A tardigrade builds its synapses last: two developmental programmes in Hypsibius exemplaris, separated by their timing and then by their chemistry

Abstract

Two public developmental transcriptomes of Hypsibius exemplaris, Levin et al. 2016's 62 single timed embryos and Yoshida et al. 2017's egg-to-adult series, have never been asked when a tardigrade transcribes its synaptic machinery: neither paper, nor two later reanalyses, mentions synapsin, synaptotagmin, VAChT, GABA or tyrosine hydroxylase. Eleven synaptic and neurotransmitter genes rise through embryogenesis in Levin's series (mean Spearman rho +0.414 against a background of −0.104; permutation p = 0.0001–0.002 across twelve parameter settings). They replicate in Yoshida's independent series (rho +0.764, all eleven rising), then fall after hatching to 0.18–0.61 of their egg-stage level. Neural patterning factors fail the same monotone test (p = 0.38–0.56), but not because they are unordered. They form a synchronised pulse: mean pairwise correlation +0.521 (p = 0.0028), all peaking within six hours, against +0.305 and twenty-six hours for the synaptic set. Patterning comes first and wiring later, and we report the failed test alongside the one that rescued it. The genes co-expressed with each set share no locus and no Pfam domain. The synaptic neighbourhood holds unannotated complexin, Shaker, synaptophysin-family and EF-hand proteins. The patterning neighbourhood is enriched for immunoglobulin-superfamily adhesion molecules (4 against 1 in 300 random proteins, P = 1.5e-6), homeodomain factors and cadherins, and includes a complete Celsr/Flamingo. Among them is BV898_05552, sister to unc-86 at 93% bootstrap: a POU4 neuronal determinant not previously reported in a tardigrade, which switches on 288-fold at about 38 hours post laying, inside the patterning pulse. Two methodological results reach beyond this question. Levin's assembly maps to only 45.5% of the 19,946 annotated nHd_3.1 loci, against 99.8% for Yoshida's, so the most finely timed dataset cannot see half the genes. And our own first time axis for Levin's series was wrong by a factor of two. The supplement's minutes are halved from the two-cell stage, and the series runs to about 101 hours. Only on the corrected axis do the programmes agree with morphological staging. Every statistic is rank-based, so no result changed; every stated hour did.