Embracing Complexity in Defensive Networks

Abstract

The neural basis of defensive behaviors continues to attract much interest, not only because they are important for survival but also because their dysregulation may be at the origin of anxiety disorders. Recently, a dominant approach in the field has been the optogenetic manipulation of specific circuits or cell types within these circuits to dissect their role in different defensive behaviors. While the usefulness of optogenetics is unquestionable, we argue that this method, as currently applied, fosters an atomistic conceptualization of defensive behaviors, which hinders progress in understanding the integrated responses of nervous systems to threats. Instead, we advocate for a holistic approach to the problem, including observational study of natural behaviors and their neuronal correlates at multiple sites, coupled to the use of optogenetics, not to globally turn on or off neurons of interest, but to manipulate specific activity patterns hypothesized to regulate defensive behaviors.

Keywords: amygdala; defensive behaviors; extinction; fear; infralimbic; medial prefrontal cortex; prelimbic.

Figure 1:

Population coding provides a comprehensive view of neuronal representations. A. The same population code can be seen from several vantage points. Active neurons are colored, while inactive ones are gray. The Ensemble View is the specific population of neurons that are activated by an event. A Population Vector View lists the activities of all sampled neurons in a particular region, irrespective of whether they are activated or not. This can be thought of as a point in a space with a dimensionality equal to the number of sampled neurons. Such spaces are difficult to visualize, so investigators often use dimensionality reduction techniques (e.g. principal component analysis) to project the high dimensional Population Vector View into a Dimensional Reduction View. B. Different task events often activate different ensembles, and their relationships can be evident from each vantage point. However, the Dimensional Reduction View (bottom) is often most helpful for visualizing their relationships. C. Two different types of events (Type 1 and Type 2) can activate overlapping ensembles, but downstream neurons are still capable of distinguishing between the events so long as the overlap is not complete. D. The relationships between multiple types of events can be visualized using a correlation matrix, by taking the mean response for each event type and measuring its correlation with the mean of another event type. Such a plot will summarize the relationships between event-evoked ensembles and sometimes reveal how events are ‘categorized’ within a region.

Figure 2:

Multiple population codes coexist within the BLA. All arrows indicate the location of the schematized relationship between ensembles in the Dimensional Reduction View. A. Presently, valence is the most studied code in the BLA. Populations encoding negatively or positively valenced events occupy different parts of space. Events of the same valence can be broken out in the same space, and will often occupy similar regions. B. Learning can shift population codes, such as making an ensemble that encodes a cue more similar to the ensemble encoding the reinforcer. Further training, such as extinction, can further alter the code. C. Besides valence, ensembles in the BLA simultaneously encode whether behaviors are active or passive, and the animal’s movement speed. Since these occupy different locations in the population space, they can be independently decoded. D. LA and BL appear to differ in how they encode events of the same valence. In LA events of the same valence tend to activate similar ensembles, while this is less so in BL.