Modulation of input sensitivity and output gain by retinal amacrine cells

  1. Department of Electrical Engineering, Stanford University, Stanford, CA 94305
  2. Department of Electrical & Computer Engineering, Department of Ophthalmology & Visual Sciences, University of Utah, Salt Lake City, UT 84112
  3. Department of Neurobiology, Stanford University School of Medicine, Stanford, CA 94305
  4. Department of Neurosurgery, Cologne-Merheim Medical Center, Witten/Herdecke University School of Medicine, Germany


  • Reviewing Editor
    Fred Rieke
    University of Washington, Seattle, United States of America
  • Senior Editor
    Lois Smith
    Boston Children's Hospital, Boston, United States of America

Reviewer #1 (Public Review):

This paper studies how amacrine cells influence retinal output signals. The approach taken is unusually direct. First, the amacrine light response is characterized. Second, the properties of signaling between the amacrine cell and ganglion cells is characterized by injecting current into the amacrine cell while measuring ganglion cell spiking. Third, the ganglion cell light response is analyzed in terms of components produced by signaling pathways that go through the amacrine cell and those that do not. Interpretation of the results relies on several important and largely untested assumptions. If some of the concerns that this dependence produces can be reduced the paper would be substantially stronger.

Linear vs. nonlinear and direct vs. indirect
Influences of an amacrine cell on the ganglion cell response are separated into direct effects - in which the amacrine cell directly produces a component of the ganglion cell response - and indirect effects - in which the amacrine cell modulates component(s) of the ganglion cell response (e.g. lines 97-99). In various places direct and indirect are equated with linear and nonlinear. Importantly, this assumption forms the basis of the analysis in the paper. It is not clear why a direct pathway through the amacrine cell should be linear. For example, it seems entirely possible that nonlinear models would capture the amacrine cell light response better than linear models. Similarly, nonlinear models may better capture the transmission of signals from amacrine cells to ganglion cells. Clarity on this issue is essential to interpret the results in the paper. One example of this issue comes up in the sentence on line 233. The definition of modulation is precise but only in the context of the above assumptions.

Components of oSTA
The set of pre-spike stimuli that are orthogonal to the "direct" STA is used to characterize the "indirect" pathways conveying signals to a ganglion cell. For the reasons noted above, it is not clear that this is accurate. In addition, the text describes the PCs of this orthogonal stimulus ensemble as features. This is introduced in the paragraph starting on line 177, and this paragraph has the disclaimer that these features do not correspond to neural pathways. That important caveat to interpretation could be reiterated in the following text - particularly in discussing the different forms of modulation.

Related to this point, the analysis of Figures 3 and 4 relies on the PCs of this orthogonal stimulus ensemble. Since the PCs themselves do not map onto pathways or mechanisms, it is not clear how to interpret some of the results. For example, when you see a polarity shift along one of the PCs, what happens along others (for example, could they also be shifting polarity such that the net effect is a change in kinetics but not a change in polarity)? This also comes up in the paragraph on line 236, as it is not clear how the separation works given the way the components used as the basis of the separation are defined.

Some of these issues are clarified in Figure 4D, and perhaps it would help to start with that description. I think this section would be much clearer if two types of modulation were noted and then it was laid out how that conclusion was reached.

Reviewer #2 (Public Review):

The authors analyze how individual amacrine cells in the salamander retina can affect the sensitivity of retinal ganglion cells to different visual features. They use simultaneous recordings of amacrine and ganglion cells and apply current injection into the amacrine cells to assess the evoked response modulation of ganglion cells. The resulting transmission filter is combined with the amacrine cell's temporal receptive field to determine a visual feature that stands for the visual signal processing from stimulus to a ganglion cell via the recorded amacrine cell. This sets the stage for analyzing how activation of this "amacrine pathway" affects the encoding of other (orthogonal) visual features by the ganglion cell.

The direct measurements of amacrine cell signals and their signal transmission to ganglion cells in challenging dual recordings is certainly a strength of this paper. In addition, the authors use an original and intriguing computational framework to analyze interactions of different visual features encoded by a ganglion cell and ask important questions about how inhibitory interneurons modulate stimulus encoding. The concept of distinct types of amacrine cell function with feature-specific modulation of input sensitivity and global modulation of output strength is thought-provoking and an interesting concept for follow-up investigations.

However, despite the emphasis on a causal approach and direct measurements of amacrine cell effects, the paper does not use actual amacrine cell signals for the main analyses, but rather a proxy given by visual signals that are consistent with the amacrine-to-ganglion signal transmission. In doing so, it is largely disregarded that visual filters of other pathways (including, e.g., fatigue or desensitization in the excitatory signals) may overlap with the deduced amacrine pathways. It thus remains unclear how much such alternative pathways may contribute to the signals assigned to the amacrine pathway and how this might influence the findings and their interpretations. In addition, the analysis and interpretation of the amacrine pathway are hard to follow and easy to misunderstand, because the paper often applies ambiguous language by referring to the visual stimulus dimension of the amacrine pathway as "amacrine output" and "amacrine effects" and by equating activation or deactivation of the amacrine pathway with hyperpolarization or depolarization of the amacrine cell.

Some other interpretations are also unclear, by taking the results a bit too far. For example, the emphasis on divisive normalization remains unclear, as divisive normalization seems more specific than the general suppressive effects described here. Similarly, the connection to the previously observed reversal of preferred contrast by ganglion cells is somewhat tenuous. Here, the potential reversal in the analyzed response nonlinearities only concerns specific features that nonlinearly interact with other features and therefore do not easily translate to the contrast sensitivity of the ganglion cell as a whole, as is suggested in the text. In addition, the two examples of reversals shown in the figures are not fully convincing.

Regarding the clustering analysis of the pairs of amacrine cells and ganglion cell features (Fig. 4), a specific concern is that it is unclear how well the analyzed parameters can actually be extracted from the firing rate response nonlinearities. From the examples in Fig. 4A, it looks like many nonlinearities do not show a clear saturation (but might still yield a good fit by the piece-wise linear model and thus be included in the analysis). It seems plausible that this could result in a bias towards lower gain (defined via the saturation level) when nonlinearities are shifted rightward (higher threshold). It is thus not entirely clear how strong the evidence is for the correlation between gain and threshold changes.

Further, minor caveats are that only 11 amacrine cells go into the analysis, and it remains uncertain to what degree they cover the diversity of amacrine cells in the retina or rather represent a specific subset of types. Also, the restriction to visual signals with no spatial structure, though understandable, limits the generality of the findings. The extracted temporal features remain rather abstract with unspecified significance, in particular since quite a large number of features are extracted per ganglion cell (a total of 321 features, which presumably come from 39 ganglion cells that had a significant amacrine transmission filter).

Reviewer #3 (Public Review):

This study aims to provide a generalizable definition of retinal amacrine cell function in visual processing. The authors used larval tiger salamander retinas and white noise stimulus to measure the retinal ganglion cell responses with multielectrode array recording, while either measuring individual amacrine cell membrane potential or stimulating the amacrine cell by injecting white noise currents using a sharp electrode. Modulatory effects of an amacrine cell on ganglion cells are analyzed by a computational framework that parses the signaling processing underlying ganglion cell responses into multiple conceptual pathways that are differentially subject to the amacrine cell signaling. The authors conclude that an individual amacrine cell can have diverse modulatory effects on ganglion cell responses. One class of effects modulates the sensitivity of the ganglion cell to specific visual features, while the other class of effects modulates the gain of responses to all features.

Amacrine cells are known for their remarkable cell type diversity and serve as key players underlying the complexity of computations performed by the vertebrate retina. However, their functions largely remain a mystery except for a few better-studied cell types. Therefore, the topic of this study is important. Furthermore, the study aims to extract general computational functions from these neurons, which will have broader applications to sensory processing beyond the retina. My main questions are centered around the interpretation of the computational analysis. First of all, the definition of a "visual feature" in this study using the white noise stimulus is different from that used in many other retinal studies using more structured stimuli than white noise. In this study, a major finding is that amacrine cells can control the sensitivity of specific visual features of the ganglion cell. However, it is difficult to gain intuition about how such feature specificity is related to the processing of other artificial and natural stimuli. More discussion along this line will help to clarify the significance of this result.

Another concern is the assumption that the somatic membrane potential of the amacrine cell represents its transmission property to ganglion cells. There are compelling examples that amacrine cells often exhibit local response properties that dramatically differ across the dendritic arbor and the soma (e.g. AIIs, Vlgut3+ ACs, starburst amacrine cells, A17s). This potential (and likely) complication should be addressed.

The dataset in this study is from 8 sustained and 3 transient amacrine cells. Immediate questions are: do all sustained or all transient cells belong to the same cell type in terms of functional properties or morphology? Is there any difference in the modulatory effects between the sustained and transient groups?

There is a rich body of literature on the functions of various amacrine cell types in the mammalian retina in shaping the receptive field properties, gain, and sensitivity of retinal ganglion cells. It would help the reader if the novelty of the current study is adequately discussed in the context of previous work.

One concern of sharp electrode recordings is the dialysis of intracellular solution into the cytoplasm, causing changes in membrane properties over time (e.g. Hooper et al., 2015). Have the authors examined the data obtained at the earlier and later phases of the recording to assess the potential effect of dialysis?

  1. Howard Hughes Medical Institute
  2. Wellcome Trust
  3. Max-Planck-Gesellschaft
  4. Knut and Alice Wallenberg Foundation