Credits: Prof. David Berson from Brown University
The following exercises will help your students get familiar with connectomic analysis of retinal neurons. There are about 40 types of retinal ganglion cells (RGCs) in the mammalian retina. Connectomic mapping is helping to provide definitive classifications of these RGC types by revealing their specific shapes, stratification patterns and synaptic connectivity. These connectivity maps are also revolutionizing our understanding of how specific patterns of synaptic input shape the functional properties of target neurons.
Using WEBKNOSSOS and the data annotated by the Berson lab, the students will explore structures of interest in their context. Next, they will be tracing (‘skeletonizing’) an unknown RGC. In the last task, they will be identifying the types of bipolar cells that provide excitatory input to a ganglion cell which has already been reconstructed. In working through these exercises, students are doing the same things professional neuroanatomists do to build comprehensive connectomes for single cells, for single cell types, and for all the cell types in that chunk of the CNS.
Setup
Open webknossos.org and create a free account. Open this link and click “Copy to my account”. You should be able to see something like the image below.
Take some time to get familiar with the WEBKNOSSOS interface. Play with the layers and settings in the left-side panel, take a look at the toolbar and menu on the top of the screen, and explore the different tabs on the right-hand side panel. Find more information on how to use WEBKNOSSOS here.
Now, open the “skeleton” tab on the right-side panel to find a hierarchical listing of skeletons (trees) in several groups (folders). Note the four groups named TASK 0, 1, 2, or 3. You can see what is inside each group by clicking the + to the left of the group, or collapse with the — symbol.
Task 0 — Explore structures of interest
Expand the first group, “Task 0,” and select an entry, such as the first one, “Muller glial cell.” The location of the viewports displaying the EM will jump to the location of the node. You are now viewing a Muller glial cell. Navigate in the EM to examine the cell in its context. Scroll through the Z slices with the mouse wheel (or use the shortcuts “D” and “F”), zoom in and out with “I” and “O,” and navigate on the planes by holding and releasing the left mouse button.
Expanding the group TASK 0, selecting the first entry, and navigating through the EM with the shortcuts D, F, I, and O.
Explore each entry in the “Task 0” skeleton group, and use the viewports to familiarize yourself with the structures and understand them in their context.
Task 1 — Trace an unknown cell
Collapse the TASK 0 group, expand the TASK 1, and select the entry “E”. This should position your cursor in the three orthogonal EM views at the location of the single node in this “starter” skeleton.
Toggling skeleton groups, finding starter position by clicking on the entry “E” of the TASK 1 group, and observing the highlighted cell that will be traced.
Watch the following video to learn how to trace (skeletonize) a cell of interest.
Tracing (skeletonizing) a cell by scrolling through the Z direction with the space bar and placing nodes by left-clicking.
Now, you are ready to start tracing!
Click on any one of the starter nodes listed with a single uppercase letter under TASK 1 in the Skeleton tab.
Find your node in the EM views and scroll back and forth through a few slices. Try to identify the plasma membrane of the dendrite the nodes lies within.
Making sure the “Skeleton tool” is selected, add your first node somewhere within the same process, not too far from the starter node. You should be able to see the new node appear in the 3D window as well as in each of the three EM planes. It will be a different color than the starter node because that starter node was entered as a branch point, from which the process extends in two directions. Typing “B” on the keyboard will mark your node as a branchpoint; “J” will bring you back to the last branchpoint marked.
Keep adding more nodes inside the same process. The distance between nodes can be whatever is a comfortable one for seeing what you’re doing. You might start close together at first, then get sparser with your node entry as you get more comfortable, which will speed things up. Choose a distance that makes it easy for you to keep track of what you are doing.
Feel free to generate skeletons for more than one starter node. If you are struggling with a particularly tricky spot, just click a different starter node under TASK 1 and start again from there. All your work will be saved in your annotation.
Spend at least 45 minutes building up one or more of these skeletons. It should be possible to enter hundreds of nodes in this amount of time, but don’t worry if you are slow and don’t manage to enter this many nodes.
WEBKNOSSOS will save your work in your account so you do not have to complete this in one sitting. When you are ready to continue, simply log back in, go to your Dashboard (top left of the screen) and select Annotations. You should see your annotation there. Click on it, and you should be right back where you left off with your skeletons available for viewing in the Skeletons tab. Because the interface is all web based, you can do this from any machine with a good internet connection by logging back into WEBKNOSSOS.
Task 2 — Assessing stratification and mapping synaptic connectivity
Now you are going to visualize a pre-traced retinal ganglion cell and examine its relationships with several types of bipolar cells.
In mice, there are 15 types of bipolar cells that provide excitatory (glutamatergic) input to the dendrites of amacrine cells and ganglion cells in the inner plexiform layer. Like other retinal neurons, bipolar cells are classified by their unique morphology, responses to light stimuli, and pattern of gene expression. The schematic below (adapted from Shekhar et al., 2016) depicts the bipolar cell types of the mouse retina. OFF cone bipolars cells (orange) target their axons to the OFF sublayer of the IPL (gray band). ON cone bipolars (purple) and rod bipolars (teal) terminate in the ON sublayer.
The “GluMIs” (‘Glutamatergic Monopolar Interneurons) are weird because they are missing the expected process in the outer plexiform layer (at top of this figure) and thus lack any direct input from photoreceptors! Though not strictly a “bipolar” cell, GluMIs seem in other respects just like another bipolar cell type.
Bipolar cell types of the mouse retina, adapted from Shekhar et al., (2016)
The Berson lab has previously reconstructed most of the bipolar axons in this volume, and your annotation file contains all skeletons for three OFF cone bipolar cell types:
GluMI
Type 1
Type 2
Notice that the axons of these 3 types terminate in the outer (upper) half of the OFF sublayer of the IPL.
1. Looking at the bipolar cells
Head to the Skeletons tab, expand the “Bipolar Cells” group to reveal three subgroups, one for each type. Each of these contains the skeletons of all the cells of that type, a digital representation of their axon terminals. If you like, you can view the individual skeletons of the type by expanding that bipolar skeleton group.
Expanding TASK 2, toggling the visibility of the first subgroup, changing the particle size, navigating in 3D.
Use the 3D display to view these skeletons from various angles, and in motion. Adjust the “particle size” in the left-hand side panel to view the skeletons more clearly in 3D. By turning on and off various trees or tree groups, you begin to visualize the real anatomical relationships between them. Do they touch? Do they stratify at the same level of the IPL, or in distinct strata?
Now, turn only the Type 2 bipolars on. You should see something like this:
Notice how the full collection of axon terminals of this bipolar type forms a mosaic. Their largely non-overlapping terminal arbors “tile” the retinal space, fitting together like the pieces of a jigsaw puzzle. This is a glimpse of how a single retinal cell type divides the retinal image (and thus visual space) up into separate processing chunks, like the pixels in a digital image.
Is this ‘tiling’ behavior unique to the Type 2 bipolar cells?
View the distribution of the axon terminals of the other two types in this same orientation. Do you see any differences in mosaic behavior? Or in any other features (e.g., terminal-arbor diameter?)
Now, click the XZ switch on the 3D view to get a side (edge-on) view of this mosaic of terminals. The retina is a bit tilted within the block, so tweak this orientation a bit with right-click-drag to get it truly perpendicular to the retinal layers. That’s when the bands of terminals are as thin as they can be and you are looking as close to true “edge-on” as possible.
Moving in the 3D viewport. Clicking XZ to get a side view of the skeleton group.
Turn on and off the three bipolar types in various combinations. You should be able to confirm what was shown earlier in the first figure. The terminal arbors of these bipolar types overlap a lot in depth — they “co-stratify”.
2. Looking at the ganglion cell
Turn off the bipolar skeletons and any others that are on and turn on the GANGLION CELL tree.
Turning off the bipolar skeletons and turning the GANGLION CELL on.
View the skeleton in both the YZ and XZ planes. Note that the cell body clues you in to retinal orientation. It lies in the ganglion-cell layer near the vitreous, while the bipolar terminals we have been looking at are as far away from the ganglion cell layer as can be.
Compare the stratification of the ganglion cell to that of the axon terminals of the three bipolars types. Is the ganglion cell stratifying in a way that would put it in a position to get synaptic input from some or all of these bipolar types? Though we haven’t got any info on actual synapses yet, can we at least agree that these processes co-stratify?
3. Looking at the ribbon synapses
Next, we are going to study the ribbon synapses that this ganglion cell actually receives. Turn the RIBBONS on and look at their relationship to the dendritic arbor of the ganglion cell. Are they all over the dendritic tree, or restricted to specific locations? (hint: you may want to adjust Particle Size).
Turning ribbon synapses on, changing the particle size, and clicking on a node to jump to its location on the EM viewports.
Now we are ready to identify the sources of these ribbon inputs.
Assignment: Mapping synaptic inputs to a neuron of interest
Assignment description:
Turn on the bipolar skeletons, along with the ganglion cell and mapped ribbons. In the ribbons folder, click one of the single-node skeletons. This will bring you to a location in the EM slices where the ribbon was mapped.
Now try to find the skeleton of the bipolar that makes this synapse. Look around the mapped ribbon within the same process for nodes from a tree that belongs to the presynaptic bipolar cell. Pay attention to clipping distance and particle size (left panel settings) if you are struggling to see the skeletons clearly.
Identify the presynaptic cell by clicking on one of those nodes making sure it is in this same cell as the ribbon. You’ve now selected a different skeleton (a bipolar). The cell-type identity of that cell should appear in the naming box at the top of the Skeletons tab (you’ll also see that your new skeleton is highlighted in the tree list). Hold this type identity in short-term memory!
Use your standard web browser ‘undo’ command (Ctrl/Cmd + Z) to jump back to the mapped ribbon whose source you just identified. If this doesn’t work, just click on the original ribbon node if you can set it, or just navigate back in the tree hierarchy to where you first found it. Enter text into the naming box at the top of Skeletons tab to rename the ribbon (1, 2, or GluMI), replacing the placeholder text “<enter type here>”. You can’t rename it by clicking directly on the tree list.
Click on a new unidentified node in the list and repeat this process for as many mapped ribbons as you can for 30 min. You should be seeing more and more identified bipolar inputs accumulating in your node list.
Total up the number of ribbon inputs you found from the three bipolar types. Does your data speak to the question of whether this ganglion cell gets its excitatory input from all the bipolar types that costratify with it?
Submission of work:
In the tree list, turn off all tree and tree groupings. Now turn on only the following trees or groups (this means you will just be sending trees that show your work not yet another copy of all those bipolar skeletons):
- TASK 1 (this contains your ganglion-cell skeletons)
- RIBBONS to ganglion cell (this is within TASK 2, but leave bipolar cells and ganglion cell unchecked)
At the top right of the Skeletons tab, click the More button with the down arrow. In the dropdown box, select “Download Selected Trees”. Save the file and change the name to include your first and last name.
You’re done! You generated a mini connectome by mapping the synaptic input to a neuron of interest. Check out the sources below that show how this work can answer questions about neural circuits. If you want more challenge, see if you can reconstruct a trickier cell than a ganglion cell by visiting TASK 3, where you’ll find starter skeletons for a bipolar cell and an amacrine cell, as well as yet another ganglion cell. Or choose any cell to reconstruct that you choose in the volume. You might discover something interesting!
Briggman KL, Helmstaedter M, Denk W. Wiring specificity in the direction-selectivity circuit of the retina. Nature. 2011 Mar 10;471(7337):183–8. doi: 10.1038/nature09818. PMID: 21390125.
Greene, M. J., Kim, J. S., & Seung, H. S. (2016). Analogous convergence of sustained and transient inputs in parallel on and off pathways for retinal motion computation. Cell reports, 14(8), 1892–1900.