A neurodevelopmental origin of behavioral individuality in the Drosophila visual system

What was read

• bioRxiv preprint 10.1101/540880 v1 (5 Feb 2019), 37 pages: main text, Figs 1–5 legends, Materials and Methods, Figs S1–S6, Movies S1–S7 captions. Read in full.
• NOT read: the published Science version (paywalled; Unpaywall/OpenAlex/FreiDok list no open copy, and FreiDok's record has no file). The published paper may contain revisions beyond the preprint; numbers below are the preprint's.

Question

Does non-heritable, stochastic variation in brain wiring during development cause stable individual differences in behaviour? The paper sets this "developmental noise" alongside genes and experience as a third source of individuality.

System

• Dorsal Cluster Neurons (DCNs): a population of commissural visual interneurons in the fly optic lobe, each hemisphere's set derived from one neural stem cell. Each DCN sends an axon to the contralateral side ending either in the lobula (L-DCN, ~75%) or the medulla (M-DCN, ~25%). The L/M choice is made by an intrinsically stochastic lateral-inhibition mechanism (Langen et al. 2013, Zschatzsch et al. 2014), so wiring differs between individuals and between the two hemispheres of one brain.
• Behaviour: Buridan's paradigm. A single fly walks on a water-moated platform between two unreachable black stripes at 180°. "Absolute stripe deviation" = how far the path deviates from the straight line joining the stripes; low = narrow, direct path, high = broad or meandering. Tracks analysed with CeTrAn plus custom Python; 36 behavioural parameters.
• Anatomy: brains of individually tracked flies dissected the day after testing, imaged by confocal, reconstructed semi-automatically in Imaris, and measured blind for DCN counts and medulla axon numbers; 37 anatomical features.

Results (with numbers)

1. Inter-individual variation in stripe deviation is large, similar in males and females (N = 50 and 48), and normally distributed.
2. Reduced genetic diversity does not reduce it. Two inbred DGRP lines (639, N = 61; 859, N = 59), inbred for 20 generations, differ from outbred Canton S in population mean but are as variable or more variable across individuals (Fig 1 F–I, S1).
3. It is non-heritable. Mating the three lowest-deviation pairs and the three highest-deviation pairs from 47 males and 37 virgin females gave offspring populations (N = 180 each) that were indistinguishable in stripe deviation (p = 0.22) and six other parameters, and that reproduced the parental population's full variability (Fig 2, S2). Single-pair version: same (N = 60 each, p = 0.4).
4. It is temporally stable. Same individuals tested every other day for three days: Pearson r = 0.74–0.77 between days for Canton S (N = 74), with left- or right-shifted approach angles also conserved; stable in four-week-old flies (r = 0.53–0.70) and in both DGRP lines (Fig 3, S3). So the trait is intrinsic, not state-dependent.
5. DCN wiring is more variable than previously thought (N = 103 brains, up to 168 hemispheres): 22–68 DCNs per hemisphere; 11–55 lobula axons; 6–23 medulla axons (mean 13.99). Left/right medulla-axon asymmetry ranges 0–10 (mean about 2.4–3.0 depending on measure). Medulla axon number does not correlate with DCN cell number (r = −0.03), so the medulla innervation is an independent stochastic outcome. Adult M-DCN pattern is stable over hours and days (Movie S7).
6. DCNs are required for the behaviour: silencing all DCNs or only M-DCNs with Kir2.1 or tetanus toxin abolishes stripe fixation (p < 0.001; flies approach stripes from the arena edge) (Fig 4 C–D, S4 B–C).
7. Structure predicts the individual: in an unbiased 36 × 37 correlation screen, left/right asymmetry in M-DCN medulla innervation is the strongest predictor of an individual's stripe deviation, r = −0.67, p < 0.001, N = 103 (more asymmetric wiring → narrower path). Related parameters (absolute angle deviation, centre deviation) also correlate; unrelated ones (distance, r = −0.17; number of full walks, r = 0.24) do not. Optic-lobe size asymmetry (r = 0.03) and asymmetry of a neighbouring atonal-positive population (r = 0.01) do not predict behaviour; lobula-axon asymmetry does only weakly (r = 0.32) (Fig 5, S5).
8. The correlation depends on DCN activity: with tetanus toxin in DCNs the anatomy–behaviour correlation vanishes (r = −0.002, p = 0.98, N = 92; control r = 0.54, N = 89).
9. Asymmetry is sufficient: painting over one eye in the 20 weakest-orienting flies (stripe deviation > 40, expected to be DCN-symmetric) improved their orientation (paired Wilcoxon p = 0.002); across all 79 flies the effect was milder (p = 0.01).

Authors' conclusion: stochastic developmental wiring produces a distribution of circuit diagrams across genetically identical individuals, and that wiring difference causes a stable behavioural individuality.

Limits

• One innate visual behaviour, one identified neuron population of a few dozen cells, in a fly. The structural readout is a count of axon branches in a genetically labelled population, not a connectome.
• The correlation explains roughly 45% of the variance (r² ≈ 0.45); the rest of the individual's behaviour is not accounted for by this feature.
• The candidate circuit was chosen by prior hypothesis (DCNs were already known to be variable and matched a predicted contralateral circuit); the "unbiased" screen was over 36 × 37 pairs without a stated multiple-comparison correction in the preprint.
• Sign inconsistencies in the preprint legends (Fig 5C gives r = 0.67 where the text says −0.67; Fig S5A calls r = 0.72 "negatively correlated"). Presumably fixed in the published version; not verified.
• The structure was read after death, from a fixed brain, but the behaviour had been measured in life on the same individual; this is the design that would be needed to calibrate any structure-to-trait decoder.

What the brief uses it for, and whether it holds

Brief section 2: "The one existence proof of structure predicting an individual trait is a single fly circuit [2]; at MRI resolution, human brain–trait associations are weak enough that reproducible studies need thousands of subjects [3]."

• Holds. The paper is exactly an existence proof: a post-mortem structural measurement of one circuit predicts a stable individual behavioural trait (r ≈ −0.67), with causal support from silencing and from induced asymmetry.
• The brief's framing is fair about scale: the "trait" is the width of a walking path, the "structure" is a branch count in a few dozen neurons, and it is a fly.
• Relevant to Variant A (section 8) in two ways the brief does not spell out. First, this individuality is non-heritable and invisible to the genome, so structure is the only place it lives; records or relatives could not supply it. Second, reading it required paired data: living behaviour and post-mortem anatomy from the same individuals (N = 103) to fit the map. That is the "brain-bank effect-size study" the brief says Variant A would need before it is revisited.
• Also relevant to 6.4 (observability): the same measurement on a different population (VCNs, optic-lobe size) carried nothing. Structure predicts a trait only where the right structure is measured.

Cross-references

• [3] Marek 2022 is the human-scale counterweight: at MRI resolution the equivalent brain–behaviour correlations are of order r ≈ 0.1–0.2 and need thousands of subjects.
• Sits beside [1] Abdou 2018: [1] says content is in synaptic plasticity, not cell membership; [2] says a coarse wiring feature can still predict a trait. Both are consistent with the brief's "constraints on a person model, not a person" (section 3).