Failures that are not failures: unobserved sensory history predicts escape behavior

Abstract

Animals often fail to respond to reliable sensory stimuli, a phenomenon commonly attributed to stochastic neural noise or modulation by internal states. In locusts, looming-evoked jump escape response (JER) fails in ~40% of trials, and these failures correlate with multiplexed features of the descending contralateral movement detector's (DCMD) response1,2. Using Hieroglyphus banian, we show that these failures are predictable from an unobserved "Light-to-Dark transition" (LDT) caused by the procedure in the assay, minutes before the looming stimulus. In the JER assay, LDT induced a long-lasting competing behavioral state, grooming, which suppressed JER. Reducing the amplitude of this transition by illuminating the tunnel decreased grooming and rescued JER. Grooming was reliably induced by LDT in the grasshopper under culture conditions. Electro-mechanical induction of grooming was sufficient to cause JER failure. The grooming state induced by either LDT or electro-mechanical stimulation reduced looming-evoked DCMD responses through a divisive gain modulation while preserving normalized variability. Grooming was rare under natural field conditions. Thus, accounting for previously unobserved stimulus history can explain a substantial component of apparent behavioral failures that would otherwise be attributed to stochastic neural noise.