Hemodynamic responses in amygdala and hippocampus distinguish between aversive and neutral cues during Pavlovian fear conditioning in behaving rats

Lesion and electrophysiological studies in rodents have identified the amygdala and hippocampus (HPC) as key structures for Pavlovian fear conditioning, but human functional neuroimaging studies have not consistently found activation of these structures. This could be because hemodynamic responses cannot detect the sparse neuronal activity proposed to underlie conditioned fear. Alternatively, differences in experimental design or fear levels could account for the discrepant findings between rodents and humans. To help distinguish between these alternatives, we used tissue oxygen amperometry to record hemodynamic responses from the basolateral amygdala (BLA), dorsal HPC (dHPC) and ventral HPC (vHPC) in freely-moving rats during the acquisition and extinction of conditioned fear. To enable specific comparison with human studies we used a discriminative paradigm, with one auditory cue [conditioned stimulus (CS)+] that was always followed by footshock, and another auditory cue (CS−) that was never followed by footshock. BLA tissue oxygen signals were significantly higher during CS+ than CS− trials during training and early extinction. In contrast, they were lower during CS+ than CS− trials by the end of extinction. dHPC and vHPC tissue oxygen signals were significantly lower during CS+ than CS− trials throughout extinction. Thus, hemodynamic signals in the amygdala and HPC can detect the different patterns of neuronal activity evoked by threatening vs. neutral stimuli during fear conditioning. Discrepant neuroimaging findings may be due to differences in experimental design and/or fear levels evoked in participants. Our methodology offers a way to improve translation between rodent models and human neuroimaging.


Introduction
Understanding the neural basis of aversive learning is a fundamental goal in neuroscience, and Pavlovian fear conditioning is the dominant experimental paradigm (Davis, 1992;Phelps & LeDoux, 2005). Typically, an initially neutral cue such as a tone is paired with a painful, unconditioned stimulus (US) such as electric shock. With sufficient training, the cue becomes a conditioned stimulus (CS+) with its own affective significance. A second cue (CSÀ), never paired with the US, can be used to assess the specificity of the 'CS + ? US' association. Subsequent presentations of the CS+ without the US leads to extinction, in which a new 'CS+ ? no US' association is formed. This simple paradigm can be used in rodents and humans in essentially the same form, and it is assumed that it engages the same brain structures.
However, several human functional magnetic resonance imaging (fMRI) studies report only transient amygdala or HPC activation, and some have failed to observe activation at all. Indeed, a recent metaanalysis failed to find convincing evidence of higher blood oxygen level-dependent (BOLD) signals evoked during CS+ vs. CSÀ trials in either structure (Mechias et al., 2010). Subsequently, it has been argued that hemodynamic signals are ill-suited or even unable to detect the sparse neuronal activity underlying fear conditioning (Bach et al., 2011). An alternative explanation, however, is that human studies may not always evoke fear levels comparable to rodent studies, and therefore may not engage the same neural circuits.
Here we investigated in freely-moving rats whether hemodynamic responses in the BLA, dorsal HPC (dHPC) and ventral HPC (vHPC) could distinguish between threat-predicting (CS+) and neutral (CSÀ) cues. We recorded hemodynamic responses in the form of tissue oxygen (T O2 ) signals, which are closely related to the BOLD signal and are dependent upon the same physiological mechanisms ( Fig. 1; Thompson et al., 2003;Offenhauser et al., 2005;Logothetis, 2007;Lowry et al., 2010;Li et al., 2011;McHugh et al., 2011). CS+ and CSÀ-evoked T O2 signals were recorded during pre-exposure, training and extinction of conditioned fear.

Subjects
This study used eight naïve male Sprague-Dawley rats (Harlan Olac, Bicester, UK), approximately 3 months old at the start of the experiment, and housed in a temperature-and humidity-controlled room under a 12 h light : dark cycle (lights on 07:00-19:00 h). Testing took place during the light cycle. Rats were housed four per cage before surgery, singly for 1 week after surgery, and two per cage thereafter, with ad libitum food and water. The experiments were conducted in accordance with the United Kingdom Animals Scientific Procedures Act (1986) under project license PPL 30/1989, and were approved by local ethical review for the University of Oxford.

Electrode construction
T O2 was recorded amperometrically using carbon paste electrodes (CPEs), as described previously . CPEs were constructed from 8T Teflon ® -insulated silver wire (200 lm bare diameter, 270 lm coated diameter; Advent Research Materials, Suffolk, UK) soldered into gold-plated socket contacts (E363/0, Plastics One, Roanoke, VA, USA). The Teflon ® insulation was slid along the wire to create a~2-mm-deep cavity, which was packed with carbon paste (2.8 g carbon powder; Sigma-Aldrich, St Louis, MO, USA; Catalogue No. 282863, mixed with 1.0 mL of silicone oil; Sigma-Aldrich;Catalogue No. 17563-3). The Teflon ® insulation on the CPEs was flush with the compacted and smoothed carbon paste such that the active part of the electrode was a flat disk with diameter 200 lm (area: 0.03 mm 2 ). Reference and auxiliary electrodes were made from 8T Teflon ® -insulated silver wire, with one end soldered into a gold-plated contact (E363/0, Plastics One).

Surgery
Rats were anesthetized with Isoflurane (3-4% in 4 L/min O 2 for induction, 1.5-2% in 1 L/min O 2 for maintenance) and placed in a stereotaxic frame with the head level between bregma and lambda. Body temperature was maintained at 37°C via a homoeothermic blanket connected to a rectal thermometer (Harvard Apparatus, MA, USA). Lignocaine (1%) was topically applied to the skin, and Viscotears ® (Novartis Pharma AG, Basel, Switzerland) was used to prevent corneal drying. An incision was made along the midline and the periosteum retracted. Skull holes were drilled to allow the insertion of electrodes and skull screws. Rats were implanted with CPEs into the BLA [anterior-posterior (A-P): À2.8; medial-lateral (M-L): +5.0; dorsal-ventral (D-V): À7.0], dHPC (A-P: À3.6; M-L: +2.2; D-V: À3.2) and vHPC (A-P: À5.3; M-L: À4.7; D-V: À6.5). Reference and auxiliary electrodes were implanted into the cortex. The gold contacts from each electrode were inserted into a six-channel electrode pedestal (MS363, Plastics One), which was secured to the skull with dental cement. Analgesia (Meloxicam,~2 mg/kg; s.c.) was given before and after surgery.

Amperometric techniques
T O2 was measured using constant potential amperometry. A constant potential (À650 mV relative to a reference electrode) was applied to the CPEs using a low-noise potentiostat ('Biostat', ACM Instruments, Cumbria, UK). The applied potential produces the electrochemical reduction of dissolved O 2 on the surface of the CPEs, inducing an electrical current, which was measured by the potentiostat. The availability of O 2 for this two-step reaction (O 2 + 2H + + 2e À ? H 2 O 2 ; H 2 O 2 + 2H + + 2e À ? 2H 2 O) is determined by the local T O2 concentration. Thus, changes in O 2 concentration around the tip of the CPE produce directly proportional changes in the measured Faradaic current (Hitchman, 1978). The spatial resolution is sufficient to detect T O2 differences between lamina in rat whisker barrel cortex, i.e. approximately~400 lm (Li et al., 2011). Accordingly, the area of sensitivity is estimated to be a sphere with diameter twice the electrode surface diameter, i.e. a 200-lm-diameter electrode has~400-lm-diameter sphere of sensitivity (see also Thompson et al., 2003).

Data recording
Rats were connected to the potentiostat via a six-channel commutator (SL6C, Plastics One) using screened cables (363-363 6TCM, Plastics One). A Powerlab ® 8/30 (AD Instruments, Oxon, UK) was used for analog/digital conversion, and data were collected on a Windows PC running Chart ® v5 software (AD Instruments). T O2 data were sampled continuously at 4 kHz.

Apparatus
The pre-exposure and training sessions were conducted in an operant chamber (ENV-008-CT; Med Associates, Lafayette, IN, USA), illuminated by a ceiling-mounted house-light (ENV-215M; Med Associates). The grid floor (ENV-005) of the chamber had 18 stainless-steel rods connected to a shock generator (ENV-414S). A speaker (ENV-224AM) and a mechanical clicker device (ENV-135M) were mounted on the same wall of the chamber, at the same height. A counter-weighted arm (PHM-110P1) attached to the rear wall of the operant chamber held a swiveling commutator (SL6C, Plastics One) that allowed free movement of the tethered rat. The operant chamber was placed inside a sound-attenuating melamine cubicle (ENV-018MD). A CCD video camera with a wide-angle lens (VPC-465B, CES AG, Zurich, Switzerland) provided a full view of the interior of the chamber. Tissue paper was used to line the waste pan and this was replaced for each rat.
The extinction session took place in a novel contexta transparent plastic box (220 mm wide, 200 mm deep, 250 mm high) placed inside a sound-attenuating melamine chamber (350 mm wide, 395 mm deep, 475 mm high). The clicker and speaker were mounted on the same wall of the box, at the same height. An infrared video camera (Maplin Electronics, UK) was mounted above the box, providing a full view of the interior. The discriminability between the novel and training contexts was enhanced in three ways. First, the floor of the novel context was solid plastic and lined with tissue paper, providing a strong tactile contrast with the metallic grid floor of the training context. Second, the novel context was not illuminated, whereas the house light was on in the training context. Third, the novel context was given a distinct odor by placing one drop of sandalwood oil (Home Fragrance Essential Oils, Body Shop, UK) onto fresh tissue paper before each session began.
CS and US delivery were controlled by a script written in MED-PC IV software (Med Associates). In each session, the rat heard five tones (2900 Hz, 87 dB, 30-s duration) and five clickers (10 Hz, 87 dB, 30-s duration) in random order, but with no more than two consecutive presentations of the same stimulus. A relatively long CS duration (30 s) was chosen so that the hemodynamic response evoked by the CS could fully evolve before the arrival of the US. Note that during extinction, CS+1 preceded CSÀ1 for 50% of the rats, and vice versa for the other 50%. The interval between CS presentations was randomly varied between 70 and 110 s (mean: 90 s; SD: 14.9 s). The US consisted of a scrambled footshock (0.5 mA, 0.5 s duration), delivered through the bars of the grid floor, that co-terminated with offset of the CS+ (i.e. US onset at 29.5 s after CS onset). The script controlled the delivery of transistor-transistor-logic pulses to the Powerlab 8/30 ADC to ensure that CS and US delivery were accurately synchronized to T O2 data acquisition (1-ms resolution).

Procedure
The aim was to adopt a fear-conditioning paradigm that resembled those used in human fMRI studies in order to be able to compare the T O2 signals with BOLD responses associated with fear-inducing stimuli. Specifically, this involved using a discriminative fear-conditioning paradigm in which a CS+ (tone or clicker, counterbalanced) was paired with an aversive US (footshock); whereas a CSÀ (clicker if tone was CS+, tone if clicker was CS+) was never paired with footshock. Thus, T O2 signals evoked by the CS+ could be compared directly with signals evoked by the CSÀ, allowing for a CS+ vs. CSÀ contrast as would be used in BOLD fMRI. Fear conditioning began 3 weeks after surgery, and the experimental design is shown in Table 1.
Rats were first pre-exposed to the tone and clicker without any footshocks (Day 1: 5 9 tone, 5 9 clicker). This pre-exposure phase was included because pre-exposure to the CS+ and CSÀ are commonly used in both human fMRI (Buchel et al., 1998(Buchel et al., , 1999LaBar et al., 1998) and rodent electrophysiology experiments (Goosens et al., 2003;Moita et al., 2003;Herry et al., 2008). After the preexposure session, the rats then received 10 training trials over 2 days in which either the tone or clicker was always followed by footshock (Days 2 and 3: 5 9 CSÀ, 5 9 CS+ footshock, per day). Rats were conditioned over 2 days because pilot testing revealed that one session was insufficient to see robust CS+/CSÀ discrimination. Training was followed by extinction in a novel context during which no shocks were given after either the CS+ or CSÀ (Day 4: 5 9 CSÀ, 5 9 CS+; no shocks).
Apart from the presence or absence of shocks, the testing procedure on each day was otherwise identical. The rat was taken to the testing room in a transport cage, connected to the recording equipment, and placed into the conditioning chamber (Days 1-3) or the novel context (Day 4). The potential was then applied to the electrodes for 15 min to ensure a stable T O2 signal. Then the 22-min experimental session began (total time in box per day: 37 min). At the end of each session, the rat was removed from the chamber, disconnected from the equipment and returned to the holding room. The chamber was cleaned with a solution of 10% (v/v) ethanol in water, and the napkin lining the waste pan (training context) or lining the floor (novel context) was replaced before the next session began.

Data analyses
Freezing behavior was measured using a script in NIH Image (NIH, Bethesda, MD, USA). The script collected one frame of video per second and compared consecutive frames for changes in pixels. If a sufficient proportion of pixels changed then the rat was adjudged to be moving; if not, then the rat was adjudged to be freezing. The script was calibrated so that pixel changes caused by random camera noise and breathing movements were not sufficient to signal a movement score. This automated system has over 80% concordance with human ratings of freezing behavior. A detailed description can be found in Richmond et al. (1998). The counts of freezing for each 30-s CS presentation were then expressed as a percentage. For example, a rat freezing for 18 s during the 30-s CS would receive a freezing score of 60%. For correlational analyses with T O2 responses, we also calculated the freezing difference score by subtracting the mean CSÀ-evoked freezing from the mean CS+-evoked freezing.
T O2 data were down-sampled from 4 kHz to 10 Hz, and the signal change from baseline (DT O2 signal) was calculated by subtracting the mean T O2 signal during the 5 s before CS onset (i.e. baseline) from the T O2 signal during the 30-s CS presentation. Then, the 30-s signal was divided into 15 9 2-s timebins (i.e. 0-2 s, 2-4 s, 4-6 s, …, 28-30 s), with each data point equal to the mean value during each 2-s timebin. To investigate the correlation with behavior, we also calculated a T O2 difference score by subtracting the mean CSÀ-evoked T O2 signal from the mean CS+-evoked T O2 signal. Difference scores were calculated separately for each rat and for each brain region for the second training day and the extinction day.

Statistical procedures
Data were analysed using t-tests, analysis of variance (ANOVA) and Pearson correlation in SPSS (version 15; SPSS, IL, USA). ANOVAs are described using a modified version of Keppel's (1982) notation in which the dependent variable is defined in the form: A 2 9 B 3 9 S 8 , where A is a factor with two levels, B a factor with three levels, and S 8 denotes that eight subjects were included in the analysis.
For each day, we calculated the mean freezing and T O2 responses for the five CS+ vs. the five CSÀ trials. T O2 responses were analysed separately for each brain region (BLA, vHPC, dHPC) using repeated-measures ANOVA, with CS type (CS+ vs. CSÀ) and timebin (1-15, covering the 30-s CS duration in 2-s timebins) as within-subject factors [i.e. CS type 2 (CS+,CSÀ) 9 timebin 15 9 S 8 ]. For the extinction session, we also investigated responses to the first and the last CS+ and CSÀ trials of the session: i.e. CSÀ1, CS+1, CSÀ5, CS+5. The rationale for this approach was that responses to CS+1 (vs. CSÀ1) reflect the pure expression of conditioned fear, which is logically unconfounded by any extinction process whereas, comparatively, responses to CS+5 (vs. CSÀ5) reflect, in part, the consequences of extinction, i.e. learning that the CS+ is no longer followed by shock. These extinction session data were analysed using the following models, for freezing: CS type 2 (CS+,CSÀ) 9 trial number 2 (CS1,CS5) 9 S 8 ; and for T O2 : CS type 2 (CS+,CSÀ) 9 trial number 2 (CS1,CS5) 9 timebin 15 9 S 8 . Separate analyses for CS+ and CSÀ stimuli, including all five trials of the extinction session, were also conducted. For all analyses, interactions were investigated using simple main effects (SME) and pairwise comparisons with the familywise error rate at a = 0.05. Pairwise comparisons for CS type 9 timebin interactions (e.g. for CS+ vs. CSÀ) are reported as time periods after stimulus onset (e.g. higher CS+ vs. CSÀ signals during timebins 4, 5 and 6 are reported as: CS+ > CSÀ, 6-12 s, P < 0.05). Figures show the mean ± 1 standard error of the mean (SEM) for CS+-and CSÀ-evoked responses (upper panels), and the difference contrast (CS+ > CSÀ) ± 1 standard error of the difference (lower panels). Note that negative values for the difference contrast indicate CSÀ > CS+. We found no evidence of a CS+ vs. CSÀ contrast in either the behavioral or T O2 data on the first day of training, and so only data from training Day 2 are shown in the main figures. Data from training Day 1 are shown in Fig. S1.

Histology
At the end of the experiments, rats were injected with Euthatal (200 mg/mL sodium pentobarbitone; 200 mg/kg i.p.) and perfused transcardially with physiological saline (0.9% NaCl) followed by 10% formol saline (10% formalin in physiological saline). The brains were then removed and placed in 10% formol saline. The brains were then transferred to 30% sucrose-formalin and frozen. Coronal sections (50 lm) were cut on a freezing microtome and stained with Cresyl violet to enable visualisation of the electrode placements.

Histology
CPEs were successfully targeted to the BLA, dHPC and vHPC in every rat, as shown in Fig. 2. There were no systematic differences in T O2 signals based on exact placement within the BLA or subfield placements within dHPC or vHPC. No rats were excluded on histological grounds.

Behavior
During the pre-exposure session, rats were initially active during CS presentations (i.e. they exhibited low levels of freezing), but became less active towards the end of the session, presumably reflecting habituation to the conditioning chamber. Importantly, behavioral responses to the CS+ and CSÀ were almost identical during the pre-exposure session (Fig. 3A).
Somewhat surprisingly, during the training sessions, there was no significant behavioral discrimination, perhaps because discriminative freezing was masked by contextual conditioning and/or the unconditioned response to the shock. Analysis of freezing responses during the 10 CS+ vs. 10 CSÀ training trials (ANOVA: CS type 2 9 trial number 10 9 S 8 ) revealed no effect of CS type, trial number or interaction (all F < 1.2; P > 0.3; see Fig. S2).
However, even though rats did not discriminate during training, they did learn to discriminate between the CS+ and CSÀ, and during the extinction session they froze significantly more during CS+ than CSÀ trials. Analysis of CS+ and CSÀ evoked freezing over pre-exposure, training and extinction (ANOVA: day 3 9 CS type 2 9 S 8 ) revealed a significant CS type 9 day interaction (F 2,14 = 3.9; P = 0.045), with higher freezing during CS+ than CSÀ trials on the extinction day (P = 0.003) but not during preexposure or training (both P > 0.8). Thus, during the extinction session, the rats demonstrated clear behavioral discrimination between the CS+ and CSÀ.
The previous analysis of the extinction day used the averaged freezing response for the five CS+ vs. the five CSÀ presentations. To dissociate fear expression from fear extinction, we also investigated responses to the first and last CS+ and CSÀ trials of the extinction session [i.e. (CS+1 vs. CSÀ1) vs. (CS+5 vs. CSÀ5)], using the rationale that rats would expect footshock following CS+1 but, after four non-reinforced trials, would be less likely to expect shock following CS+5. Consistent with this prediction, all eight rats froze more during CS+1 than CSÀ1, but only two out of eight rats froze more during CS+5 than CSÀ5 (Fig. 3B). Analysis of these data (ANOVA: CS type 2 9 trial number 2 9 S 8 ) revealed a significant CS type 9 trial number interaction (F 1,7 = 7.6; P = 0.03), with higher freezing during CS+1 than CSÀ1 (F 1,7 = 41.3; P < 0.001) but no difference between CS+5 and CSÀ5 (F 1,7 = 0.1; P > 0.9). Thus, rats froze more during the CS+ than the CSÀ at the start of the session but not at the end, indicating that the CS+ was no longer treated as a threat stimulus. Furthermore, separate one-way ANOVAs, conducted across all five trials of the extinction session, revealed a significant main effect of trial number for the CS+ trials (F 4,28 = 3.0; P = 0.03), which reflected the general decrease in levels of freezing across the extinction session, but no significant main effect of trial number for the CSÀ trials (F 4,28 < 1; P = 0.4).

BLA T O2 signals respond to changes in 'CS+ ? US' contingency
During the pre-exposure session, stimulus onset evoked a smallamplitude, positive-going T O2 response in the BLA to both stimuli (Fig. 4). There were no differences in T O2 responses to the 'to-beallocated' CS+ and CSÀ (no effect of CS type or CS type 9 timebin interaction: all F < 0.2; P > 0.9). Training led to a marked change in the amplitude and shape of the CS+-and CSÀ-evoked T O2 signals (see Figs 4 and S1) and, on the second training day, CS +-evoked signals were significantly higher than CSÀ-evoked signals (main effect of CS type: F 1,7 = 6.9; P = 0.03; CS type 9 timebin interaction: F 14,98 = 2.0; P = 0.03; SME: CS+ > CSÀ, 2-16 s, P < 0.05; 20-30 s, 0.05 < P < 0.1). Thus, during training Day 2, BLA T O2 signals were significantly higher during CS+ than CSÀ trials, and this was found in the absence of significant behavioral discrimination.

BLA T O2 signal correlates with behavioral discrimination during training
Next, we investigated the relationship between BLA T O2 signals and freezing behavior. We calculated a T O2 difference score based on the CS+ > CSÀ contrast from human fMRI studies (i.e. CS+evoked T O2 signal minus CSÀ-evoked T O2 signal) for each rat and plotted this against the corresponding freezing difference score (i.e. CS+-evoked freezing minus CSÀ-evoked freezing). Thus, each data point in Fig. 5 represents the BLA T O2 difference score vs. the freezing difference score for each rat (n = 8). BLA T O2 difference scores were significantly correlated with freezing difference scores during training (r = 0.7; t 6 = 2.5, P < 0.05; Fig. 5A). Thus, although at a group level freezing was not significantly higher during training for CS+ than CSÀ trials, the BLA T O2 signal contrast was significantly correlated with the magnitude of behavioral discrimination.

BLA T O2 responses change to reflect 'CS+ ? no US' association during extinction in a novel context
During the extinction session, BLA T O2 responses averaged over the five CS+ vs. the five CSÀ trials were not statistically different (F < 1.1; P > 0.3; see Fig. S1). However, as illustrated in Fig. 4, . BLA T O2 signals were higher during CS+1 than CSÀ1, but were actually lower during CS+5 than CSÀ5. Analysis of these BLA T O2 responses revealed a significant three-way interaction between CS type, trial number and timebin (F 14,98 = 5.1; P = 0.001). Dissection of this three-way interaction using analysis of SME revealed three key findings.
First, BLA T O2 signals were significantly higher during CS+1 vs. CSÀ1 of extinction (CS type 9 timebin interaction: F 14,98 = 3.9; P = 0.001; CS+1 > CSÀ1, 26-30 s, P < 0.05), consistent with what was seen behaviorally and confirming that the higher CS+evoked T O2 signals seen during training were also present at the start of extinction.
Third, T O2 signals were actually significantly lower during CS+5 than CSÀ5 (CS type 9 timebin interaction: F 14,98 = 2.3; P = 0.01; CSÀ5 > CS+5, 0-4 s and 18-28 s, P < 0.05). Thus, at the start of extinction T O2 signals were significantly higher during CS+ than CSÀ trials, but by the end of extinction they were significantly higher during CSÀ than CS+ trials. This is consistent with the idea that the 'CS+ ? no US' association formed during extinction leads to inhibition of the CS+-evoked BLA T O2 signal.

BLA T O2 signals correlate with the reduction in freezing during extinction
Next, we investigated the relationship between CS+-evoked BLA T O2 signals and freezing behavior specifically during extinction. Six out of eights rats froze more during CS+1 than CS+5, indicative of extinction. In contrast, two out of eight rats did not exhibit reduced freezing from CS+1 to CS+5. Interestingly, in these two rats there was very little difference (or a negative difference) in the BLA T O2 signal contrast between CS+1 and CS+5 (i.e. CS+1 minus CS+5). In other words, like the freezing behavior, the BLA T O2 signals were as high (or higher) during CS+5 than CS+1 in these two rats. Conversely, in the six rats that froze more during CS+1 than CS+5, the corresponding BLA T O2 signals were higher during CS+1 than CS+5. To analyse these data, we calculated difference scores for freezing behavior (CS+1-evoked freezing minus CS+5-evoked freezing) and BLA T O2 signals (CS+1-evoked T O2 signals minus CS+5-evoked T O2 signals), and plotted these against each other (Fig. 5B). The difference between CS+1-and CS+5evoked BLA T O2 signals was highly correlated with the decrease in freezing from CS+1 to CS+5 (r = 0.8; t 7 = 3.1, P = 0.02). In other words, the decrease in the T O2 signal from CS+1 to CS+5 predicted the reduction in freezing behavior seen across the extinction session.

HPC T O2 signals are higher during CS+ than CSÀ trials during acquisition
During pre-exposure, CS+-and CSÀ-evoked T O2 signals in dHPC and vHPC were different to those seen in the BLA, with CS onset leading to negative-going signals in both HPC subregions (compare leftmost traces in Fig. 4 vs. Fig. 6A and B). Importantly, there were no differences between the 'to-be-allocated' CS+-and CSÀ-evoked signals in either dHPC or vHPC during pre-exposure (no effect of CS type or CS type 9 timebin interaction: all F < 1.2; P > 0.3). However, as with the corresponding amygdala T O2 responses, training changed the direction and shape of responses in both HPC regions and, by the second training day, CS+-evoked vHPC T O2 signals were significantly higher than CSÀ-evoked signals (CS type 9 timebin interaction: F 14,98 = 5.1; P = 0.001; CS+ > CSÀ, 4 -8 s, P < 0.05). This did not reach significance in dHPC, but there was a trend toward higher CS+-evoked signals (main effect of CS type: F 1,7 = 3.9; P = 0.09; no interaction: F 14,98 = 1.3; P > 0.2). Thus, like the BLA, vHPC T O2 signals during training were higher during CS+ than CSÀ trials. However, unlike the BLA, neither vHPC nor dHPC difference signals (i.e. the CS+-minus CSÀ-evoked T O2 signal) were significantly correlated with the corresponding freezing difference scores during training (dHPC: r = 0.59, P > 0.1; vHPC: r = À0.57, P > 0.1).  The graph shows BLA T O2 signal difference contrast (CS+ minus CSÀ) vs. freezing difference contrast (CS+ minus CSÀ) for Day 2 of training. Rats that had higher CS+ than CSÀ T O2 signals exhibited higher freezing to the CS+ than CSÀ. (B) The BLA T O2 signal difference contrast (CS+1ÀCS+5) vs. the corresponding freezing difference contrast (CS+1ÀCS+5) during extinction. Larger decreases in BLA T O2 signal from CS+1 to CS+5 were associated with larger decreases in freezing behavior from CS+1 to CS+5. In both graphs, each data point represents the T O2 vs. freezing difference score for individual rats (n = 8). CS, conditioned stimulus.
dHPC and vHPC T O2 signals are lower during CS+ than CSÀ trials during extinction During extinction, dHPC and vHPC T O2 responses were again markedly different from those seen in the BLA (see Figs 6 and S1). As shown in Fig. 6, CS+-evoked T O2 responses in both dHPC and vHPC decreased below baseline after 12-14 s and were lower (on average) than CSÀ-evoked responses. Indeed, mean dHPC T O2 signals (averaged over the five CS+ vs. the five CSÀ trials of the session) were significantly lower during CS+ than CSÀ trials (main effect of CS type: F 1,7 = 10.5; P = 0.01; CS type 9 timebin interaction: F 14,98 = 7.8; P = 0.001; CSÀ > CS+, 14-30 s, P < 0.05). The same pattern of lower CS+-evoked signals was seen in vHPC (main effect of CS type: F 1,7 = 5.9; P = 0.045; CS type 9 timebin interaction: F 14,98 = 4.1; P = 0.001; CSÀ > CS+, 16-28 s, P < 0.05). Note that unlike signals in the BLA, there was no pronounced change in response from CS+1 to CS+5 in the HPC signals across the extinction session, so it is unlikely that the lower CS+than CSÀ-evoked responses were a consequence of extinction (see Fig. S1). Nevertheless, both dHPC and vHPC T O2 signals clearly discriminated between the discrete auditory stimuli (CS+ and CSÀ) during the extinction session.

Summary of results
Behavioral responses and T O2 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 CSÀ trials during the extinction session in a novel context but not during training. In the BLA, T O2 signals were higher during CS+ than CSÀ trials during training and at the start of extinction, but were significantly lower during CS+ than CSÀ 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+/CSÀ discrimination present in the BLA T O2 signals were strongly correlated with behavioral discrimination, even though at a group level the rats did not freeze significantly more during CS+ than CSÀ trials. Also, during extinction, the decrease in the BLA T O2 signal from CS+1 to CS+5 was significantly correlated with the decrease in freezing responses from CS+1 to CS+5. T O2 signals in dHPC and vHPC also discriminated between the CS+ and CSÀ, 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. CSÀ (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+/CSÀ discrimination in the BLA T O2 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., , 2001Armony & Dolan, 2002;Tabbert et al., 2005Tabbert et al., , 2006Knight 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(Tabbert et al., , 2006Knight 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 CSÀ.
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 CSÀ had higher CS+than CSÀ-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., 2006Cheng et al., , 2007. A robust finding from these studies is the positive correlation between differential CS+/CSÀ SCRs and differential amygdala BOLD signals. This is consistent with our study, which found a strong correlation between differential freezing responses and differential BLA T O2 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 CSÀ may be the critical determinant of differential amygdala activation.

HPC T O2 signals discriminate between the CS+ and CSÀ
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 T O2 responses to discrete auditory cues. HPC T O2 signals discriminated between CS+ and CSÀ trials, both during training (vHPC) and during extinction (vHPC and dHPC). Moreover, HPC T O2 signals were markedly different from BLA T O2 signals, which is important because it shows that T O2 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 T O2 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 CSÀ-evoked signals is transient (i.e. the CS+ response is greater than the CSÀ 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 CSÀ-evoked signals in both HPC subregions during extinction (i.e. there was a CSÀ > CS+ contrast). We do not know why the CS+-evoked HPC T O2 response is negative (compared with both the CSÀ 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 CSÀ, is associated with this negative HPC response is not immediately obvious and requires further investigation. Nevertheless, higher CSÀ-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 T O2 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 fearconditioned 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 T O2 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 T O2 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 T O2 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 T O2 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 T O2 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.

Conclusion
To conclude, in rats, hemodynamic responses in the BLA, dHPC and vHPV can detect the differential patterns of neuronal activity evoked by threatening vs. neutral stimuli.

Supporting Information
Additional supporting information can be found in the online version of this article: Fig. S1. Behavioral (A) and tissue oxygen (T O2 ) responses in the basolateral amygdala (BLA; B), dorsal hippocampus (dHPC; C) and ventral hippocampus (vHPC; D) during pre-exposure, training and extinction of Pavlovian fear conditioning. Columns 1-4 show mean responses to the five CSÀ and five CS+ presentations during each day of phase I (pre-exposure, training I, training II and extinction, respectively). Columns 5 and 6 show responses for the first and last CS+ and CSÀ trials of extinction (i.e. column 5: CSÀ1 vs. CS+1; column 6: CSÀ5 vs. CS+5). (A) Mean (±SEM) percentage freezing during the 30 s of the CSÀ (blue bars) or CS+ (red bars). (B-D) Mean (±SEM) T O2 change from baseline (DT O2 ) during the 30 s of CSÀ (blue traces) and CS+ (red traces) in 2-s timebins. n.s., no significant difference in paired t-test, or no main effect of CS or CS 9 timebin interaction; *significant difference in paired t-test (P < 0.05) or main effect of CS (P < 0.05); # CS 9 timebin interaction (P < 0.05). Fig. S2. Freezing responses to the CSÀ (blue) and CS+ (red) during the last five trials of training. There was no evidence of discriminative freezing (i.e. higher freezing during CS+ trials) during any of the training trials.