Summary of results
Behavioral responses and TO2 signals from the BLA, dHPC and vHPC were recorded simultaneously in freely-moving rats during the acquisition, expression and extinction of conditioned fear. Rats froze more during CS+ than CS− trials during the extinction session in a novel context but not during training. In the BLA, TO2 signals were higher during CS+ than CS− trials during training and at the start of extinction, but were significantly lower during CS+ than CS− trials by the end of extinction, a pattern that may reflect the expression of the ‘CS+ → no US’ association learned during extinction. Moreover, during training, the levels of CS+/CS− discrimination present in the BLA TO2 signals were strongly correlated with behavioral discrimination, even though at a group level the rats did not freeze significantly more during CS+ than CS− trials. Also, during extinction, the decrease in the BLA TO2 signal from CS+1 to CS+5 was significantly correlated with the decrease in freezing responses from CS+1 to CS+5. TO2 signals in dHPC and vHPC also discriminated between the CS+ and CS−, with higher CS+-evoked signals in vHPC during training and lower CS+-evoked signals in vHPC and dHPC during extinction. Thus, in rats, hemodynamic responses in the BLA and HPC can detect the distinct patterns of neuronal activity evoked by aversive vs. neutral cues.
Role of the amygdala in fear conditioning: from rodent studies to human neuroimaging
Rodent lesion and electrophysiological studies demonstrate an essential role for the BLA in fear conditioning and extinction (LeDoux et al., 1990; Maren et al., 1996; Herry et al., 2008; Sierra-Mercado et al., 2011), and several authors have proposed the BLA as the critical locus for CS+ → US associations (Davis, 1992; Romanski et al., 1993; Fanselow & LeDoux, 1999; Maren, 2001). Overall, in rodents the evidence for the role of the BLA in fear conditioning is compelling.
Given this evidence, the failure of many human fMRI fear studies to detect differential amygdala activation is surprising (Mechias et al., 2010). However, the interpretation that group-averaged hemodynamic signals are insensitive to the differential patterns of neural activity elicited by the CS+ vs. CS− (for discussion, see Bach et al., 2011) is not consistent with the present data. Even with a relatively small sample, we found that 10 training trials were sufficient to produce robust CS+/CS− discrimination in the BLA TO2 signal. So, at least in rats, amygdala hemodynamic responses can discriminate between aversive and neutral stimuli.
Nevertheless, the question remains as to why some fMRI studies do not observe differential amygdala activity. First, it is important to emphasize that many studies have reported robust differential amygdala activity (Morris et al., 1998, 2001; Armony & Dolan, 2002; Tabbert et al., 2005, 2006; Knight et al., 2009). Thus, our data are consistent with a subset of the fMRI literature. Importantly, our study shares certain design features with some of the studies cited above (Tabbert et al., 2005, 2006; Knight et al., 2009) that are not found in the majority of the studies discussed by Mechias et al. (2010) and Bach et al. (2011). First, unambiguously aversive USs were used (i.e. shock or 100 dB white noise). Second, 100% reinforcement between the CS+ and US was used (i.e. the CS+ was always followed by the US during training), not partial reinforcement. This may be particularly important as amygdala activation increases as a function of the probability of reinforcement (Dunsmoor et al., 2007), and the majority of human fear-conditioning studies have used partial reinforcement to avoid conflation between CS+- and US-evoked responses. These design features may be necessary, although perhaps not sufficient, to evoke differential amygdala responses to the CS+ vs. the CS−.
A further consideration concerns the levels of fear evoked in participants. Humans are aware that they will not come to any real harm during the experiment, which presumably limits their subjective experience of fear during scanning. Although we cannot know for sure, it is a reasonable assumption that rodents experience higher levels of fear than humans during fear-conditioning experiments. Equating levels of fear across species may not be possible, but a recent human fMRI study offers insights into the relationship between fear and amygdala activation. van Well et al. (2012) found that only subjects who exhibited differential fear-potentiated startle (FPS) responses to the CS+ vs. the CS− had higher CS+- than CS−-evoked amygdala BOLD signals. Subjects that did not exhibit differential FPS did not show differential amygdala activity, even when they showed normal US expectancy (i.e. they could correctly predict which CS would be followed by shock). Thus, conditioning, as indexed through correctly learning the stimulus contingencies, is not sufficient to elicit differential amygdala BOLD signals.
While van Well et al. (2012) were the first to measure FPS during human fMRI, many previous studies have used the skin conductance response (SCR), an autonomic measure of arousal, to index fear conditioning (LaBar et al., 1998; Phelps et al., 2004; Knight et al., 2005; Carter et al., 2006; Cheng et al., 2006, 2007). A robust finding from these studies is the positive correlation between differential CS+/CS− SCRs and differential amygdala BOLD signals. This is consistent with our study, which found a strong correlation between differential freezing responses and differential BLA TO2 signals. If subjects (rodent or human) do not show differential fear responses (as indexed by FPS, freezing or SCRs), it is unlikely that they will show differential amygdala activation. Thus, the levels of fear evoked by the CS+ compared with the CS− may be the critical determinant of differential amygdala activation.
HPC TO2 signals discriminate between the CS+ and CS−
The precise role of the rodent HPC in fear conditioning remains debated. Early studies suggested that the HPC was required for contextual but not discrete cue conditioning (Selden et al., 1991; Phillips & LeDoux, 1992). However, both dHPC and vHPC lesions can reduce freezing to discrete cues under some circumstances (Richmond et al., 1999; Maren & Holt, 2004; Maren, 2008; Quinn et al., 2008; Zelikowsky et al., 2012).
The present data show that both dHPC and vHPC exhibit differential TO2 responses to discrete auditory cues. HPC TO2 signals discriminated between CS+ and CS− trials, both during training (vHPC) and during extinction (vHPC and dHPC). Moreover, HPC TO2 signals were markedly different from BLA TO2 signals, which is important because it shows that TO2 signals do not simply reflect general changes due to freezing behavior or the systemic physiological changes that accompany fear conditioning (e.g. heart rate, blood pressure, blood flow, etc.). Notably, there is good concordance between our vHPC TO2 data and anterior HPC BOLD signals recorded during human fear conditioning in terms of the timing and shape of the response (Alvarez et al., 2008; note that anterior HPC is the primate equivalent of rodent vHPC). In particular, in both humans and rodents the difference between CS+- and CS−-evoked signals is transient (i.e. the CS+ response is greater than the CS− response, but only during the first few seconds of stimulus presentation). Thus, differences may not be detected in the HPC when the BOLD contrast is based on analysis of longer durations.
Another notable finding is that CS+-evoked signals were significantly lower than CS−-evoked signals in both HPC subregions during extinction (i.e. there was a CS− > CS+ contrast). We do not know why the CS+-evoked HPC TO2 response is negative (compared with both the CS− and the pre-CS baseline), but recent evidence suggests that negative BOLD signals are associated with suppression of neuronal activity (Boorman et al., 2010). Exactly why the CS+, and not the CS−, is associated with this negative HPC response is not immediately obvious and requires further investigation. Nevertheless, higher CS−- than CS+-evoked HPC BOLD signals have also been reported during extinction in humans (Phelps et al., 2004). Thus, our HPC findings during extinction are consistent with human fMRI data.
Using TO2 amperometry as a translational tool for BOLD fMRI
The present study is the first to employ time-resolved hemodynamic measurements in freely-moving rats, but previous studies have measured hemodynamic activity in rats after fear conditioning (LeDoux et al., 1983; Holschneider et al., 2006). For example, using cerebral blood flow (CBF)-autoradiography, Holschneider and colleagues found significantly increased CBF in the lateral amygdala in a fear-conditioned compared with a tone-alone group, but they also found significantly decreased CBF in the anterior BLA and central amygdala (Holschneider et al., 2006). While autoradiographic techniques can image the whole brain, their disadvantage is that they have no temporal resolution and offer only a snapshot of CBF changes in the brief (<1 min) period before the animal is killed. As a result, autoradiographic approaches are incompatible with the types of discriminative fear-conditioning designs used in human fMRI and the present study. In contrast, the TO2 approach used in the present study offers a high temporal resolution signal for studying hemodynamic changes associated with behaviour.
There are strong theoretical and empirical reasons for arguing that TO2 is closely related to BOLD (Zheng et al., 2002; Logothetis, 2007; Lowry et al., 2010). Nevertheless, there are several methodological differences between our approach and human fMRI studies of fear conditioning, including the species and the higher temporal and spatial resolution (but limited spatial sampling) of TO2 compared with BOLD. Importantly, however, our data are consistent with a subset of human fMRI studies that share certain key design features with our study, suggesting that these methodological differences are not critical.
Moreover, the TO2 approach utilized here is applicable to many other areas of neuroscience. Most of our knowledge of the cellular and molecular mechanisms that underlie behavior comes from animal experiments that for ethical and practical reasons are impossible in humans. Thus, animal models offer the best hope for understanding neuronal mechanisms underlying the dysfunctional brain states associated with neurological and psychiatric disorders. Techniques that facilitate comparisons between animal and human research are therefore essential if basic research is to translate to clinical benefits. One of the key advantages of our TO2 approach is that it can be combined with invasive methods, such as genetic modification, brain lesions, local drug infusions and electrophysiology; in other words, the techniques that allow us to uncover neuronal mechanisms. Thus, tissue oxygen amperometry has the potential to improve the translation between rodent models and human neuroimaging.