Take your time to decide: to swim or not to swim

Animals use sensory evidence to decide what to do and when to do it. In fact, when presented with certain stimuli, it may take up to several seconds for an animal to react, even though action potentials, which govern the flow of information in the brain, last only a few milliseconds. Why?

HFSP Cross-Disciplinary Fellow Ruben Portugues and colleagues
authored on Thu, 18 June 2015

One possible explanation is that the brain is “accumulating evidence before deciding what to do,” i.e. given the visual motion, should it send a command to move left or right?  These mechanisms imply a three-step process.  Initially, sensory evidence is transformed into a sensory drive describing the instantaneous strength of the stimulus.  Secondly, this drive is integrated or accumulated over time (hundreds of milliseconds or seconds) to represent the evidence that lends support to a particular motor output.  Finally, when this evidence reaches a threshold, the decision is made to perform the behavior.  In fact, a large volume of work exists that investigates correlates of these various steps in the context of perceptual decision making in primates.

In this behavioral study we ask whether there may be a different mechanism underlying the initiation of motion in larval zebrafish, in particular a mechanism that does not integrate evidence over time but may nevertheless result in latencies in the order of seconds.  Larval zebrafish, like many other animals, try to keep in the same place with respect to their visual environment: if this moves left or right, they turn and swim left or right, and if it moves forward, they swim forward with it.  Nevertheless, if it moves forward slowly, it may take them up to 4 or 5 seconds before they start swimming.  As described above, it may be that they are accumulating evidence over this period.  In contrast, we propose that the long reaction times may arise from a non-history dependent process.  The sensory drive (the forward speed) may be used to set up the rate of a stochastic Poisson process.  This rate can be set up instantaneously (as opposed to the drive being integrated over time): slow speeds result in a low rate and high speeds in a high rate.  No integration over time is necessary.

We assume that the sensory drive is a function solely of the speed of the grating and then test our model by showing larval zebrafish gratings with speeds that vary in time and are either constantly accelerating or decelerating.  The reaction times we measure for these experiments, together with the latency distributions to the constant speed gratings support the idea that this stochastic model describes the behavior better than a history-dependent accumulation of evidence model. 

Animals often perform behaviors that appear to involve non-local cues in either space or time.  Flies fly towards the rotting fruit across the room or in our example fish take 4 seconds before “deciding” to start swimming.  In fact, these behaviors can be explained with local rules that are implemented here and now: flies follow local odor gradients and fish may set up the instantaneous probability of initiating a swim.  Setting up these systems with neurons is in fact easier.


Whole-field visual motion drives swimming in larval zebrafish via a stochastic process, Ruben Portugues, Martin Haesemeyer, Mirella L. Blum and Florian Engert. (2015) J Exp Biol 218: 1433-1443.

Link to article

Pubmed link