confocal · inferior olive · mouse

THE OLIVE

A sorting game built from one neuroanatomy paper.

← the textbook "curly" kindthe textbook "straight" kind →

You have a stack of microscope images of neurons from a small part of the brainstem that the old anatomists named the inferior olive, because from the outside it looks like one. For a century the textbooks have said these cells come in two kinds: curly ones, whose branches twist and double back in a small ball around the cell body, and straight ones, whose few branches run far out. Round one: sort twenty cells. Round two: find the groups they form. One score at the end.

Labeling method

Built from Vrieler et al. (2019), Brain Structure and Function. About this game

VIRUScell 1 / 20

Curly = twists and branches densely close to the cell body. Straight = sparse, reaching far. If it's neither, say so — that answer counts too.

Each cell you sort drops onto a ruler of straightness: how far the tree reaches from the cell body, divided by the length of its longest branch. 0.1 is a tight ball, 0.9 a nearly straight line.

curlycan't tellstraight
round one · result

Your eye against the ruler

curlycan't tellstraight
— CONSISTENT

cell bodies · one slab0 / 12 markers

Every dot is one cell body in a slab of the olive about a third of a millimetre across. The paper says these cells form small groups that are wired tightly to each other — roughly ten to twelve groups in a slab this size, each a dozen or so cells. Tap the slab to drop a marker in the middle of each group you think you see; tap a marker to remove it. Then score.

session complete

What the olive taught you

—
your eye, consistency
—
groups, somata only
—
groups, dendrites lit
— SCORE

One. The two kinds don't exist. Ninety reconstructed cells, 25 measures, no split anywhere: curly and straight are the ends of a line, and a single number — how far a tree reaches divided by how long its longest path is — carries what the anatomists' eyes were doing.

Two. The cell bodies don't tell you where the groups are. At 40 µm the whole nucleus is one cluster. The structure is in which way the dendrites grow: most trees are lopsided, with the soma at the edge of its own field, and neighbours either share a field or avoid each other entirely. A few long straight cells reach across — the paper's "bridges".

Vrieler N., Loyola S., Yarden-Rabinowitz Y., Hoogendorp J., Medvedev N., Hoogland T.M., De Zeeuw C.I., De Schutter E., Yarom Y., Negrello M., Torben-Nielsen B., Uusisaari M.Y. (2019). Variability and directionality of inferior olive neuron dendrites revealed by detailed 3D characterization of an extensive morphological library. Brain Structure and Function 224:1677–1695. doi:10.1007/s00429-019-01859-z. Game built from this paper; the paper is the authority where the two disagree.

how to play

The Olive, explained

The setting

The inferior olive is a small structure in the brainstem, named for its olive-shaped bulge, that helps the cerebellum keep movements precisely timed. Its neurons talk to each other through direct electrical contacts on their branches, called dendrites. So the shape of a cell's branches, and where they go, decides who it talks to.

For about a hundred years anatomists have described olive neurons as two kinds. Curly cells have branches that twist, loop and double back inside a small ball around the cell body. Straight cells have a few branches that run out a long way. A 2019 paper measured ninety of these cells in 3D to ask whether the two kinds are real, and where the cells' wiring groups actually are. This game lets you repeat both questions with your own eyes and hands.

Round one: sort

You see twenty cells, one at a time, drawn as the microscope would show them. For each, tap Curly, Straight, or Can't tell. There is no right answer per cell; the game is watching how consistent your eye is.

After each choice the cell drops onto a ruler of straightness: how far the tree reaches from the cell body divided by the length of its longest branch. A value near 0.1 is a tight ball; near 0.9 is almost a straight line. At the end you see your three piles on that ruler and a consistency score: the share of your curly/straight calls that a single cut on the ruler can reproduce. The question the round answers is whether your "curly" and "straight" piles are two things or two ends of one thing.

Round two: find the groups

Now you see a slab of the olive, about a third of a millimetre wide, with every cell body as a dot. The paper argues the cells form small, tightly wired groups — in a slab this size roughly ten to twelve groups of a dozen or so cells. Tap the slab to drop a marker where you think the middle of each group is; tap a marker to remove it; then Score. The score is the share of cell pairs you grouped the way the olive does.

Then Show the dendrites: every cell's branches light up. Cells talk where their branches touch, so a cell belongs with the cells its branches grow towards, not with the bodies lying next to it. Move your markers to where the branches bunch, and score again. The difference between the two scores is the point of the round.

The score

One number out of 1000, built from your consistency in round one and your two scores in round two. There is no winning; there is a better score next time, and a different cell library (the three labeling buttons on the start screen) if you want new cells.

about this game

The Olive

A beta built from Vrieler et al. (2019), Brain Structure and Function 224:1677–1695, doi:10.1007/s00429-019-01859-z. Game built from this paper; the paper is the authority where the two disagree. Every number the game uses sits in the config block at the top of the page source, each tagged with the table or figure it came from.

What the paper hands over

Primarily an estimate: 90 reconstructed neurons, 25 morphometric measures (Table 1), their ranges per labeling method (Table 4), their correlations with the subjective curly/straight label and with "straightness" (Table 5), a PCA whose first component tracks straightness (Fig. 3), and the positions of 11,800 somata with a cluster search (Fig. 5). Secondarily an argument: the discussion proposes that dendrite directionality delineates tightly coupled groups and that straighter cells bridge them (Figs. 6–7).

The two reversals

One. Believed: two cell types, curly and straight. Paper: a continuum, captured by one number, reach divided by longest path. Two. Believed: coupled groups would show in where the cell bodies sit. Paper: somata form one mesh down to 40 µm; the structure is in the direction the dendrites grow.

Where the player sits

The experimenter. Round one replays the paper's own method: sort, then measure, then see your sorting against the ruler. Round two replays its last section: look for groups in the somata, fail, then see the dendrites. The runner-up concept — the neuron as an agent spending a dendrite budget on curl versus reach — was recorded for a second pass; it needs a payoff the paper's results don't supply.

Presets

Round one draws from one of the paper's three labeling methods, each with its own straightness range (Table 4): Virus (0.13–0.81, median 0.37), Alexa (0.19–0.81, median 0.49), Biocytin (0.21–0.79, median 0.39).

What is synthetic

Every neuron is generated, not traced. Trees are grown branch by branch with a target straightness drawn from the paper's range; branch count, stem count, tip count, total length, hull size and contraction follow straightness with the signs and roughly the strengths of Table 5. A headless check on 600 generated cells per method recovers those correlations within about 0.2 (branches −0.5 vs −0.7 in the paper; stems −0.3 vs −0.4; hull +0.6 vs +0.5; contraction +0.6 vs +0.6) and a unimodal straightness distribution covering the paper's range.

The round-two slab is invented to match Figs. 5–7: twelve dendritic fields tile the slab; each group's cells sit at the rim of their shared field with trees growing into it (soma eccentric, Fig. 4b); three long straight cells span two fields (Fig. 7). The paper reports examples, not group sizes or counts; ours are choices, listed in the config block. The check confirms that a 40 µm distance-based clustering sees one cluster (Fig. 5c) while the dendritic overlap recovers the planted groups.

What is cut

The 25-measure PCA; electrophysiology and subthreshold oscillations; labeling chemistry; the dorsal cap of Kooy; the cluster-bridge hypothesis as something the player can drive (it appears only as the round-two reveal).

Divergences

Neurons are planar, not 3D. Twenty cells per round, not ninety. Round-two trees are drawn at reduced length so fields read as fields. The score is our construction.

For the author

Look at round one first: does the generated library look like your library? Every number sits in the config block at the top of the page source, each tagged with its table or figure. We were least sure about the round-two slab, which formalises your discussion rather than your results.