Paper: Vrieler et al. (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. Only the PDF was available; no code or reconstructions.
Primary: estimate. 90 reconstructed neurons, 25 morphometric measures (Table 1), their ranges per dataset (Table 4), their correlations with the subjective curly/straight label and with “straightness” (Table 5), a PCA whose first component tracks straightness (Fig. 3), soma positions of 11,800 cells with a DBSCAN cluster search (Fig. 5).
Secondary: argument. The discussion proposes that dendrite directionality delineates tightly coupled clusters and that straighter, less directional cells act as “bridges” between them (Figs. 6–7). No model, one illustrative sample. Formalising this needs choices; see section 8.
Two.
A. The experimenter (sort, then cluster). Position: the anatomist with a confocal stack. Lever: label neurons; draw cluster boundaries. Feedback: histogram fills; boundaries scored. Surprise: your categories don’t exist; your clusters were in the wrong place.
B. The neuron (grow your dendrites). Position: one IO neuron, cast as agent. Lever: a dendrite budget split between local curl and straight reach, plus a direction. Feedback: gap-junction coupling with neighbours lights up as trees overlap. Surprise: the curly build couples you tightly to a few; the straight build couples you weakly to many, and the network needs both. Payoff (coupling → synchronised firing) is in the introduction, not the results: our construction.
C. The forecaster (bet on the slice). Position: a reader before the figures. Lever: bets on distributions (two humps or one?), on cluster detection, on where somata sit in the dendritic volume. Feedback: the paper’s figure. Surprise: each bet reveals a reversal. Thin.
A. It is the paper’s own method replayed, both reversals land on the player through their own hands, and it needs no invented payoff. Runner-up: B, which is the better toy and the better second pass once the author has confirmed that the cluster-bridge story is what they want foregrounded. Switch to B if the author says the game is “too much about the method, not enough about the olive.”
Act one, by labeling dataset (Table 4 columns): Virus (fullest trees, straightness median 0.37), Alexa (sparser, median 0.49), Biocytin (median 0.39). Act two, by subnucleus (Fig. 1g, Fig. 5): PO, MAO, DAO — same mechanism, different soma fields.
The 25-measure PCA (reduced to straightness plus the three or four correlates the player can see); electrophysiology; the cluster-bridge hypothesis as a mechanism (it appears only as the act-two reveal, not as something the player can drive); subthreshold oscillations; labeling chemistry; the DCK subnucleus.
The confocal stack. Black field; neurons in fluorescent green (viral), red-orange (Alexa) or amber (biocytin) with a soft glow; the olive border as a white dotted line; a 20 µm scale bar; monospace captions in the corner like a microscope overlay. Motion: trees grow from the soma when they appear; in act two the dendrites “develop” like a long exposure.
Result (600 generated cells per method, seed 20191677): straightness unimodal and spanning Table 4 (virus 0.15–0.83, median 0.41); Spearman rho with branches −0.53 (paper −0.70), stems −0.37 (−0.41), tips −0.54 (−0.63), hull +0.68 (+0.50), contraction +0.57 (+0.57), total length −0.34 (−0.47). Signs all right; branches and length weaker than the paper by ~0.15. Act two: DBSCAN at 40 µm sees one cluster in 18 of 20 slabs; seeds at the true field centres score F1 > 0.85 in 20 of 20.
Planned: Generate 1,000 synthetic neurons per dataset; re-measure straightness, branches, stems, tips, hull; check that Spearman correlations match Table 5 in sign and within ±0.15; check that the straightness distribution covers Table 4’s min–max and is unimodal; check that the act-two soma field, run through a simple DBSCAN at D = 40 µm, yields one cluster (Fig. 5c) while the dendrite-overlap graph yields the planted groups.