Avoidance and Approach: Mechanisms of Goal-Directed Learned Behavior

Active avoidance is an intriguing learned behavior

The video shows a rat performing active avoidance.

When the light on the right side turns on, the rat receives an innocuous sensory stimulus in its whisker pad through a pair of subcutaneous wires (the light is just for you to know when the whisker pad stimulus occurs).  The whisker pad stimulus signals that a threat will occur (mild footshock) if the rat does not move to the other side of the cage within 7 sec (if you put your fingertips on the grid, the footshock feels like a tingling sensation).  The rat has already learned that moving to the other side of the cage avoids the threat. In signaled active avoidance, the subject predicts and controls the upcoming threat. Initially, the whisker pad sensory stimulus had no meaning and was ignored. After acquiring new meaning (predicting the threat), the whisker pad sensory stimulus evokes the avoidance response of moving to the other side of the cage. This is fascinating and leads to many questions we are able to tackle using modern experimental tools that allow controlling and monitoring the activity of specific neurons in behaving animals performing signaled active avoidance.  These tools are particularly useful when combined with genetically modified mouse strains, and thus our work now employs mice.

  • What neural circuits mediate active avoidance behavior? Despite being studied for almost a century, the circuits that mediate signaled active avoidance are not well known. Many neural circuits in the brain are modulated (activate or deactivate) by this complex learned behavior, we are defining those circuits that are essential to mediate the behavior. There is much sensorimotor, cognitive, and emotional integration during signaled active avoidance behavior, including sensory stimulus detection (perception), acquisition of new meaning (learning and memory), and its ability to drive a motor response anew (locomotor control). It is an excellent model system to study these issues.
  • What changes in these circuits to code the significance of the sensory stimulus?  Before deciphering what changes in the brain to acquire a new conditioned behavior, it is essential to define the neural circuits that mediate the behavior.  This is best done by selectively manipulating the neural activity of those circuits (gain and loss of function) in behaving mice. Once the neural circuits that mediate the behavior are well defined, it should be more straightforward to decipher what changes in these circuits to acquire (learning) and remember (memory) the behavior. This is also an essential step to understanding how these behaviors go awry during disease.
  •  How do these essential circuits differ when the same motor behavior is either avoidance or approach depending on the motivational state?  It is not well known how neural circuits adapt to produce the apparently same behavior when it is motivated differently.  By studying the same motor behavior motivated differently (avoidance or approach), we are deciphering these intriguing issues.
  • Why deciphering the neural circuits that underlie avoidance and approach is important? During a normal day, we perform active and passive avoidance behaviors constantly (think of the many behaviors that avoid harm). In many cases, these behaviors are signaled (e.g., crossing the street at a light). Failing to correctly perform these behaviors can lead to severe consequences but we tend to perform them with low anxiety due to our ability to predict and control the threats by adjusting our behavior.  In some psychiatric disorders, such as anxiety disorders, subjects may avoid non-threatening situations, which can be highly problematic. In addition, signaled and unsignaled movement is impaired in many neurological disorders. Thus, understanding how these normal behaviors are generated will be useful to decipher new, more effective ways to address these problems.  

Continue reading below to learn what we have discovered about signaled active avoidance.

The sensory cortex is robustly activated by the sensory conditioned stimulus but…

The video above also shows field potential (population) activity recorded from the somatosensory cortex as the animal performs the behavior.  It shows the cortical response evoked by the whisker pad sensory stimulus (innocuous 10 Hz electrical stimulus delivered through a pair of subcutaneous wires; the cortical response evoked by the first 10 stimuli of the 10 Hz train are shown in the video). One intriguing finding is that the cortical activity evoked by the sensory stimulus changed as a function of the behavioral state of the rat (more about this on the behavioral state page). Based on the sensory cortex responsiveness during the signaled active avoidance behavior, one would assume that this part of the brain is critically important for the animal to detect the whisker pad sensory stimulus and respond correctly to avoid the threat.

…signaled active avoidance does not require the sensory thalamus or cortex

A surprising discovery was that lesions of the somatosensory thalamus, which block neural signals to the sensory cortex, do not block the ability of the animals to respond effectively (avoid) to the whisker pad conditioned stimulus (WCS).  The same result occurred in other experiments when animals trained to avoid the footshock in response to an auditory stimulus (ACS, 8k Hz tone) were subjected to lesions of the auditory thalamus.  This indicates that the sensory thalamus and its targets (e.g. sensory cortex) are not required for active avoidance. In other words, the animals can detect the sensory stimulus without sensory cortex involvement. What other neural circuits may detect the sensory conditioned stimulus to drive the avoidance behavior?

When the sensory thalamus is absent, due to a lesion, the superior colliculus in the midbrain fully mediates active avoidance

We discovered that the superior colliculus can detect the sensory stimulus to avoid the threat. The superior colliculus can do this on its own, without the sensory cortex, as long as the sensory stimulus is salient (easy to detect). For less salient stimuli, we discovered that the sensory cortex and the superior colliculus must work together to detect the sensory stimulus and avoid the threat. The synergy between distributed neural circuits is an important feature of sensory detection. In the case of the sensory cortex and superior colliculus, this synergy is likely to occur through corticotectal connections, which drive sensory responses in the superior colliculus via the cortex.

Synergy may occur through corticotectal connections

We found that when a whisker sensory stimulus is delivered, neural activity ascends to the sensory cortex and the superior colliculus causing fast responses in both areas, but the superior colliculus produces a characteristic second response, which we called peak2. The sensory cortex drives peak2 in the superior colliculus through corticotectal fibers and this provides a mechanism for synergy. The image below shows recordings of responses in the superior colliculus driven by a sensory stimulus (whisker deflection). There are 2 characteristic peaks (peak1 and peak2). Peak1 is very fast occurring before the cortical sensory response. Peak2 follows the cortical response and depends on it; it is abolished if the cortical response is blocked. This means that the superior colliculus receives information from the sensory cortex about the stimulus that just occurred. When we lesion the thalamus, the cortex receives no sensory inputs. Therefore, the superior colliculus lacks the information provided by peak2 limiting its ability to detect low saliency stimuli.

There is a neural correlate of active avoidance in the superior colliculus

We recorded cells in the superior colliculus during active avoidance and found that cells ramp up their activity during avoids.

The substantia nigra pars reticulata (SNr) controls active avoidance

The output of the basal ganglia through the SNr consists of tonically active GABAergic neurons that project to the superior colliculus, among other places. Using optogenetics and DREADDs, we discovered that modulating the firing of SNr cells controls signaled active avoidance. Increasing SNr firing blocks signaled active avoidance while decreasing SNr firing drives avoidance (even in the absence of an external sensory signal).  Exciting SNr cells selectively blocked avoidance responses without affecting the ability of the animals to escape the foot-shock; the animals ignore the sensory stimulus that signals the threat. The image below (left) shows GABAergic SNr cells controlled using optogenetics. In SNr cells that express ChR2, cell firing increases as the frequency of the blue light train (1 ms pulses) increases and is maximal when the blue light is continuous (SNrChR2 mice). In SNr cells that express ArchT, cell firing decreases during continuous green light (RosaArch mice). The other image shows (right) the effect of optogenetics on the ability of the mice to perform signaled active avoidance. Blue light delivered in the SNr of SNrChR2 mice during the avoidance interval, simultaneously with the sensory stimulus (ACS+LCS) that signals the threat (auditory tone; ACS), suppresses the ability of the mice to avoid the threat as a function of the blue light frequency; maximal abolishment of avoidance responding occurs with continuous blue light. In contrast, green light delivered in the SNr of RosaArch mice facilitates the ability of the sensory stimulus to drive avoidance responses (ACS+LCS) and, remarkably, is also able to substitute for the sensory stimulus if the sensory stimulus is not presented (LCS alone). Thus, increasing SNr firing during the avoidance interval abolishes signaled active avoidance while decreasing SNr firing drives active avoidance, even in the absence of an external sensory signal.

SNr firing does not suppress signaled active avoidance by inhibiting the superior colliculus but by inhibiting the midbrain tegmentum

After finding that GABAergic SNr cells could control signaled active avoidance, we determined which output pathways of the SNr controlled active avoidance. We discovered that SNr pathways to the thalamus and superior colliculus were not responsible for controlling signaled active avoidance. Instead, projections from the SNr to the midbrain pedunculopontine tegmentum (PPT) control active avoidance. Importantly, inhibiting excitatory cells in the PPT is sufficient to block signaled active avoidance!

Also, excitation of GABAergic afferents in PPT from local neurons, SNr neurons, or zona incerta neurons is sufficient to abolish signaled active avoidance. Moreover, excitation of excitatory neurons in PPT produces fleeing responses that can be tuned, in a frequency-dependent manner, to resemble fast escape responses (driven by the foot-shock) or slow avoidance responses (driven by the sensory stimulus that predicts the threat). The image below shows the effect of inhibiting CaMKII-Arch expressing neurons in PPT with green light. Note the abolishment of signaled active avoidance at the higher green light powers without affecting the ability of the animals to escape the footshock (which ensues at 7 sec).

Neurons in the pedunculopontine tegmentum have a full representation of signaled active avoidance

Using calcium imaging fiber photometry we discovered that the neurons that block signaled active avoidance when they are inhibited, discharge at the onset of the auditory conditioned stimulus, and discharge further during the occurrence of the avoidance response. The discharge of these neurons follows closely the avoidance behavior of the animals in different versions of active and passive avoidance procedures.

So, what drives these midbrain pedunculopontine tegmentum neurons to discharge and generate avoidance responses? Could it be SNr cells?

SNr firing codes but does not drive avoidance responses

One possibility is that SNr cells deactivate to drive avoidance responses by disinhibiting cells in the pedunculopontine tegmentum. Although the firing of SNr cells is modulated by avoidance responses (both activation and deactivation occur), we discovered that SNr firing does not drive avoidance responses.  If we interfere with SNr activity during avoidance responses, these responses occur normally. The modulation of basal ganglia activity output via the SNr must serve another purpose other than driving avoidance responses. One possibility is that this activity serves to inform target structures about the ongoing response.

If SNr deactivation does not drive avoidance responses, then what neurons drive the pedunculopontine tegmentum to generate avoidance responses?  Stay tuned, we are exploring some intriguing possibilities.