Spatiotemporal properties of glutamate input support direction selectivity in the dendrites of retinal starburst amacrine cells

Curation statements for this article:
  • Curated by eLife

    eLife logo

    Evaluation Summary:

    This is an important paper that addresses a key mechanism that underlies the canonical computation of direction selectivity in the retina. By using fluorescence imaging of glutamate release from excitatory interneurons combined with a computational model of dendritic integration, the authors make a convincing case that the kinetics of glutamate release contributes to the direction-selectivity of individual neural processes in retinal neurons. This work will appeal to visual neuroscientists as well as cellular physiologists interested in dendritic computations.

    (This preprint has been reviewed by eLife. We include the public reviews from the reviewers here; the authors also receive private feedback with suggested changes to the manuscript. Reviewer #3 agreed to share their name with the authors.)

This article has been Reviewed by the following groups

Read the full article See related articles

Abstract

The asymmetric summation of kinetically distinct glutamate inputs across the dendrites of retinal ‘starburst’ amacrine cells is one of the several mechanisms that have been proposed to underlie their direction-selective properties, but experimentally verifying input kinetics has been a challenge. Here, we used two-photon glutamate sensor (iGluSnFR) imaging to directly measure the input kinetics across individual starburst dendrites. We found that signals measured from proximal dendrites were relatively sustained compared to those measured from distal dendrites. These differences were observed across a range of stimulus sizes and appeared to be shaped mainly by excitatory rather than inhibitory network interactions. Temporal deconvolution analysis suggests that the steady-state vesicle release rate was ~3 times larger at proximal sites compared to distal sites. Using a connectomics-inspired computational model, we demonstrate that input kinetics play an important role in shaping direction selectivity at low stimulus velocities. Taken together, these results provide direct support for the ‘space-time wiring’ model for direction selectivity.

Article activity feed

  1. Author Response

    Reviewer #1 (Public Review):

    According to the space-time wiring hypothesis proposed by (Kim, Greene et al. 2014), the BC-off SAC circuit mimics the structure of a Reichardt detector; BCs closer to SAC soma have slower dynamics (they can be more sustained, have a delay in activation or slower rise time), while BCs further away are more transient. Later studies confirmed the connectivity and expanded the model on SACs (Ding, Smith et al. 2016, Greene, Kim et al. 2016). However, physiological studies that used somatic recordings to assess the BC properties at different dendritic distances were inconclusive (Stincic, Smith et al. 2016, Fransen and Borghuis 2017). Here, the authors used iGluSnFR, a glutamate sensor to measure the signals impinging on SAC dendrites. Their experimental findings align with the space-time wiring hypothesis, revealing sustained responses closer to SAC soma (mediated by prolonged release from type 7 BCs, and only slightly affected by amacrine cells), which according to their simulated SAC should produce a substantial increase in direction selectivity (DS).

    I find the work to be clear and well presented. However, I do have some reservations with the findings:

    Main points:

    1. Very low number of cells examined in the key experiment presented in the first figure. The authors used a viral approach to express flex- iGluSnFR in SACs in Chat-Cre mice. Sometimes (apparently twice) the construct was expressed in individual SACs - this is a very underpowered experiment! The low number of successes precludes adequately judging the validity of the findings.

    We agree with the reviewer that measuring iGluSnFR signals from single starburst dendrites is a powerful approach to confirm space-time wiring hypothesis. To bolster our data, we doubled our n number (updated Figure 1C and D, n = 66 ROIs; 20 dendrites and 4 retinas/FOVs). It should be noted that the results from these experiments are also validated on a larger scale across the starburst plexus (Figure 2).

    1. The model doesn't represent key known properties of BC-SACs and the interactions within SAC dendrites. First, the authors decided to construct a ball and stick model that doesn't consider the dendritic morphology of the starburst cell. A stimulus moving over a SAC is expected to engage multiple dendrites with complex spatiotemporal patterns that are expected to have a substantial effect on the voltages recorded on the investigated dendrite (Koren, Grove et al. 2017). For example, the dendrites in the orthogonal orientation will be activated at about the same time as the proximal dendrites; how such strong input will affect dendritic integration is unclear but should be taken into account in the model. Second, the authors assume a similar peak BC drive between proximal and distal inputs. However, a recent study found an enhanced glutamate release from proximal BCs, mediated by cholinergic SAC drive ((Hellmer, Hall et al. 2021); not cited). How different release amplitude would affect the conclusions of the model?

    It is well established that individual starburst dendritic sectors are relatively electrically isolated from each other (Miller & Bloomfield, 1983; Euler et al., 2002; Tukker et al., 2004, Poleg-Polsky et al., 2018) and thus we used a simple ball and stick to model direction selectivity in starburst dendrites.

    Related to the Reviewer’s point, in the paper we explicitly acknowledge that the simple ball and stick model will not capture important network interactions that are expected to impact direction selectivity (e.g. SAC-SAC inhibition). We suggest this as a future line of investigation.

    The idea of different synaptic weights across the starburst dendrite is an interesting one. If the proximal inputs are stronger relative to distal ones as the Reviewer suggests, it might be expected that the direction selectivity will be further enhanced. However, in a preliminary analysis, we did not find strong evidence for directionselectivity or sensitivity to MLA, to support the idea of cholinergic modulation.

    1. Another reason for including an accurate dendritic morphology is in the differences in the number of BCs that target a cell. Because SAC dendrites cover the entire receptive field area, type 7 BCs, which occupy the proximal third of the dendrites (Ding, Smith et al. 2016, Greene, Kim et al. 2016), are expected to cover only 11% of the area covered by SAC dendrites (1/3 x 1/3 = 1/9) and correspondingly mediate just 11% of the BC drive. A nonbifurcating model presented here would dramatically overrepresent their contribution to SAC responses. ??

    We have estimated BC numbers directly from the connectomics data which takes into account starburst morphology (Ding et al., 2016). To capture the heterogeneity of BCs that might be encountered at the level of single dendrites, in the revised manuscript, we have averaged responses over many trials in which the precise BC numbers varied according to the probability density functions observed in the connectomics data set (Ding et al., 2016). The details of the model parameters are now provided in the Methods section.

    1. (Fransen and Borghuis 2017) found that off-SACs have a more pronounced distinction in the time to peak than on-SACs. I found it surprising that given the large body of work demonstrating the effectivity of the viral approach in expressing iGluSnFR in off BC (Borghuis, Marvin et al. 2013, Franke, Berens et al. 2017, Szatko, Korympidou et al. 2020, Gaynes, Budoff et al. 2021, Strauss, Korympidou et al. 2021), that the authors did not compare between on and off SAC populations.

    It is possible that the kinetic differences are more pronounced for inputs to OFF starbursts. However, we observed a weaker iGluSnFR expression in the OFF starburst layer and the S/N was below what was required for our analysis. Therefore, we focused on the ON starburst.

    1. Recent work (Gaynes, Budoff et al. 2021) suggests that BCs' responses to motion and to static flashes have distinct dynamics. However, the current manuscript tests responses to flashed stationary stimuli experimentally, and then combines them in a simulation modeling a moving stimulus. This potential limitation of the study should at least be discussed.

    The Reviewer correctly points out that static and motion stimuli might have distinct dynamics (especially ‘emerging’ stimuli). We now describe this limitation of our study and discuss the findings of Gaynes et al. (2022)

    We have revised our model to take BC release rates truncated according to stimulus velocity and size to more appropriately represent the duration of the stimulus.

    Reviewer #2 (Public Review):

    The authors present a nice series of imaging experiments confirming previous anatomical and electrophysiological evidence for the "space-time wiring" model for directionally selective responses in SAC dendrites. Fluorescence measurements with a genetically encoded glutamate indicator show that excitatory inputs to proximal SAC dendrites are more sustained than distal dendrites. Although the signals are shaped by surround inhibition, the fundamental differences persist with inhibition blocked, suggesting intrinsic differences in the synaptic release processes in different cone bipolar cell types.

    The authors examine iGluSnFR dynamics in individual SACs (Figure 1) and in a population of SACs (Figure 2). The latter is possible because distal inputs to all SACs occur deeper in the IPL and so can be imaged separately from the proximal inputs, and it permits the measurement of many more synapses in each experiment. The former approach is particularly powerful, however, because it allows careful mapping of the different types of inputs along the dendritic axis of individual SACs. This experiment was performed in only seven dendrites in two retinas, however; consequently, the confidence intervals for any spatial fitting would be quite broad. This experiment would be strengthened with additional data from more dendrites.

    We have now the increased n number for Figure 1 in the revised manuscript (updated Figure 1C and D, n = 66 ROIs; 20 dendrites and 4 retinas/FOVs). Please see response to the Reviewer #1.

    It is very interesting that white noise stimuli do not pull out the kinetic differences - interesting enough to merit inclusion in the primary figures rather than as a supplement. These results seem valuable to our understanding of DS processing, but the implications remain unclear. Is it really the case that DS is eliminated - or even substantially degraded - when motion stimuli are presented atop some background (i.e., conditions in which the circuit is continuously stimulated)? Are the distinct kinetics are brought about by abrupt, large changes in luminance - if so, wouldn't one expect much weaker DS in response to drifting sinusoidal gratings?

    The question of how direction is encoded in the natural scene is very interesting but beyond the scope of this study. We presented a preliminary white noise analysis to show that our recordings are consistent with other recent reports (e.g. Strauss et al., 2022) which cast doubt on the space-time wiring model, rather than to directy address this specific issue. It should also be noted that complementary inhibitory and/or other intrinsic dendritic mechanisms may ensure that dendrites continue to remain DS in regimes in which BC mechanisms appear to be ineffective.

    In the introduction (p. 3A) the authors suggest that space clamp errors could distort EPSC kinetics, causing EPSCs arriving distally to appear more transient than those arriving more proximally. This seems contrary to what one would typically expect: cable theory would predict that more distal inputs ought to be filtered more, therefore appearing more prolonged, not more transient, than proximal inputs. It does not seem necessary to cast doubt on the previous results (Fransen and Borghuis, 2017) to motivate sufficiently the present experiments. One might simply point out that electrophysiological recordings do not provide precise information regarding the anatomical location of synaptic inputs.

    In the revised manuscript, we changed the text according to the Reviewer’s suggestion.

    Reviewer #3 (Public Review):

    In the study "Spatiotemporal properties of glutamate input support direction selectivity in the dendrites of retinal starburst amacrine cells", Srivastava, deRosenroll, and colleagues study the role of excitatory inputs in generating direction selectivity in the mouse retina. Computational and anatomical studies have suggested that the "space-time-wiring" model contributes to direction-selective responses in the mammalian retina. This model relies on temporally distinct excitatory inputs that are offset in space, thereby yielding stronger responses for motion in one versus the other direction. Conceptually, this is similar to the Reichardt detector of motion detection proposed many decades ago. So far, however, there is little functional evidence for the implementation of the space-time-wiring model. Here, Srivastava, deRosenroll and colleagues use local glutamate imaging in the ex-vivo mouse retina combined with biophysical modeling to test whether temporally distinct and spatially offset excitatory inputs might generate direction-selective responses in starburst amacrine cells (SACs). Consistent with the space-time-wiring model, they find that glutamatergic inputs at proximal SAC dendrites are more sustained than inputs at distal dendrites. This finding was consistent across different sizes of stationary, flashed stimuli. They further linked the sustained input component to the genetically identified type 7 bipolar cell and showed that the difference in temporal responses across proximal and distal inputs was independent of inhibition, but rather relied on excitatory interactions. By estimating vesicle release rates and building a simple biophysical model, the authors suggest that next to already established mechanisms like asymmetric inhibition, excitatory inputs with distinct kinetics contribute to direction-selective responses in SACs for slow and relatively large stimuli.

    In general, this study is well-written, the data is clearly presented and the conclusion that (i) the temporal kinetics of excitatory inputs varies along SAC dendrites and that (ii) this might then contribute to direction selectivity is supported by the data. The study addresses the important question of how excitation contributes to the generation of direction-selective responses. There have been several other studies published on this topic recently, and I believe that the results will be of great interest to the visual neuroscience community.

    However, the authors should address the following concerns:

    • They should demonstrate that differences in response kinetics between proximal and distal dendrites are unrelated to differences in signal-to-noise ratio.

    In response to the Reviewer’s comment, we have now added new plots to supplementary Figure S1 (A, B) that show that the response kinetics are not strongly related to signal strength.

    • To demonstrate consistency across recordings/mice, the authors should indicate data points from different recordings (e.g. Fig. 2C).

    In the updated Figure 2C-E, we have now added the average values for each recording to indicate the consistency/variability in the data.

    • The authors mention in the introduction that the space-time-wiring model is conceptually similar to other correlation-type motion detectors that have been experimentally verified in different species. It would be great to expand on the similarity and differences of the different mechanisms in the Discussion, especially focusing on Drosophila where experimental evidence at the synaptic level exists.

    It should be noted that the results of the influential Nature paper describing the spacetime wiring of inputs to T4 DS neurons in the fly system were not reproduced by the same group. A new paper from Axel Borst’s group, however, shows a distinct source of spatially offset excitation (glutamate-mediated by disinhibition) may underlie the multiplicative operation. Nevertheless, to our knowledge, no studies have mapped out the spatiotemporal properties of inputs across single T4/T5 DS neurons as we have done for the starburst. In the revised manuscript we briefly summarize the fly literature in response to the Reviewer’s suggestion.

    • The authors use stationary spot stimuli of different sizes to characterize the response kinetics of excitatory inputs to SACs. I suggest the authors add an explanation for choosing only stationary stimuli for studying the role of excitatory inputs in direction selectivity/motion processing.

    Please see the above response to Reviewer #1.

    In addition, the authors use simulated moving edges to stimulate the model bipolar cells. They should provide details about the size of the stimulus and the rationale behind using this size, given their previous results.

    For simulation experiments, bipolar cell inputs were triggered by a 400 µm wide bar moving over a range of velocities (0.1 – 2 mm/s). We have now added more details in the Methods section and main text in the revised manuscript.

    • Using the biophysical model, the authors show that converting sustained bipolar cell inputs to transient ones reduces direction selectivity in SACs. I suggest the authors also do the opposite manipulation/flip the proximal and distal inputs or provide a rationale why they performed this specific manipulation.

    Thanks for the suggestion. We have now updated Figure 6B showing DSi vs velocity plots for several different bipolar cell input distributions – (i) sustained-transient, (ii) transient-sustained, (iii) all transient, and (iv) all sustained.

    • In each figure, the authors should note whether traces show single trial responses or mean across how many trials. If the mean is presented (e.g. Suppl. Fig. 2a), the authors should include a measure of variability - either show single ROIs in addition and/or add an s.d. shading to the mean traces.

    In the revised manuscript, we have now indicated the mean and number of trials for each figure. We have also added S.E.M values to the mean traces in the figures.

  2. Evaluation Summary:

    This is an important paper that addresses a key mechanism that underlies the canonical computation of direction selectivity in the retina. By using fluorescence imaging of glutamate release from excitatory interneurons combined with a computational model of dendritic integration, the authors make a convincing case that the kinetics of glutamate release contributes to the direction-selectivity of individual neural processes in retinal neurons. This work will appeal to visual neuroscientists as well as cellular physiologists interested in dendritic computations.

    (This preprint has been reviewed by eLife. We include the public reviews from the reviewers here; the authors also receive private feedback with suggested changes to the manuscript. Reviewer #3 agreed to share their name with the authors.)

  3. Reviewer #1 (Public Review):

    According to the space-time wiring hypothesis proposed by (Kim, Greene et al. 2014), the BC-off SAC circuit mimics the structure of a Reichardt detector; BCs closer to SAC soma have slower dynamics (they can be more sustained, have a delay in activation or slower rise time), while BCs further away are more transient. Later studies confirmed the connectivity and expanded the model on SACs (Ding, Smith et al. 2016, Greene, Kim et al. 2016). However, physiological studies that used somatic recordings to assess the BC properties at different dendritic distances were inconclusive (Stincic, Smith et al. 2016, Fransen and Borghuis 2017). Here, the authors used iGluSnFR, a glutamate sensor to measure the signals impinging on SAC dendrites. Their experimental findings align with the space-time wiring hypothesis, revealing sustained responses closer to SAC soma (mediated by prolonged release from type 7 BCs, and only slightly affected by amacrine cells), which according to their simulated SAC should produce a substantial increase in direction selectivity (DS).

    I find the work to be clear and well presented. However, I do have some reservations with the findings:

    Main points:
    1. Very low number of cells examined in the key experiment presented in the first figure. The authors used a viral approach to express flex- iGluSnFR in SACs in Chat-Cre mice. Sometimes (apparently twice) the construct was expressed in individual SACs - this is a very underpowered experiment! The low number of successes precludes adequately judging the validity of the findings.
    2. The model doesn't represent key known properties of BC-SACs and the interactions within SAC dendrites. First, the authors decided to construct a ball and stick model that doesn't consider the dendritic morphology of the starburst cell. A stimulus moving over a SAC is expected to engage multiple dendrites with complex spatiotemporal patterns that are expected to have a substantial effect on the voltages recorded on the investigated dendrite (Koren, Grove et al. 2017). For example, the dendrites in the orthogonal orientation will be activated at about the same time as the proximal dendrites; how such strong input will affect dendritic integration is unclear but should be taken into account in the model. Second, the authors assume a similar peak BC drive between proximal and distal inputs. However, a recent study found an enhanced glutamate release from proximal BCs, mediated by cholinergic SAC drive ((Hellmer, Hall et al. 2021); not cited). How different release amplitude would affect the conclusions of the model?
    3. Another reason for including an accurate dendritic morphology is in the differences in the number of BCs that target a cell. Because SAC dendrites cover the entire receptive field area, type 7 BCs, which occupy the proximal third of the dendrites (Ding, Smith et al. 2016, Greene, Kim et al. 2016), are expected to cover only 11% of the area covered by SAC dendrites (1/3 x 1/3 = 1/9) and correspondingly mediate just 11% of the BC drive. A non-bifurcating model presented here would dramatically overrepresent their contribution to SAC responses.
    4. (Fransen and Borghuis 2017) found that off-SACs have a more pronounced distinction in the time to peak than on-SACs. I found it surprising that given the large body of work demonstrating the effectivity of the viral approach in expressing iGluSnFR in off BC (Borghuis, Marvin et al. 2013, Franke, Berens et al. 2017, Szatko, Korympidou et al. 2020, Gaynes, Budoff et al. 2021, Strauss, Korympidou et al. 2021), that the authors did not compare between on and off SAC populations.
    5. Recent work (Gaynes, Budoff et al. 2021) suggests that BCs' responses to motion and to static flashes have distinct dynamics. However, the current manuscript tests responses to flashed stationary stimuli experimentally, and then combines them in a simulation modeling a moving stimulus. This potential limitation of the study should at least be discussed.

  4. Reviewer #2 (Public Review):

    The authors present a nice series of imaging experiments confirming previous anatomical and electrophysiological evidence for the "space-time wiring" model for directionally selective responses in SAC dendrites. Fluorescence measurements with a genetically encoded glutamate indicator show that excitatory inputs to proximal SAC dendrites are more sustained than distal dendrites. Although the signals are shaped by surround inhibition, the fundamental differences persist with inhibition blocked, suggesting intrinsic differences in the synaptic release processes in different cone bipolar cell types.

    The authors examine iGluSnFR dynamics in individual SACs (Figure 1) and in a population of SACs (Figure 2). The latter is possible because distal inputs to all SACs occur deeper in the IPL and so can be imaged separately from the proximal inputs, and it permits the measurement of many more synapses in each experiment. The former approach is particularly powerful, however, because it allows careful mapping of the different types of inputs along the dendritic axis of individual SACs. This experiment was performed in only seven dendrites in two retinas, however; consequently, the confidence intervals for any spatial fitting would be quite broad. This experiment would be strengthened with additional data from more dendrites.

    It is very interesting that white noise stimuli do not pull out the kinetic differences - interesting enough to merit inclusion in the primary figures rather than as a supplement. These results seem valuable to our understanding of DS processing, but the implications remain unclear. Is it really the case that DS is eliminated - or even substantially degraded - when motion stimuli are presented atop some background (i.e., conditions in which the circuit is continuously stimulated)? Are the distinct kinetics are brought about by abrupt, large changes in luminance - if so, wouldn't one expect much weaker DS in response to drifting sinusoidal gratings?

    In the introduction (p. 3A) the authors suggest that space clamp errors could distort EPSC kinetics, causing EPSCs arriving distally to appear more transient than those arriving more proximally. This seems contrary to what one would typically expect: cable theory would predict that more distal inputs ought to be filtered more, therefore appearing more prolonged, not more transient, than proximal inputs. It does not seem necessary to cast doubt on the previous results (Fransen and Borghuis, 2017) to motivate sufficiently the present experiments. One might simply point out that electrophysiological recordings do not provide precise information regarding the anatomical location of synaptic inputs.

  5. Reviewer #3 (Public Review):

    In the study "Spatiotemporal properties of glutamate input support direction selectivity in the dendrites of retinal starburst amacrine cells", Srivastava, deRosenroll, and colleagues study the role of excitatory inputs in generating direction selectivity in the mouse retina. Computational and anatomical studies have suggested that the "space-time-wiring" model contributes to direction-selective responses in the mammalian retina. This model relies on temporally distinct excitatory inputs that are offset in space, thereby yielding stronger responses for motion in one versus the other direction. Conceptually, this is similar to the Reichardt detector of motion detection proposed many decades ago. So far, however, there is little functional evidence for the implementation of the space-time-wiring model.

    Here, Srivastava, deRosenroll and colleagues use local glutamate imaging in the ex-vivo mouse retina combined with biophysical modeling to test whether temporally distinct and spatially offset excitatory inputs might generate direction-selective responses in starburst amacrine cells (SACs). Consistent with the space-time-wiring model, they find that glutamatergic inputs at proximal SAC dendrites are more sustained than inputs at distal dendrites. This finding was consistent across different sizes of stationary, flashed stimuli. They further linked the sustained input component to the genetically identified type 7 bipolar cell and showed that the difference in temporal responses across proximal and distal inputs was independent of inhibition, but rather relied on excitatory interactions. By estimating vesicle release rates and building a simple biophysical model, the authors suggest that next to already established mechanisms like asymmetric inhibition, excitatory inputs with distinct kinetics contribute to direction-selective responses in SACs for slow and relatively large stimuli.

    In general, this study is well-written, the data is clearly presented and the conclusion that (i) the temporal kinetics of excitatory inputs varies along SAC dendrites and that (ii) this might then contribute to direction selectivity is supported by the data. The study addresses the important question of how excitation contributes to the generation of direction-selective responses. There have been several other studies published on this topic recently, and I believe that the results will be of great interest to the visual neuroscience community.

    However, the authors should address the following concerns:
    - They should demonstrate that differences in response kinetics between proximal and distal dendrites are unrelated to differences in signal-to-noise ratio.
    - To demonstrate consistency across recordings/mice, the authors should indicate data points from different recordings (e.g. Fig. 2C).
    - The authors mention in the introduction that the space-time-wiring model is conceptually similar to other correlation-type motion detectors that have been experimentally verified in different species. It would be great to expand on the similarity and differences of the different mechanisms in the Discussion, especially focusing on Drosophila where experimental evidence at the synaptic level exists.
    - The authors use stationary spot stimuli of different sizes to characterize the response kinetics of excitatory inputs to SACs. I suggest the authors add an explanation for choosing only stationary stimuli for studying the role of excitatory inputs in direction selectivity/motion processing. In addition, the authors use simulated moving edges to stimulate the model bipolar cells. They should provide details about the size of the stimulus and the rationale behind using this size, given their previous results.
    - Using the biophysical model, the authors show that converting sustained bipolar cell inputs to transient ones reduces direction selectivity in SACs. I suggest the authors also do the opposite manipulation/flip the proximal and distal inputs or provide a rationale why they performed this specific manipulation.
    - In each figure, the authors should note whether traces show single trial responses or mean across how many trials. If the mean is presented (e.g. Suppl. Fig. 2a), the authors should include a measure of variability - either show single ROIs in addition and/or add an s.d. shading to the mean traces.