In three fly nerve cord connectomes, EM transmitter predictions broke the hemilineage rule in the same early-born neurons, and secondary neurons kept the rule

Question: In the fly ventral nerve cord, all neurons of one hemilineage are thought to use one fast transmitter (Lacin et al. 2019). Do electron microscopy (EM) transmitter predictions agree with this rule when the models did not get it as a training label? Do the exceptions repeat between animals? Data: neuron-level predictions from three separately trained models in MANC v1.0 (male), male CNS v1.0 (male) and BANC v888 (female). We used 12,963 MANC neurons from 31 hemilineages with a known transmitter, matched one-to-one to male CNS (mancBodyid) and BANC (manc_match) neurons. Most of these neurons had a training label that follows the rule in the male CNS or BANC. Results: Labeled neurons agreed with the rule in 96.5% (MANC), 98.1% and 99.7% of cases. Without a label, secondary neurons agreed in 91-97%, but primary and early secondary neurons agreed in only 42-50% in all three animals. The breaks repeated. Of unlabeled neurons with a male CNS break, 93% also broke the rule in BANC, against 0.2% of those that kept it; 99% of double breaks had the same transmitter. Of 9,216 neurons with predictions in all three animals, 272 had the same non-hemilineage transmitter in all three. Independent errors give 1.1, and a permutation within hemilineage and birth time gives 62 (maximum 81 of 1,000). Of the 272, 93% were primary or early secondary and 96% had no training label. Breaks went in all directions (glutamate to acetylcholine 120, GABA to acetylcholine 77, acetylcholine to glutamate 37, others 38), mainly in 09B, 19A, 21A and 07B; 76 types broke the rule in all three animals. A sign table from the hemilineage rule gave the opposite of the three-animal majority sign at 5.1% of presynaptic sites (09B 56%, 21A 30%, 19A 25%, 07B 15%). Each dataset's own predictions differed from the majority at 0.07-1.3%. Motor neurons are glutamatergic, but BANC predicted GABA for 62% of 391 matched motor neurons, and the male CNS gave 'unclear' for 86%. Thus the rule describes secondary neurons well, and many early-born neurons have the same other transmitter in three animals. Limits: all calls are EM predictions without new experiments; the models share one method family and partly share label sources; hemilineage and birth time come from annotations; neuron matches can contain errors; there is one female nerve cord.

Drosophila · ventral nerve cord · connectome · neurotransmitter prediction · hemilineage · birth order · MANC · male CNS · BANC · Dale's law · signed connectome models · motor neurons · comparative connectomics