Variability in automatically generated raw activity scores.

The limitations in the use of raw activity scores in the analysis of conditioned fear are clearly stated by Anagnostaras et al. (2010): raw activity scores such as those provided by the Video-freeze system do exhibit more between-subject variability than freezing scores. Figure 1 shows distributions of automatically generated freezing and activity scores for 44 Swiss Webster mice, which have been transformed into z scores in order to allow comparisons in a common metric1: Although freezing during baseline was absent for most subjects, raw activity scores do indeed show considerable variance.

Figure 1

Raw activity and freezing scores under baseline and conditioned stimulus conditions (. Raw activity end percent freezing scores (freezing threshold = 18), were automatically generated by the Video-freeze© system. Data were then transformed into z scores in older to allow for comparisons between the two scales.

This increased variance is not limited to baseline data, but extends to conditioned responding, as can be seen in the right-hand side of Figure 1, which compares freezing and raw activity scores for the presentation of a conditioned stimulus (blinking house light), after two light-shock pairings (0.7 mA, 2 s. scrambled foot-shock). This increased variability of raw activity scores in and of itself does not constitute a limitation of this type of data. On the contrary, it may offer a unique opportunity to model individual differences through fear-conditioning preparations: one of the distinctive features of anxiety disorders such as post-traumatic stress syndrome is precisely the variability in the response to stressors. Such between-subject variation has been successfully replicated in animal models (Bush et al., 2007; Walker et al., 2008).

Raw activity scores may constitute better predictors of response to stressors precisely because of their greater dynamic range, especially when comparing several strains of mice. Incorporating baseline differences into preclinical modeling will indeed require slight departures from current practices in the data analysis of fear-conditioning data (see for example Robles and Vargas-Irwin, 2010), but may by achieved with procedures as straight forward as using raw baseline data as covariates. In summary, analyzing raw scores presents new computational challenges, and further modeling and analytical developments are needed to take advantage of these opportunities.