Cerebellum regulates the timing of periaqueductal grey neural encoding of fear memory and the expression of fear conditioned behaviours

The pivotal role of the periaqueductal grey (PAG) in fear learning is reinforced by the identification of neurons in rat ventral (vPAG) that encode fear memory through signalling the onset and offset of an auditory conditioned stimulus during retrieval. Within this framework, understanding of cerebellar contributions to survival circuits is advanced by the discovery that: (i) reversible inactivation of the medial cerebellar nucleus (MCN) during fear consolidation (a) reduces the temporal precision of vPAG offset, but not onset responses and (b) increases rearing behaviour during retrieval, and (ii) chemogenetic inhibition of the MCN-vPAG projection during fear acquisition (a) reduces the emission of fear-related ultrasonic vocalisations and (b) slows the extinction rate of fear-related freezing. These findings show that the cerebellum regulates fear memory processes at multiple timescales and in multiple ways. The current findings indicate that dysfunctional interactions in the cerebellar-survival network may underlie fear-related disorders and comorbidities. Impact Statement Cerebellar-periaqueductal grey interactions contribute to fear conditioned processes and, as such, provide a novel target for treating psychological conditions including post-traumatic stress disorder.


Introduction
The periaqueductal grey (PAG) lies at the hub of central networks that co-ordinate survival, including 3 coping behaviours; from those subserving homeostatic and reproductive functions to those mediating 4 defensive, fear-evoked coping responses. Neurons in the functionally distinct longitudinal columns of 5 the PAG co-ordinate different aspects of survival behaviours. Of particular relevance to the current 6 study, fear-evoked freezing is mediated by neurons in its ventral sector (vPAG; Vianna et al., 2001;7 Walker and Carrive, 2003; Tovote et al., 2016;Watson et al., 2016). Fear-related behaviours co-8 ordinated by the PAG also extend to the expression of 22kHz ultrasonic vocalizations (USVs; Kim et al., 9 2013; Ouda et al., 2016) and risk assessment activity such as rearing (Sandner et al., 1987;Clelland et 10 al., 2010). Furthermore, at a cellular level, electrophysiological studies have found that neurons in 11 vPAG encode associatively conditioned fear memory (Watson et al., 2016;Wright et al., 2019). 12 Central nervous system survival networks involving the PAG are well documented (Tovote et al., 2015). 13 Until recently, the cerebellum was generally considered not to be part of this network. However, there 14 is good evidence in cats and rodents that vermal regions of the cerebellum contribute to motor and 15 autonomic components of defensive states; inactivation of vermal cerebellar cortex (lobules IV-VIII), 16 or one of its main output nuclei (medial cerebellar nucleus, MCN, aka fastigial nucleus), leads to 17 deficits in fear-related behaviours such as context conditioned bradycardia (Supple et al., 1990) and conditioned fear unmasks risk assessment rearing behaviour (Koutsikou et al., 2014). 23 The cerebellum is reciprocally connected to many brain regions associated with survival networks 24 (Sacchetti et  In particular, an emerging concept is that the cerebellar vermis and its output nucleus MCN are 32 involved in the control of an integrated array of fear-related functions, including fear memory and 33 fear-induced behaviours such as freezing. Important insights into the role of the murine MCN-vPAG 1 pathway in modulating vPAG fear learning and memory have been provided recently (Vaaga et al., 2 2020; Frontera et al., 2020). However, it is not known whether PAG neural encoding of fear memory 3 is dependent on the integrity of its cerebellar input. More generally, given the well-established role of 4 the cerebellum in the co-ordination of movements and in particular, the representation of temporal 5 relationships (e.g. Xu  are that: (i) vPAG encodes temporally precise information about the onset and offset of a fear 16 conditioned stimulus and that these two neural signals may be generated by independent 17 mechanisms; (ii) MCN output regulates the temporal accuracy of vPAG encoding of fear conditioned 18 stimulus offset but not onset during retrieval (early extinction) of the conditioned response; (iii) MCN 19 output also regulates multiple aspects of survival behaviour at different times during fear 20 conditioning: during acquisition the emission of USVs and the suppression of rearing and also the rate 21 of extinction of conditioned freezing during retrieval. In summary, the present study provides 22 evidence that the cerebellum regulates the ability of the PAG to encode a fear memory trace with 23 temporal precision and also regulates the time of occurrence of appropriate patterns of behaviour 24 during fear acquisition and retrieval. 25 26 27 Results 28 vPAG unit responses during an auditory cued fear conditioning paradigm 29 As a first step, an auditory cued fear conditioning paradigm (Fig. 1A) was used to investigate how single 30 unit activity in the vPAG (n=11 animals) responded to fear acquisition and subsequent retrieval and 31 extinction of fear conditioned responses (Fig. 1B-E). In seven of these animals a dual microdrive 32 included tetrodes that were also implanted in the contralateral MCN to record local field potential 33 activity. The remaining four animals had a single microdrive for vPAG recording combined with 1 bilateral cannulae implanted to target MCN. In all animals the position of tetrodes (and where 2 appropriate cannulae) was histologically verified. The majority of PAG tetrode recording sites were 3 located in the vPAG, and cannulae or tetrodes implanted in the cerebellum were located close to or 4 within MCN ( Supplementary Fig. 1A, B and C). Since similar electrophysiological data were recorded 5 from the PAG obtained from animals with a dual microdrive (PAG and MCN) and those with a single 6 microdrive (PAG, that received a saline infusion into MCN), the results have been pooled in the analysis 7 below (termed 'control', Figs 2 and 3). There was no significant difference between baseline firing 8 rates across habituation, acquisition, and extinction training ( Supplementary Fig. 2, p=0.649, one-way 9 ANOVA). Thus, it seems reasonable to assume that single unit recording was stable over time and 10 comparable between groups.

15
A total of 32 vPAG units were recorded during habituation (Fig. 1 B). The large majority (75%, n=24/32) 1 were unresponsive to the unconditioned auditory tone; the remainder responded (see Methods for 2 definition) to tone onset (either increasing, n=6/32 units or decreasing firing rate, n= 2/32 units); and 3 in some of these cases (15.6%) also to tone offset (either increasing 3/32 units or decreasing firing 4 rate 2/32). 5 We recorded the activity of 50 vPAG units during acquisition where the CS tone was paired with the 6 US footshock at tone offset (Fig. 1C). A small proportion of units (18.5%) responded to CS onset. 7 Following the US, some units (20%) also displayed an increase in firing rate which presumably reflects 8 sensory afferent drive to the PAG as a result of the aversive peripheral stimulus (e.g. (Sanders et al., 9 1980;Heinricher et al., 1987;Sharma et al., 1999). 10 During retrieval and extinction of the conditioned response we recorded a total sample of 55 vPAG 11 units in control animals. In retrieval during early extinction (EE) 29 units (52.7%) displayed a transient 12 increase in activity during presentation of the unreinforced conditioned stimulus (CS+, Fig. 1D, E). In 13 keeping with a previous classification of PAG unit activity during extinction training (Watson et al., 14 2016) these units are therefore defined as type 1. The pattern of response was typically a phasic 15 increase in activity at CS+ onset but an additional feature not previously reported for vPAG units was 16 also a phasic increase in activity at CS+ offset (Fig. 1D 1 onset responses this reduction was not evident for mean peak z-score (Fig. 2D, decrease of 13.8% 33 from EE to LE; paired t test; p=0.197) but was evident for type 1 offset responses (Fig. 3D, decrease of 1 67.0% from EE to LE; paired t test; p = 0.011). This distinction between onset and offset responses 2 provides additional evidence to suggest they may be evoked by separate mechanisms.  have been expected to occur 1 sec later ( Supplementary Fig. 3). Although caution is clearly needed 20 when interpreting a small sample, our results are consistent with vPAG activity at CS+ offset signalling 21 the end of the conditioned stimulus rather than the time of the expected occurrence of the 22 unconditioned stimulus. 23

The effect of temporary MCN inactivation during fear consolidation on vPAG activity 24
In 4 additional animals, muscimol was infused into the MCN to reversibly block cerebellar output 25 during consolidation of the fear associative memory prior to extinction training (termed 'muscimol' in 26 extinction sessions, Figs 2 and 3). Effects of the muscimol infusion on general motor co-ordination 27 were carefully monitored. Immediately after the infusion all animals (n=4) displayed ataxia, providing 28 a positive control that the muscimol was disrupting cerebellar activity. The severity of the ataxia 29 gradually reduced over several hours and the animals were behaviourally normal after 24hrs. After a 30 total delay of 48hrs, to ensure complete washout of the drug, the animals were exposed to the 31 unreinforced CS+ in extinction training. There was no significant difference between baseline firing 32 rates across habituation, acquisition, and extinction training ( Supplementary Fig. 2, p=0.079, one-way 33 ANOVA). A total of 14 type 1 vPAG units were recorded during extinction. Similar to the control results 34 described above, the majority (71%, 10/14 units) of type 1 onset units (Fig. 2B) showed a reduction in 35 responsiveness during extinction training as measured by mean change in integrated area of response 36 ( Fig. 2C, on average the integrated area of response was 77.8 % smaller between EE and LE, p = 0.002, 1 paired t test). In keeping with the findings from control animals, no significant difference was found 2 for mean peak z-score ( Fig. 2D, peak 0.7% smaller between EE and LE, p = 0.971, paired t test). There 3 was also no statistically significant difference between control and muscimol animals for CS+ onset 4 responses in EE or LE for either measure of response ( Fig. 2C and D, mean integrated area in EE, p = 5 0.689, unpaired t test; mean integrated area in LE, p = 0.351, unpaired t test; mean peak z-score in EE, 6 p = 0.919, unpaired t test; mean peak z-score in LE, p = 0.654, unpaired t test,). 7 By marked contrast, inspection of Figure 3A 3A,B shows there was also a reduction in mean peak z-score (muscimol mean peak was 60% of control) 12 but this was not statistically significant ( However, muscimol animals (n=4) displayed a significant increase in rearing activity during EE 1 compared to controls (Fig. 2G, n=8 animals, median 2.0 rears during CS+ by comparison to control of 2 0 rears, p = 0.018, Mann Whitney test). This difference was not evident in LE (p = 0.333, Mann Whitney 3 test). In muscimol animals, rearing activity reduced in LE, but this was not significantly different from 4 EE (p = 0.250, Wilcoxon test). 5 In terms of ultrasonic vocalisations (USVs) 2/8 (25%) of control animals and 2/4 (50%) of muscimol 6 animals emitted USVs during presentation of the CS+. In terms of the mean total number of USVs per 7 animal in control versus muscimol groups across all extinction training there was no significant 8 difference (Fig. 2H, p = 0.321, Mann Whitney test). 9 In summary, for the various behavioural measures studied there was no detectable difference 10 between control and muscimol animals during extinction training with the exception of rearing, where 11 in EE (retrieval), MCN inactivation resulted in a significant increase during presentation of the CS+. 12

(ii) Effects after presentation of the CS+ in extinction training 13
To investigate whether MCN inactivation during consolidation has effects on fear behaviour after CS+ 14 offset, the same behavioural measures studied during delivery of the CS+ were quantified during the 15 inter-CS+ interval. In terms of freezing behaviour during the inter-CS+ interval, both control and 16 muscimol animals showed extinction learning similar to that found during presentation of the CS+ ( The mean total number of USVs across all extinction blocks per animal in muscimol groups was 27 increased during the inter-CS+ interval but this was not significantly different from control (Fig. 3H, p 28 = 0.26, Mann Whitney test). Therefore, based on the available data no systematic differences were 29 detected in the inter CS+ interval between control and muscimol groups in terms of extinction learning 30 or numbers of USVs but a significant difference was detected in terms of an increase in rearing 31 behaviour during EE (retrieval). 32

Population activity in the MCN and vPAG 1
To assess changes in neural population activity during extinction, auditory event related potentials 2 (ERPs) were recorded simultaneously from the MCN and vPAG (n= 5 animals, Fig. 4). The ERP in the 3 vPAG at CS+ onset had a significantly shorter onset latency then the ERP recorded in the same animals 4 in MCN (vPAG onset 6.0 ms ± 1.14 SEM; MCN onset 25.3 ms ± 2.43 SEM, p=0.006, paired t-test). 5 However, latency to peak was not significantly different (PAG peak 65.4 ms ± 1.4 SEM; MCN peak 74.5 6 ms ± 6.6 SEM, p=0.294, paired t-test). At CS+ offset the ERP in the vPAG was also significantly shorter 7 than the ERP in MCN (vPAG offset 28.6 ± 6.3 SEM; MCN offset 47.6 ms ± 2.7 SEM, p=0.037; paired t-8 test), while latency to peak was similar (vPAG peak 89.6 ± 5.7 SEM; MCN peak 99.0 ms ± 6.5 SEM, 9 p=0.310, paired t-test). The difference in onset latency between ERP onset and offset responses 10 recorded within both vPAG and MCN was statistically significant (vPAG onset vs offset, p=0.024, Welch 11 unpaired t test; MCN onset vs offset, p=0.0003, unpaired t test). This difference suggests that ERP 12 onset and offset responses are likely to be generated by different neural pathways. training. D) Same as C but for CS+ offset (n= 5 rats, Paired t-test, *p<0.05, ** p<0.01). 9 1

Inhibition of direct MCN to vPAG projection during fear acquisition and early consolidation 2
Given that during extinction training: (i) population activity in MCN resembles changes in vPAG 3 population and unit activity; and (ii) the finding that global inactivation of MCN in rats can disrupt 4 encoding in vPAG, it was of interest to determine the extent to which a direct projection exists 5 between the MCN and vPAG. In 7 rats a fluorescently tagged anterograde virus was injected into the 6 MCN ( Supplementary Fig. 4). In every case terminal projections in the PAG were primarily localised to 7 its ventrolateral region on the contralateral side. presentation of the unreinforced CS+ (Fig. 6A, EE p=0.645, LE p=0.362, unpaired t test,) and the inter-10 CS+ interval (Fig. 6B, EE p=0.501, LE p=0.422, unpaired t-test). Both groups also showed similar levels 11 of extinction learning during CS+ (control, p =0.0103; DREADD, p <0.0001, paired t test) and the inter-12 CS+ interval (control, p = 0.0042; DREADD, p <0.0001, paired t tests). However, the rate of extinction 13 during the CS+ was significantly slower in DREADD animals (Fig. 6C, p=0.042, unpaired t test), but not 14 during the inter-CS+ interval (Fig. 6D, p=0.229, unpaired t test). The latter finding suggests the effect 15 is not a general disruption of the expression of freezing behaviour. 16 With regard to rearing behaviour no differences were found between control and DREADD animals 17 both within early and late extinction (Fig. 6E,F  case this produced ataxia, although this was generally less severe and shorter lasting (~1h) than was 28 observed in the muscimol animals. For example, immediately after the infusion of muscimol, animals 29 were unable to perform the beam walking task, while the DREADD animals were able to perform the 30 task, although deficits were evident by comparison to baseline performance ( Supplementary Fig. 5F, 31 slower time to traverse, p=0.006, unpaired t-test; increased foot slips, p=0.086, Mann-Whitney test). 32 Discussion. 33 We have shown that vPAG neurons (type 1 units) encode temporally precise information about both 34 the onset and offset of a fear conditioned auditory stimulus and that these two neuronal signals may 35 be generated by independent mechanisms. This is because in vPAG during retrieval and extinction: i) 36 some single units only respond to CS+ onset or only to CS+ offset; ii) unit onset and offset responses 37 exhibited different characteristics during extinction training; iii) MCN inactivation disrupted the vPAG 1 pattern of unit activity at CS+ offset but not onset; and (iv) the latency of the event related potential 2 (ERP) at CS+ onset was significantly shorter than the ERP recorded at CS+ offset. Together, these 3 findings imply different neural pathways generate onset and offset responses. 4 Importantly, vPAG units displayed little or no response to an auditory tone during habituation but 5 displayed robust activity at tone onset and/or offset when the same tone was classically conditioned. 6 This provides evidence that the responses were related to the associative conditioning rather than the 7 sensory stimulus. Consistent with a recent study in mice (Frontera et al., 2020) the present results in 8 rats found that inactivation of the MCN-vPAG pathway during acquisition reduces the subsequent rate 9 of extinction of conditioned freezing behaviour during retrieval, and that MCN inactivation during 10 consolidation has no detectable effect on fear conditioned freezing. However, the present behavioural 11 results advance understanding beyond effects on freezing. Fear state involves a multiple pattern of 12 defence-related responses, including (but not limited to) USVs and risk assessment behaviour such as were generally similar in early and late extinction (ie were extinction resistant) and therefore may 5 contribute to the persistence of fear memory after extinction. 6 In terms of previous reports of CS+ offset responses in vPAG, Wright and McDannald (2019) identified 7 a distinct population of units in an auditory fear discrimination task they termed ramping units. These 8 units were related to threat probability and also to fear output as determined by freezing behaviour. 9 Ramping units progressively increase activity over the duration of the auditory cue presentation, 10 reaching a peak around sound offset and ramping down in activity thereafter. However, this pattern 11 differs markedly from our type 1 offset units whose phasic activity was precisely coupled to CS+ offset. 12 Indeed, we found no evidence of ramping type activity in our sample of vPAG units. This is perhaps 13 unsurprising because the experimental paradigm of Wright and McDannald (2019) differed from ours 14 in a number of important ways. In particular, they used a trace fear conditioning protocol where the 15 aversive footshock was paired in each session with the auditory cue with a varying probability of 16 occurrence. The closest comparison with our results is between our acquisition sessions and their 17 trials when the probability of a footshock was 100% (their Fig. 1). We observed an increase in activity 18 immediately after the footshock and observed a progressive reduction in activity that resembled the 19 change in firing after peak in their ramping units (our Fig. 1C). 20 An unanswered question concerns the origin of CS+ offset responses in vPAG. Our ERP data are 21 consistent with offset responses being generated by synaptic input to the PAG, and we estimate that the cerebellum not only regulates the timing of movement to enable coordinated behaviour and 1 motor learning, but that this timing regulation extends to other functions of the CNS, including 2 perceptual tasks that require the precise timing of salient events (Spencer et al., 2013). The present 3 study extends this concept to the encoding of fear memory by vPAG. Our findings indicate that the 4 cerebellum is important for the regulation of fear memory processes at multiple timescales: at the 5 millisecond timescale to control CS+ offset timing within vPAG, and at longer timescales (hours/days) 6 to regulate the rate of fear extinction and timing of expression of multiple fear-related behaviours. 7 We provide evidence that MCN is not only involved in fear conditioned freezing behaviour but that it 8 is also involved in the expression of USVs during acquisition and rearing behaviour during retrieval. It 9 is tempting to speculate that MCN regulation of vPAG encoding of CS+ offset underlies some if not all 10 of these behavioural effects, but this remains to be determined. but also a concomitant increase in risk assessment behaviours including rearing (Koutsikou et al., 20 2014). This suggests that expression of survival behaviours is regulated by the cerebellum in a context 21 dependent manner. Our results also suggest that activity in the MCN-vPAG pathway during acquisition 22 regulates the subsequent rate of fear extinction. This is consistent with a previous study in mice 23 (Frontera et al., 2020) and suggests this is a function that is conserved across species. Outbred strains 24 of rats have been shown to demonstrate different behavioural phenotypes during fear extinction (Ji 25 et al., 2018) where a proportion of animals show faster rates of extinction than others. Interruption 26 of MCN-vPAG interaction during acquisition or early consolidation may therefore contribute to an 27 anxiety-like behavioural phenotype, with wider implications for possible neural mechanisms that 28 underly psychiatric disorders such as PTSD. 29 In summary, MCN regulates precise temporal encoding of fear memory within vPAG and also regulates 30 the expression of different survival behaviours depending on the phase of Pavlovian fear conditioning. 31 During early extinction MCN output supresses rearing, while a direct pathway to the vPAG appears to 32 be important in eliciting fear related USVs during acquisition, and the rate of expression of freezing 33 during early extinction. The cerebellum through its interactions with the survival network might 1 therefore be coordinating the most appropriate behavioural response at the most appropriate time. Rats were anaesthetised initially with gaseous isoflurane, followed by intraperitoneal injections with 12 ketamine and medetomidine (5mg/100g of Narketan 10 and Domitor, Vetoquinol). Each animal was 13 mounted in a stereotaxic frame with atraumatic ear bars and surgery was performed under aseptic 14 conditions. Depth of anaesthesia was monitored regularly by testing corneal and paw withdrawal 15 reflexes with supplementary doses of ketamine/medetomidine given as required. A midline scalp 16 incision was made, and craniotomies were performed above the cerebellum and/or the PAG as 17 required in each line of experiment. At the end of every surgery, the rat was administered the 18 analgesic Metacam (Boehringer Ingelheim, 1 mg/Kg) and the medetomidine antidote Atipamezole 19 (Antisedan, Vetoquinol 0.1mg I.P.). Tetrodes, cannula and/or viral injections were carried out as 20 described below depending on the experiment. Animals were handled for 1 week prior to surgery and 21 during recovery before any behavioural paradigms or electrophysiological recordings. 22 Electrophysiological recording (n= 15 rats): 23 1. Dual microdrive experiments (n=7 rats). Two in-house built microdrives, designed to slot closely 24 next to each other, were positioned over craniotomies to allow tetrodes to be independently 25 advanced into the right MCN (11.4 mm caudal from bregma, 1 mm lateral from midline, depth of 26 4mm) and contralateral vPAG (7.5 mm caudal from bregma, 1 mm lateral from midline, depth 4.8mm). 27 The microdrives were attached to the skull with screws and dental acrylic cement. Each microdrive 28 contained 3-4 tetrodes for local field potential (LFP) and single unit recording (0.0008-inch Tungsten 29 wire 99.95% CS 500 HML, insulated with VG Bond, 20 µm inner diameter, impedance 100-400 KΩ after 30 gold plating; California Fine Wire). 2. Single microdrive experiments (n=8 rats). These implants were 31 the same as described above except only one microdrive was implanted to record single units from 1 the vPAG, and infusion cannulae were implanted bilaterally (n=8) to target the MCN (details below). animal where the timing of the US was delayed 1sec after the CS. This was followed by a session of 28 extinction training (day 2). During extinction, 5 blocks of 7 tone presentations (trials) were repeated. 29 The first two blocks (trials [1][2][3][4][5][6][7][8][9][10][11][12][13][14] were defined as early extinction (EE), when the animal was exhibiting 30 high levels of freezing, while the last two blocks (trials [21][22][23][24][25][26][27][28][29][30][31][32][33][34][35] were defined as late extinction (LE), 31 when the animals were exhibiting low levels of freezing. 32 Balance beam (n= 18 rats). This test was used to assess general motor coordination and balance 1 (Luong et al., 2011). Animals were trained for 3 consecutive days to cross 6 times a 160 cm long beam 2 that ended on an enclosed safety platform. On each day, the beam was progressively thinner in width 3 (6 cm, 4 cm and 2 cm). The 2 cm beam was then used for the test day. Baseline performance was 4 recorded and then CNO (Clozapine N-oxide) was given either i.p. or by intracranial infusion (for details 5 see below) and after an interval of 15 minutes the animal was re-tested on the beam. Beam balance 6 performance was manually scored using Solomon Coder software (© 2019 by András Péter), by 7 scoring the time to cross the beam for each trial and the number of foot slips. 8 Open field (n= 14 rats). This test was used to assess both general motor behaviour and anxiety levels. 9 Animals were exposed for the first time to the arena (round arena with 90 cm diameter, and hight of 10 51 cm) on the test day. They were placed at the perimeter of the arena and were allowed to explore 11 for 10 minutes. Exploratory behaviour was recorded for the whole session. DeepLabCut (Wei et al., 12 2018) was used to track the animal behaviour in the video, the tracking output was then used to 13 calculate total distance covered and time spent in two equivalent areas of the arena, a central and a 14 peripheral one. 15 Elevated plus maze (n= 18 rats). This test was used as an additional assessment of anxiety (Pellow et 16 al., 1985). Animals were placed in a plus shaped maze, 1m above the floor, with 2 open and 2 closed 17 arms (10 cm wide and 50 cm length) and allowed to explore the maze for 5 minutes. Elevated plus 18 maze performance was manually scored using Solomon encoder software (© 2019 by András Péter) 19 to calculate total time spent in open versus closed arms and the number of entries into each arm. 20

Data acquisition and analysis: 21
Electrophysiological recording. Multisite electrophysiological data were recorded using a Blackrock 22 Microsystems (Utah, USA) data capture system synchronised with OptiTrak software. Neural data 23 were sampled at 30kHz and band pass filtered online between 300Hz-6kHz. 24 Ultrasonic vocalisation recordings. USVs emitted at 22kHz were recorded using an ANL-940- 1 25 Ultrasonic Microphone and Amplifier (Med Associates, Inc.) connected to the Blackrock Microsystems 26 (Utah, USA). Although USVs were recorded as an aliased signal (the maximum sampling rate of the 27 recording system was 30kHz, while the optimal sampling rate was 44kHz) we were able to reliably 28 identify all USV events. For analysis, USVs were visualised using Spike7 software (Cambridge Electronic 29 Design Limited) and individual USV emissions manually identified and the total number during each 30 recording session was counted. 31 Auditory cued fear conditioning. Video recording of animal behaviour during the fear conditioning 1 paradigm was captured with an OptiTrak camera and software, allowing synchronisation with neural 2 data. Fear-related freezing behaviour was manually scored using Solomon Coder software (© 2019 by 3 András Péter). Freezing behaviour was identified as periods of time longer than 2s in which the animal 4 had an absence of movement (except those associated with respiration and eye movements, 5 Blanchard & Blanchard, 1969) while typically maintaining a crouching position. Percentages of time 6 spent freezing were calculated for each trial during CS+ presentations and during inter-CS+ intervals. 7 To evaluate the extinction rate of freezing behaviour, the slope of the intercept of freezing % for all 8 trials was calculated. Rearing activity was counted as events in which the animal was standing upright 9 on its rear limbs. The number of rearing events during CS+ presentations and inter-CS+ intervals was 10 counted. 11 zero. During acquisition, the footshock caused electrical interference so it was not possible to analyse 22 unit activity during the 0.5 sec period of stimulus delivery. The following analysis was performed on 23 MATLAB. In all experimental groups PETHs of unit activity to the unreinforced conditioned tone (CS+) 24 during extinction training were constructed for individual units over all trials  and also separately 25 for early extinction (trials 1-14) and late extinction trials (21)(22)(23)(24)(25)(26)(27)(28)(29)(30)(31)(32)(33)(34)(35). PETHs were z-score normalised to a 26 5 second baseline recording of unit activity before tone onset and data grouped to show the average 27 of all unit responses and separated into different unit response types. A significant response was 28 defined as 1 or more consecutive 40 ms time bins where the z-score was ± 2 SD from baseline mean. 29 The peak response was measured as maximal value found within the first 25x40 ms bins after tone 30 onset or offset. The area of the response was calculated as the trapezoidal numerical integration of 31 the first 25x40 ms bins after tone onset or offset. 32 Data analysis of Local field potential. Local field potential (LFP) data were averaged in relation to tone 1 onset and offset using MATLAB (n= 14 trials per mean of each animal). The tetrode recording site 2 yielding the largest mean peak to trough response was identified in each animal and used to calculate 3 group average data of peak amplitude of LFP response recorded in the PAG and cerebellum. 4 Statistical analysis. Statistical analysis and graphs were performed with GraphPad Prism 9. Data was 5 shown as mean ± S.E.M., except for rearing behaviour and USVs that were shown as median ± IQR. 6 Paired t-tests or Wilcoxon test (for USVs and rearing behaviour) were used for within group 7 comparison, while unpaired t-tests or Mann Whitney test (for USVs and rearing behaviour) and 8 ANOVA were used to compare groups. Differences were considered significant at p < 0.05. 9 Acknowledgments: We gratefully acknowledge Ms Rachel Bissett for her help with histological 10 processing and the Wolfson Bioimaging Facility for their support and assistance. We would like to 11 thank Bryan Roth and Edward Boyden for supplying the viral vectors used in this work. This work was 12 supported by BBSRC grant BB/MO19616/1 and a Wellcome Trust PhD studentship 203775/Z/16/Z. 13 Competing interests: The authors declare that no competing interests exist.