In Depth: Untangling Starburst Amacrine Cells

In a bet to unravel one of the longest standing controversies in visual neuroscience, and thus in neuroscience period, Briggman and friends have pioneered a new combination of calcium imaging and electron microscopy. This is the arguably the more impressive sibling of the paper by Bock and friends that I reviewed ealier. The techniques used are in essence the same, but there are a few key differences that helped this paper make some stronger conclusions.

Rather than working in the mouse visual cortex, where higher visual processing occurs, Briggman and friends chose to try untangling circuitry in the mouse retina. Historically, the retina is thought of as the site where visual information is broken down into its most basic parts – different colors, points of light/dark, and lines of specific orientations. Thus, you find many retinal ganglion cells – the output cells of the retina that project into the brain proper – that are selective for specific, basic visual stimuli. The thinking is that once the retina breaks a visual scene down into its basic parts, that info is transmitted into the visual cortex where the scene begins to be reconstructed, and as you move deeper into the visual processing streams in the brain, you eventually get single cells that have more and more complex preferred stimuli like directions of movement, patterns, shapes, etc. Even further down the line you get cells that respond to things like objects and faces, the most famous of which are cells that have been found in certain individuals to respond only to the likes of Jennifer Aniston, Halle Berry, or, every one’s favourite, OJ Simpson. However, like I said, the idea is that the retina only deals with the most basic components of the scene. Interestingly though, there are a minority of retinal ganglion cells in the retina that respond to movement of lines in a specific direction, notably, a population of direction selective retinal ganglion cells that respond to one of Up-, Down-, Left- or Right-ward movement. Movement in the opposite direction of the preferred stimulus direction, called the “null direction,” produces no response in the direction selective retinal ganglion cells.

There has been an unanswered question in the field of retinal circuitry since the discovery of direction selective retinal ganglion cells (DSGCs) 50 years ago; namely, what makes these cells direction selective? It turns out that the cells responsible for conferring this selectivity onto DSGCs are Starbust Amacrine Cells (SACs)  and they are pretty strange birds. First weird thing: they use 2 different neurotransmitters that have opposite actions on postsynaptic targets – GABA (inhibitory) and Acetylcholine (excitatory). The rule of thumb is that neurons only use one type of neurotransmitter, but SACs are an exception. Secondly, they have dendrites that both receive synapses and make synapses. Some of these synapses are onto DSGCs. Third, not only are SACs already direction selective themselves, but their individual dendrites respond preferentially (with calcium elevations) to movement in specific directions.

A number of things are also known about how SACs interact with DSGCs:

1.) SAC dendritic trees splay out and overlap with the dendritic trees of DSGCs.

2.) SACs make inhibitory GABAergic synapses with DSGCs.

3.) Without inhibitory input from SACs, DSGCs are not direction selective.

4.) SACs on the null side of a DSGC (ie the side from which the null stimulus approaches) inhibit that DSGC much more strongly than SACs on the preferred side (the side from which the preferred stimulus approaches).

This last point brings up an important question about how SACs confer direction selectivity onto DSGCs. The setup makes perfect sense; the layout of cells in the retina corresponds topographically with the visual field, meaning that the visual field falls directly onto the photorecptors, and the photoreceptors project perpendicularly and directly onto the SACs and DSGCs (via bipolar cells). The result is that when a stimulus moves through the visual field, the activity that it evokes in the photorecptors moves along with it, and thus so does the corresponding activity in the DSGCs and SACs. So, if you consider a single DSGC, when its preferred stimulus moves across the retina, the stimulus will hit an adjacent SAC first, but this preferred-side SAC does not inhibit that DSGC, so the DSGC becomes active when the stimulus hits it. When the stimulus gets past the DSGC, the null side SAC is activated, but the DSGC is already active and that SAC can’t inhibit the DSGC. However, for movement in the null direction, the strong, null-side SAC is activated first, and that strong inhibitory input that it makes onto the DSGC preempts activation, and as a result the DSGC doesn’t respond. This property of the circuit is called asymmetric input, and there are 2 theories as to how it arises:

1.) Null side SACs make more synapses onto the DSGC in question, making their inhibition stronger. This is a structural asymmetry.

2.) If the structure of the circuit is symmetrical, with all SACs making an equal number of synapses onto neighboring DSGCs, the asymmetry could be in synaptic strengths, with null side SACs making the same number, but stronger synapses onto the DSGC.

It is this asymmetry that Briggman and friends seek to understand. Their experimental design is very similar to that of Bock and friends’ work in the cortex. To start with, Briggman and friends also functionally imaged DSGCs using calcium indicators. To do this they presented the retina with lines moving in specific directions while imaging calcium responses. (The retina has the express advantage over the visual cortex in that it will still respond to visual stimuli when it has been removed from the eye and placed in a dish.) Then, the group used electron microscopy to take very fine scale images of retinal circuitry (lateral resolution of 16.5 x 16.5 nm). However, rather than slicing the retina into very thin slices and then imaging them like Bock and friends, Briggman and friends used a more modern approach termed serial block-face electron microscopy (SBEM). This involves imaging the retina as it is sliced, fixing the retina under the microscope, shaving off thin slices (23nm thick in this case) and then imaging the top of the remaining block of tissue after every slice is shaved off. This integrates the slicing and imaging processes, automating the process and making it much quicker. Following the imaging, the resulting 3D reconstruction was matched to the functionally imaged cells, allowing Briggman and friends to find their DSGCs in the piece of retina that they reconstructed. Of the 634 neurons that were functionally imaged, 25 of them responded to either up-, down-, left- or rightward movement. They found these 25 cells in the reconstructed volume (300 x 300 x 60 cubic microns), and traced 6 of them. They identified the SAC synapses onto these 6 DSGCs by the stereotypical bulkiness of their presynaptic terminals. They only found 24 synapses with this approach, but it allowed them to back trace each axon to a different SAC. So now they had 6 DSGCs and 24 presynaptic SACs. They then fully traced each SAC, finding the full fall of dendrites for each (remember that, aside from recieving input, SACs’ dendrites make presynaptic terminals onto DSGCs). To identify the rest of the synapses in between the 24 SACs and 6 DSGCs, they considered every point at which an SAC came within 1.5 microns of a DSGC a potential synapse, and then wittled the 9260 cases of this down to 831 true synapses based on the presence of one of those bulky presynaptic “varicosities.” Since the synaptic cleft separating pre- and postsynaptic terminals are generally somewhere on the order or 10 to 50 nm wide, I take issue with this process. There is definitely room in a 1.5 micron gap for a completely separate terminal to be made, and I wonder how closely they examined all of these large synapses for the presence of another postsynaptic terminal. Complicating the matter is the fact that they used a simplified staining protocol for the SBEM, which only labeled the surface of the neurons. Evidently a referee was unsure about this staining protocol and had them verify that the varicosities were indeed synapses by doing a similar but smaller scale analysis with a more traditional stain that labels pre- and postsynaptic machinery and vesicles, making it much easier to identify synapses with high fidelity. Briggman and friends found that yes, the varicosities they used to identify synapses contain marks of synapses every time. However, this doesn’t circumvent the likelihood that other synapses might squeeze into their 1.5 micron space.

Ignoring this potential confound, they go on to look at the location of the 24 SACs relative to the DSGCs they synapsed onto. They assess the layout of the circuitry in two ways. First they compare the direction in which the dendrites of SACs project relative to the location of the DSGCs and find that SAC dendrites tend to synapse on DSGCs that have a preferred stimulus that move in the opposite direction to the direction in which that branch of the SAC’s dendritic tree projects. Keep in mind that the direction in which a dendritic branch projects depends on the location of the cell body from which it emanates, so this finding essentially indicates that SACs preferentially make synapses with DSGCs with whom they have null side relationships. The second way that Briggman and friends examined the connections between these two cell types was to look at the circuitry from the synapses’ points of view. The group generated “dendrite angle” vectors that projected from each synapse toward the SAC cell body that made that synapse. What they found was that the direction of the these dendrite angle vectors was on average almost exactly opposite to that of the preferred direction of the DSGC in question (the mean difference between the two is 165° with a standard deviation of ~52°, so quite variable, but what likely counts in neural circuit computations is overall trends of connectivity). Both of these analyses come together to help Brigmman and friends make the conclusion that the asymmetry in synaptic input from SACs to DSGCs is indeed a structural asymmetry; the asymmetry is in the physical connectivity, not synaptic strengths.

And thus comes to a close a half-century of debate on how direction selectivity arises in the retina. But by no means is the retina solved! There are still something to the tune of 40 or 50 types of amacrine cells that could probably use some examining, so we will anxiously await full retinal connectome and see what comes as we approach it.

This entry was posted in Exciting Research, Home, In Depth, Uncategorized. Bookmark the permalink.

15 Responses to In Depth: Untangling Starburst Amacrine Cells

  1. Pingback: In Depth: Functional Connectomics of the Primary Visual Cortex |

  2. Pingback: Crash Course In Functional Connectomics |

  3. Pingback: In Depth: Neurons that Change Shape! Implications for Treatment of Depression? |

  4. Denise says:

    What makes the Starburst Amacrine Cells direction selective?

    How do the SACs know which RGCs to make more synapses with?

    Is there anything known about the development of this direction selectivity?

  5. Richard says:

    You wrote: “To identify the rest of the synapses in between the 24 SACs and 6 DSGCs, they considered every point at which an SAC came within 1.5 microns of a DSGC a potential synapse”…. ” I take issue with this process. There is definitely room in a 1.5 micron gap for a completely separate terminal to be made”
    This is indeed a problem with the Briggmann et al set up. SAC varicosities are within this range, however, Briggmann et al did not rule out the possibility of another SAC dendrite (most under 1 micron in diameter) to be laying directly between the indentified SAC varicosity and the DSGC dendrites.

  6. Richard says:

    Briggmann wrote: “To identify the preferred directions of On–Off DSGCs, we labelled the ganglion cell layer of an adult mouse retina by bulk electroporation
    with the membrane-impermeable formofOregon Green 488BAPTA-1,
    a calciumindicator. This avoids the damage thatwould inevitably result
    from the pipette penetration needed for acetoxymethyl ester-based
    load”. They used a bulk staining technique of the entire inner plexiform layer not penerating the individual starburst for the tracing.

  7. Richard says:

    You Wrote:’ Keep in mind that the direction in which a dendritic branch projects depends on the location of the cell body from which it emanates, so this finding essentially indicates that SACs preferentially make synapses with DSGCs with whom they have null side relationships.” The DSGCs are highly convoluted trees. How does this explain their results?

    • cksalmo says:

      Yes, you’re right that the trees are complex, but it seems to me that, although the branches are complex, individual branches don’t seem to overlap a whole lot. Thus, each branch covers a roughly separate field of photoreceptor input (via bipolar cells). Taking that into account, the individual SACs can receive specific input. Does that answer your question.

      • Richard says:

        Not really. Cofasciculation of SACs dendrties raises the intriguing possibility of asymmetrical bipolar input impinging on ACs within the plexus of the network, and which could play a functional role in directional selectivity. I think the 1.5 micron is void is not a void but they actually took all synapses within this range, so it is not a caveat. However, it is not clear how they recorded from DSGCs to decipher the null-preferred axis in trees that are “convoluted” meaning they zig-zag to the left and right, up and down.

  8. cksalmo says:

    1.) For their synapse identification, I was under the impression that they counted any presynaptic bouton that was 1.5um or less away from the dendrite as a synapse. And if you count a bouton that is 1.5um away from the dendrite as a synapse onto that dendrite, I think you are most definitely generating an overestimate of the number of synapses (since synaptic clefts are generally less than 40nm wide). However, when they traced that dendritic tree they did a line trace, not a contour trace. If they were measuring 1.5um from the line trace, that is better. But, as you say, an SAC dendrite can get down to 1um in diameter, so even if they measured from the line trace, they may have been oversampling.

    Although I think I understand your point – since SAC dendrites are cofasciculated, you would argue that all the synapses in the vicinity of particular dendrite, although not necessarily synapsing onto that single dendrite, are likely giving rise to the same direction selectivity in the cofasciculated SACs. That makes me somewhat less concerned with the validity of the conclusions made by studying these synapses. However, this process is still oversampling, so while the relative asymmetry in the input would likely be the same, their synapse counts for a single SAC are largely overestimated. This doesn’t matter too much for their conclusion that there is indeed a structural asymmetry giving rise to direction selectivity. But they did not generate this conclusion by accurately reproducing the wiring diagram – they had to oversample and create an unrealistic model rather than making conclusions from the actual network – in my opinion the study is somewhat weakened by this.

    2.) They recorded from the DSGCs with a calcium dye that they had to bulk-electroporate into the cells. They then give their directional stimulation and record the calcium responses from all the cells in a single plain. They are just looking at Ca responses in the the cell body as far as I can tell (the same group saw dendrite specific Ca responses in a previous paper, but that isn’t what they looked at here). The dye would have gotten into every cell type, but they identified the SACs afterward with their SBEM. So at that point they have the functional calcium-indicator recordings for direction selectivity, and then they also have the morphology to ID the SACs. I don’t see the problem there.

  9. Richard says:

    In the rabbit, these SACs can reach down to 0.2 um in their thinnest arbors. If they included any presynaptic bouton that was 1.5um or less away from the DSGC dendrite then their results did not show what was in the void or “gap”? If not then they indeed have oversampled. However, and more importantly their asymmetry constraint will longer be conclusive because those boutons that were “oversampled” are not necessarily connected to the DSGC if another SAC is inserted within the ‘”gap”.

  10. cksalmo says:

    Precisely the point I made in my post, although I made it less clearly. Thanks for hashing that one out.

  11. Richard says:

    Reading the absract to another Nature article piublished in 1997 we at a similar crossroads. They were sure that they ablated the SACs and yet DS still remained, until Yoshida et al (2001) published in NEURON proved them wrong. I believe the same will come out of this preliminary study given that they did oversample their data.

    Retinal direction selectivity after targeted laser ablation of starburst amacrine cells
    Shigang He & Richard H. Masland
    Howard Hughes Medical Institute, Massachusetts General Hospital, Boston,
    Massachusetts 02114, USA
    . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
    Directionally selective retinal ganglion cells respond strongly
    when a stimulus moves in their preferred direction, but respond
    little or not at all when it moves in the opposite direction1,2. This
    selectivity represents a classic paradigm of computation by neural
    microcircuits, but its cellular mechanism remains obscure. The
    directionally selective ganglion cells receive many synapses from a
    type of amacrine cell termed ‘starburst’ because of its regularly
    spaced, evenly radiating dendrites3,4. Starburst amacrine cells have
    a synaptic asymmetry that has been proposed as the source of the
    directional response in the ganglion cells5,6.Here we report experiments
    that make this unlikely, and offer an alternative concept of
    the function of starburst cells.We labelled starburst cells in living
    retinas, then killed them by targeted laser ablation while recording
    from individual directionally selective ganglion cells. Ablating
    starburst cells revealed no asymmetric contribution to the ganglion
    cell response. Instead of being direction discriminators, the
    starburst cells appear to potentiate generically the responses of
    ganglion cells to moving stimuli. The origin of direction selectivity
    probably lies with another type of amacrine cell.

  12. Richard says:

    How many retinas did they sample to arrive at their oversampled data?


Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s