).

    This paper has relatively minor weaknesses. (1) It does not particularly resolve why its results are different from prior, similar experiments, though it carefully discusses and highlights those differences. (2) The functional implications of these results, and especially the two cell types that stand out from the crowd, are not discussed with much depth. This likely reflects an issue beyond this paper: that the functions of different RGC types for visual processing in the brain remain poorly understood, which in turn makes it difficult to understand functional implications of these results.

    We thank the reviewer for their clear summary.

    Reviewer #2 (Public Review):

    This paper describes careful work on the effects of light adaptation on the spatio-temporal receptive fields of retinal ganglion cells assessed using an ex vivo preparation on a multielectrode array. The results are analyzed systematically and clearly but do not provide a clear insight into this important aspect of retinal function, in part because of doubts about how they might extrapolate to the situation in vivo.

    We thank the reviewer for their clear summary.

    Strengths

    - A potentially interesting question: Do changes in average luminance alter spatio-temporal receptive fields of different types of retinal ganglion cell (RGC) in different ways? Or are different types of RGC affected in similar ways and to similar extents? Previous work going back decades has investigated the effects of luminance adaptation in a piece-meal way but this study makes a more systematic comparison across six types of RGC.

    - Experiments and analyses have been carried to a very good level.

    - Thorough description of the results with clear figures.

    - A scholarly presentation that includes important literature.

    We thank the reviewer for their assessment.

    Weaknesses

    - Some of the effects observed in the six types of RGC, such as shortening of integration times at higher luminance, have been recognized for some time (as is clearly discussed).

    We agree, most of the general effects we observe have been observed in prior literature. However, the literature is also conflicted on most of the effects we observe. For example, whether or not the surround disappears at low light levels has been argued in prior literature. Similarly, the extent to which temporal receptive fields switch from differentiating to integrating (biphasic to monophasic) across light levels also has conflicting evidence in prior literature. Thus, we think our experiments add some clarity to this by presenting a unified set of stimuli to six simultaneously measured cell types.

    - The most novel finding is probably that one of the six types of RGC, the ON brisk-sustained, exhibited a proportionately larger change in integration time than other types, but this has not been placed in a functional context. The impact of the paper in terms of understanding retinal processing is unclear.

    We agree, the observation would be stronger if placed in a functional or behavioral context and we would love to do this! We are currently developing methods for inactivating particular RGC types and measuring the impact on behavior, but as the reviewer can appreciate, this is challenging. We hope in future studies to make intracellular measurements in these different RGC types to understand how some of these changes are reflected in (or produced by) their synaptic inputs.

    - A major drawback of all work on an MEA, usually ignored by users of this technique, is that diffusible neuromodulators are removed by perfusion. Neuromodulators can profoundly adjust processing through the retina and their levels can change in response to changes in luminance. The retinal network on an MEA is therefore likely to be far from its normal state. It is hard to know how far the ability of a retina on MEA to adapt reflects adaptation in vivo.

    100% agree! This is a very important point, and we have added some discussion on this topic as a critical caveat (lines 551-557). We have been performing experiments in mice with agonists and antagonists of D1 and/or D2 receptors to see if we can manipulate particular aspects of adaptation across different RGC types. So far, we have struggled to see much of an effect on spatial or temporal receptive field properties across light levels. The effects we do see are mostly limited to spike latencies and the gain of the contrast response functions, but these are quite subtle effects in our experiments thus far. That said, we think this is very likely an issue of not yet landing on the right drug concentrations, stimulus conditions, and possibly not waiting long enough for the effects of the neuromodulators to take full effect. Understanding how neuromodulators impact retinal adaptation is certainly important and a downside of our ex vivo experiments is that they may remove some of that from the equation. Nevertheless, we think this is a caveat that can and should be addressed in the discussion, not a fatal flaw of the experimental design.

    - An important control, reversing the direction of the luminance change, has not been carried out. If stepping back from a higher light level to a lower level does not (at least partially) reverse the changes caused by an increase in luminance, it may be that the lack of a pigment epithelium is affecting the results more severely than estimated by calculation.

    We also agree with this. We raised the light level because we are not doing our measurements with the pigment epithelium (PE) attached. If we start at high light levels and work backwards, the retina minimally sensitizes to the lower light levels, especially when switching from cone to rod-mediated conditions. We are measuring RGC responses without the PE because we cannot isolate the retina and PE from the sclera in rat (or mouse). Measuring RGC responses with the sclera attached presents problems with flattening the retina against the electrode array and presents a major diffusion barrier for oxygen to reach the retina. This is a limitation of the approach, however, as stated in the Discussion, we think that our measurements at 0.1-10 R*/rod/s and 10,000 R*/rod/s (low scotopic and low photopic respectively) are minimally subjected to pigment bleaching and thus are likely to accurately reflect in vivo steady-state conditions. The core result, that adaptation is largely ‘proportional’ still holds even if results at the light levels which are subject to photopigment bleaching are removed from the study.

    Reviewer #3 (Public Review):

    There have been variable reports on the degree to which light adaptation can modify a retinal ganglion cell's receptive field properties. The classic description suggested that the receptive field would become more low-pass in both space and time in dim light, better to integrate when signals are weak. More recent studies, particularly Tikidji-Hamburyan et al. (2015), suggested more radical changes in receptive fields across light levels, raising questions about how downstream circuits could keep track of highly variable output of the same ganglion cell, depending on the mean luminance.

    The current study revisits this problem with a straightforward set of experiments: measuring spikes in rat retinal ganglion cells, using a multi-electrode array, and presenting either spatiotemporal white noise or drifting gratings while varying the mean intensity over a one-million fold range. The technical achievement of this work is the careful and quantitative characterization of receptive field properties of specific cell types and the careful tracking of individual cells across light levels using each cell's electrophysiological image.

    The current study doesn't exactly agree with either of the extreme views from the literature. That is, the six ganglion cell types studied here mostly adapted in a similar way - becoming tuned to higher temporal and spatial frequencies in bright light. However, the receptive field did not become extremely low-pass at the dimmest light level, and there were exceptions to how individual cell types adapted, with some types showing a unique pattern, e.g., the ON brisk-sustained cell shows a particularly large shift in the peak temporal frequency in bright light. It's not exactly clear what to make of these unique patterns, which were rare, and it suggests that downstream circuits must somehow be attuned to these patterns or at least there is some benefit to these patterns that we do not currently understand.

    The study offers many technical strengths and is written clearly. If there is a weakness, it is that the new analysis does not clarify some of the discrepancies from previous studies, which had used various species, cell types and light levels that do not completely overlap with the current study. Still, if future studies followed the standard set in the current one, the field would certainly benefit.

    We thank the reviewer for their clear summary.