A Brainstem reticulotegmental neural ensemble drives acoustic startle reflexes

The reticulotegmental nucleus (RtTg) has long been recognized as a crucial component of brainstem reticular formation (RF). However, the function of RtTg and its related circuits remain elusive. Here, we report a role of the RtTg in startle reflex, a highly conserved innate defensive behaviour. Optogenetic activation of RtTg neurons evokes robust startle responses in mice. The glutamatergic neurons in the RtTg are significantly activated during acoustic startle reflexes (ASR). Chemogenetic inhibition of the RtTg glutamatergic neurons decreases the ASR amplitudes. Viral tracing reveals an ASR neural circuit that the cochlear nucleus carrying auditory information sends direct excitatory innervations to the RtTg glutamatergic neurons, which in turn project to spinal motor neurons. Together, our findings describe a functional role of RtTg and its related neural circuit in startle reflexes, and demonstrate how the RF connects auditory system with motor functions.

L ocated in the brainstem, the reticular formation (RF) has long been recognized as one of the most important part of the central nervous system (CNS). As its name suggested, the RF resembles a net-like structure composed of numerous ascending and descending nerve fibers along with various heterogeneous nuclei extending from the spinal cord to the thalamus 1 . This widespread projections allow the RF to act as a "command center" of the CNS. The midbrain RF together with its ascending tractsalso known as the reticular activating system (RAS)project diffusely to the cerebellum, thalamus, cerebral cortex, and limbic system, functioning to promote the sleep-wake cycle, arousal, and consciousness 2,3 . Lesions to RAS result in impaired consciousness and even coma in some patients 4,5 . Descending fibers from the RF, mainly reticulospinal tracts, project to the spinal cord to coordinate respiratory control, pain perception, posture maintenance, and muscular reflexes [6][7][8] .
In mammals, the RF comprises over 100 individual nuclei. Most of the nuclei embedded in this structure have no precise cytoarchitectural boundaries 1 , rendering functional studies of these RF nuclei a challenging task. Situated at the ventral extreme of the pontine tegmentum, the reticulotegmental nucleus (RtTg) is one of the least investigated neural ensembles with clear delineations of territory in RF 9 . Early studies in primates indicated that RtTg might be involved with the oculomotor circuits crucial for smooth-pursuit eye movement 10 . On the other hand, controversial conclusions have been demonstrated regarding to the functional relationship between RtTg and locomotion. Unit recordings in monkeys and cats have revealed increased RtTg neuronal activities during limb movements 11,12 . Another early study in mice also observed consistent activation of RtTg during motor seizure 13 . Still, other researchers have observed increased forward locomotion in rats following electronic lesion of the RtTg 14 .
Although sporadic studies on RtTg have been reported over the past few decades, the precise functional role of the RtTg and its brain-wide upstream and downstream projections have not been investigated thoroughly. Recently, the development of cell-typespecific viral tracing approaches as well as optogenetic/chemogenetic tools has enabled more accurate discovery and elucidation of the potential roles and circuit connections in a target brain region 15 . Using these cutting-edge techniques combined with in vitro/in vivo electrophysiological recordings, we surprisingly found that the RtTg controls a conservative startle response, and identified a neural circuit from the cochlear nucleus (CN) to spinal motor neurons that underlies this innate defensive behaviour.

Results
Optogenetic activation of RtTg neurons elicits startle reflexes. We first conducted optogenetic experiments 16 to examine the behavioural readout for RtTg neuronal activation. We unilaterally injected adeno-associated viruses (AAV) expressing Channelrhodopsin-2 (AAV-hSyn-ChR2-EGFP) into the RtTg of mice, followed by implantation of an optical fiber above the injection site (Fig. 1a). Mice transfected with AAV-hSyn-EGFP were used as controls. Brain slice patch-clamp recordings showed that a light-pulse train triggered action potentials in the RtTg ChR2 + neurons, indicating the effectiveness of RtTg neuron activation (Fig. 1b, c). Strikingly, optogenetic activation of RtTg neurons reliably evoked robust startle reflexes, which were defined as involuntarily innate defensive response occurring rapidly (often in milliseconds) after the onset of aversive stimuli (Fig. 1d-f and Supplementary Movie 1). Moreover, the startle amplitudes were positively correlated to the laser intensities (Fig. 1g). As a control, no startle reflex was observed upon photostimulation of RtTg neurons expressing EGFP only ( Fig. 1f, g and Supplementary Movie 2). It is worth noting that the optogenetic manipulation was strictly limited to the RtTg region without influencing the adjacent caudal pontine reticular nucleus (PnC) ( Supplementary Fig. 1), which has been already recognized to be critical for startle reflex by previous studies [17][18][19] .
Previous reports demonstrated that through all mammals, the stereotypical startle pattern consists of a fast contraction of neck, facial and skeletal muscles 17 . We therefore examined the electromyographic (EMG) activities in the neck and hindlimb of mice during optogenetic activation of RtTg neurons. As expected, we consistently detected robust EMG activities in both neck and hindlimb muscles following every single optogenetic stimulation pulse (Fig. 1h, i and Supplementary Fig. 2a, b). The EMG activities occurred very shortly after optogenetic activation of RtTg neurons, with a mean latency of 5.75 ± 0.38 ms for neck muscle (Supplementary Fig. 2c) and 9.43 ± 0.12 ms for hindlimb muscle (Fig. 1j) respectively.
RtTg inactivation specifically suppresses acoustic startle reflexes. We next used chemogenetic suppression-based loss-offunction experiments to further verify the potential role of RtTg in acoustic startle reflex (ASR), a major form of startle behaviours. Specifically, the AAV virus expressing mCherry-tagged hM4Di 20 an inhibitory designer receptor exclusively activated by the designer drug (DREADD)was bilaterally injected into the RtTg (Fig. 2a). We then employed a well-established paradigm 21 to trigger and measure the ASR in mice. Transient suppression of RtTg neurons by intraperitoneal (i.p.) injection of clozapine-N-oxide (CNO) significantly decreased ASR amplitudes ( Fig. 2b) without affecting other innate behaviours including motor coordination ( Fig. 2c-h) and pain perception (Fig. 2i).
We then examined the effects of Caspase3-based ablation 22 of RtTg neurons on ASR. Specifically, a mixture of AAV-hSyn-Cre and AAV-flex-taCasp3-TEVp was bilaterally co-injected into the RtTg ( Supplementary Fig. 3a). Immunohistochemical analysis showed a remarkable reduction in the number of cells stained positive for neuronal nuclear antigen (NeuN + ) in the targeted RtTg region, confirming efficient ablation (Supplementary Fig. 3b-f). Compared with the control group, mice with RtTg lesions exhibited significantly decreased ASR amplitude (Supplementary Fig. 3g). In contrast, the RtTg ablation did not affect motor coordination (Supplementary Fig. 3h-m), arousal (Supplementary Fig. 3n), and pain perception ( Supplementary Fig. 3o). Together, these results suggest that the RtTg plays an essential role specifically in startle behaviours.
The activities of RtTg glutamatergic neurons augment during ASR. We next examined the expression of c-fos, a marker for recent neural activity 23 , in the RtTg of mice exposed to 95 dB acoustic stimuli. In addition to several brain nuclei with wellcharacterized functional involvement in the central auditory systemincluding the CN and the auditory cortex (AuC) 24 we also observed robust acoustic-stimuli-induced activation of RtTg neurons (Supplementary Fig. 4 and Fig. 3a, b). Pursuing this, we generated a clear map of the c-fos expression pattern of RtTg during ASR by collecting all consecutive coronal sections covering the entire RtTg area. Quantification analysis revealed a greater density of c-fos + cells in the rostral and medial parts of the RtTg (Fig. 3c and Supplementary Fig. 5). A non-acoustic stimulus such as air puff can also elicit startle reflexes in rodents 25 . However, we did not detect any changed c-fos expression in the RtTg of mice following air puff-induced startle reflexes (Supplementary Fig. 6).
To further examine whether RtTg neurons are selectively responsive to strong stimuli that drive the startle or just generally sensitive to auditory cues, we performed RtTg c-fos staining after mice were introduced to white noise stimuli with weaker intensities (45,50,55, 60 dB). These weak stimuli have been shown to be audible in mice 26 , but were insufficient to elicit startle reflex ( Supplementary Fig. 7). As shown in Supplementary  Fig. 8a, the c-fos expression level of RtTg under 45-60 dB noise stimuli was extremely low and had no significant correlation with these weak white noise intensities. In comparison, the RtTg total c-fos + cell number was well correlated with startle-eliciting white noise intensities ( Supplementary Fig. 8b) as well as startle amplitudes ( Supplementary Fig. 8c). These results demonstrate that the activity of RtTg indeed correlates with the startle response.
Screening via the Allen Brain Database indicated that RtTg neurons feature a high expression of vesicular glutamate transporter 2 (VGLUT2) mRNA, a classic marker for excitatory glutamatergic neurons. Consistently, our immunostaining results revealed that nearly 90% of c-fos + RtTg neurons co-labelled with glutamate ( Fig. 3d), confirming the excitatory nature of these ASR-activated RtTg neurons.
To further monitor the dynamics of RtTg glutamatergic neurons during ASR, we injected a genetically encoded Credependent calcium indicator (AAV-DIO-Ef1α-GCamp6m) into the RtTg of Vglut2-Cre mice 27 and performed fiber photometry 28 (Fig. 3e). Mice transfected with AAV-DIO-Ef1α-EGFP were used as controls. Fluorescent signal fluctuations were recorded simultaneously with ASR data. As with our initial observations based on c-fos experiments, there were significantly elevated calcium signals in RtTg glutamatergic neurons upon the onset of 95 dB stimuli ( Fig. 3f-i), whereas no signal changes were detected in control mice expressing EGFP (Supplementary Fig. 9 and Fig. 3j). The changes in calcium signals of RtTg glutamatergic neurons exhibited a positive correlation with the acoustic stimulus intensities ( Supplementary Fig. 10). In contrast, we did not detect Given that startle reflex occurs rapidly in response to abrupt aversive stimuli, we next performed in vivo recording of RtTg neurons during ASR to determine whether RtTg neurons also response with short latency compatible with the initiation of the startle reflex. Mice were introduced to white noise stimuli followed by sampling the RtTg neuronal firing rate of multiple units. RtTg neuronal activity increased significantly upon the onset of acoustic stimulus with a mean spike latency of 5.12 ± 0.47 ms (Fig. 3k). Furthermore, we observed a gradual increase in firing rates with elevated stimulus intensities ( Supplementary Fig. 12). These results indicate that RtTg neurons can quickly response to acoustic stimuli.
Chemogenetic inhibition of RtTg glutamatergic neurons suppresses ASR. We further examined whether chemogenetic inhibition of RtTg glutamatergic neurons could reduce ASR amplitudes. A Cre-dependent AAV expressing mCherry-tagged hM4Di was bilaterally injected into the RtTg of Vglut2-Cre mice ( Fig. 3l and Supplementary Fig. 13). In acute RtTg brain slices, bath application of CNO reduced the membrane potential as well as the firing frequency of hM4Di + RtTg neurons, confirming the efficacy of hM4Di-mediated inhibition ( Supplementary Fig. 14). Furthermore, i.p. injected CNO significantly decreased ASR amplitudes (Fig. 3m). Note that these effects cannot be directly attributed to CNO or its metabolites, as the mCherry control mice injected with CNO exhibited no such decrease in startle response (Fig. 3n). Taken together, these results indicate that the activity of RtTg glutamatergic neurons is capable of, and necessary for driving startle reflexes.
Chemogenetic inhibition of RtTg GABAergic neurons enhances ASR. According to the Allen Brain database, neurons in the central region of the RtTg are predominantly glutamatergic, whereas neurons expressing high levels of mRNA transcripts for inhibitory GABAergic neuron marker, vesicular GABA transporter (VGAT), were mainly distributed in the pericentral region of the RtTg ( Supplementary Fig. 15a). Consistently, in this study, immunostaining of the brain sections from GAD1-EGFP transgenic mice 29 further demonstrated the shell-distribution pattern of the inhibitory neurons embracing glutamatergic neurons in the central RtTg ( Supplementary Fig. 15b). Moreover, chemogenetic inhibition of the pericentral RtTg GABAergic neurons in Vgat-Cre mice resulted in significantly enhanced ASR amplitudes, suggesting a negative regulatory role of RtTg GABAergic neurons in startle reflexes ( Supplementary Fig. 15c, d). i Tail-flick latency of hM4Di-infected mice after i.p. injection of saline or CNO (n = 8 per group, P = 0.8269). Error bar represent mean ± s.e.m. Significance was assessed using two-sided unpaired t-tests in c,i, and two-way ANOVA combining with FDR corrections in b, e-h. * P < 0.05; ** P < 0.01; *** P < 0.001; ns, not significant. ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-26723-9 RtTg receives direct projections from CN. Given that the CN is the initial central auditory structure to receive input from the cochlear via the auditory nerve 30 , our discovery that the RtTg mediates ASR motivated us to examine if the RtTg is a downstream target of the CN. We conducted projection tracing experiments, wherein AAV-hSyn-EGFP 31 was unilaterally injected into the CN (Fig. 4a). In addition to brain regions such as the lateral superior olivary nuclei (LSO) and the inferior colliculus (IC) known to receive direct CN inputs 24,32 (Supplementary Fig. 16a-c), confocal images of serial coronal sections revealed dense EGFP + terminals in the RtTg ( Fig. 4b and Supplementary   Fig. 17). Indicating a strong trend for cross-hemispherical projection, we noted that the EGFP signal was much stronger in contralateral RtTg than ipsilateral RtTg (Fig. 4c).
To further confirm the existence of CN-RtTg projections, we unilaterally injected Retro-AAV-EGFP 35 into the RtTg (Fig. 4h, Supplementary Fig. 20). Abundant EGFP + cells were observed in the CN, with EGFP + cells number much higher in the contralateral CN than that in the ipsilateral CN ( Fig. 4i, j). Together, these results demonstrate that the CN directly innervates the RtTg.
The glutamatergic CN-RtTg projections mediate startle reflexes. To further investigate the nature of CN-RtTg neural projections, we first tested whether the CN innervate RtTg glutamatergic neurons, which has been proved to mediate startle reflexes. Anterograde transsynaptic tracer AAV2/1-hSyn-Cre was unilaterally injected into the CN of Ai9 mice (Fig. 5a), and consecutive RtTg coronal sections were collected for glutamate immunostaining. As expected, nearly 90% of tdT + RtTg neurons contain glutamate, suggesting that CN neurons project extensively to RtTg glutamatergic neurons ( Fig. 5b and Supplementary Fig. 21).
We next examined the type of neurotransmission between CN and RtTg by performing whole-cell patch-clamp recordings of the RtTg brain slices prepared from mice that received virallyencoded ChR2 injection into their CN (Fig. 5c). Photostimulation of CN ChR2 + terminals evoked excitatory postsynaptic currents (EPSCs) in most (6 of 7) RtTg neurons treated with Tetrodotoxin (TTX) and 4-aminopyridine (4-AP) 36 . Bath application of the glutamate receptor antagonist kynurenic acid (KYNA) completely blocked the light-evoked EPSCs (Fig. 5d). Collectively, these results demonstrate that CN neurons form monosynaptic glutamatergic synapses with RtTg neurons.
We then employed projection-based optogenetic-activation and chemogenetic-inactivation approaches to explore the functional contribution of this excitatory CN-RtTg innervation in freely moving mice. A Cre-dependent ChR2 or hM4Di virus was unilaterally/bilaterally injected into the CN of Vglut2-Cre mice, followed by implantation of an optical fiber or a guide cannula above the RtTg (Fig. 5e and Supplementary Fig. 22a, b). As expected, optogenetic activation of ChR2 terminals, but not EGFP terminals, in the RtTg elicited robust startle reflexes (Fig. 5f). On the contrary, chemogenetic inhibition of hM4Di terminals in the RtTg dramatically decreased ASR amplitudes ( Supplementary  Fig. 22c). These results further support that the CN-RtTg glutamatergic projections mediate ASR.
Spinal motor neurons are downstream targets of the RtTg. Given the above data indicating that optogenetic activation of the RtTg instantly evoked EMG activities in hindlimb muscles, we questioned whether the RtTg neurons specifically project to spinal motor neurons (MNs), as these neurons are known to mediate muscular movement. To define the segments of the spinal cord where RtTg neurons may terminate, we unilaterally injected the anterograde transsynaptic tracer AAV2/1-hSyn-Cre into the RtTg of Ai9 mice, followed by imaging of the spinal cord transverse sections from cervical to sacral segments (Fig. 6a). We observed that tdT + neurons are predominantly distributed in cervical and lumbar spinal cord segments, which have been previously demonstrated to harbor most of the neurons that innervate the limb muscles 37,38 . Immunostaining showed that these tdT + neurons were co-labelled with an MN-specific marker, choline acetyltransferase (ChAT) (Fig. 6b and Supplementary Fig. 23, 24).
Given previous studies reporting that some spinal interneurons also express ChAT 39 , to confirm that the RtTg indeed innervates muscle-controlling spinal MNs, we next performed rabies virusbased retrograde mono-transsynaptic tracing experiments 40 . Specifically, Retro-AAV encoding G protein and EGFP was injected into the hindlimb muscles of mice, followed by the injection of RV-ΔG-mCherry 3 weeks later (Fig. 6c). Using this approach we could specifically label the muscle-projecting spinal MNs. Starter MNs co-expressing EGFP and mCherry were found in the ventral horn of the spinal lumber segment (Fig. 6d). Furthermore, trans-synaptically labelled premotor neurons (expressing mCherry only) were abundantly found in the RtTg (Fig. 6d). These results indicate the existence of monosynaptic connections between RtTg and spinal muscle-controlling MNs.
To further examine whether the MN-projecting RtTg neurons also receive CN innervation, we unilaterally injected AAV2/1-hSyn-Cre into the CN and AAV-DIO-EGFP-T2A-Synaptophysin:mCherry 41 into the contralateral RtTg (Fig. 6e). After confirming that transsynaptic spreading of Cre successfully labelled the RtTg neuron cytosol (EGFP) (Fig. 6f), we observed that the mCherry + presynaptic terminals of these RtTg neurons closely contact ChAT + spinal MNs in the spinal cervical and lumbar segments (Fig. 6g). Thus, we demonstrated a brainstem reticulotegmental neural ensemble that connects the auditory system with the motor functions, which mediates startle reflexes.

Discussion
Although sporadic reports from human and animal studies have revealed involvement of RtTg in smooth-pursuit eye movement as well as limb movements [10][11][12][13][14] , the research about RtTg is still very limited. In this study, we demonstrate that the RtTg acts as a startle-controlling nucleus. This conclusion is strongly supported by sufficient data from the present study: (1) Optogenetic activation of RtTg neurons elicited robust startle reflexes; (2) Chemogenetic inactivation and viral ablation of RtTg neurons suppressed startle amplitudes; (3) RtTg neurons exhibited increased calcium activities and firing spikes during ASR; (4) Activation of the RtTg induced EMG activities of neck and hindlimb muscles within millisecond latency, consistent with the nature of startle reflexes, a quick alert to sudden aversive stimuli.
Numerous early studies demonstrated that the PnC nucleus might play a critical role in startle reflexes [17][18][19]42,43 . According to the mouse brain atlas 44 , the RtTg situates closely adjacent to the PnC. Therefore, in this study, when manipulating the RtTg such as Casp3-based lesion, optogenetic activation as well as chemogenetic inhibition, we always checked whether there was any collateral influence on the PnC neurons, and any data with crosscontamination of the PnC were eliminated from the analysis. In other words, the identification of functional roles of the RtTg in the startle reflex should not be artifacts caused by destruction or stimulation of the adjacent PnC nucleus. Thus, based on previous and our studies, both the PnC and the RtTg are critically involved in mediating the startle reflex. In fact, the phenomenon that two different brain nuclei mediate the same behaviour does exists 45,46 . It is worth noting that in most previous studies, PnC neurons were activated by strong acoustic stimulus (≥110 dB) [47][48][49] , while in the present study, we used relatively lower acoustic stimulus (75-95 dB) to trigger the startle reflex and also observed the activation of RtTg neurons. Thus, it is possible that the PnC and the RtTg may have different sensitivities to acoustic stimulus patterns, such as intensity, frequency, duration, and sound type. Further investigation needs to be done to explore these possibilities and the potential structural as well as a functional connection between the RtTg and the PnC. Although innate, startle reflexes can be modulated by fear emotional state, known as fear-potentiated startle (FPS) 50 . The abnormally exaggerated startle reflex is a common feature across different fear-associated neuropsychiatric disorders including posttraumatic stress disorder, panic disorder, and specific phobia [50][51][52] . To date, however, circuit mechanisms for FPS remain unclear, and the therapeutic treatments to relieve this symptom are obscure. Previous studies indicate that the amygdala-PnC neural projections may engage in the modulation of the startle reflex by fear emotions. As a newly identified startle-controlling nucleus, the RtTg may also receive projections from fear-encoding upstream nuclei and contribute to the development of FPS. For instance, in the present study, when performing retrograde viral tracing from the RtTg, we did observe several fear-related upstream nuclei, such as the dorsal raphe nucleus and the lateral habenular nucleus ( Supplementary  Fig. 20). Thus, dissection of the circuit basis of RtTg-mediated startle reflex should provide insight for further study of the mechanism underlying abnormal exaggerated startle symptoms observed in many psychiatric disorders.

Methods
Animals. All animal experiments were conducted following protocols approved by the Institutional Animal Care and Use Committee of the University of Science and Technology of China (USTC). C57BL/6 J, Vglut2-Cre, Vgat-Cre, and GAD1-EGFP mice were purchased from Beijing Vital River Laboratory or Jackson Laboratories. Mice were group-housed (3-5 animals per cage) in a stable environment (23-25°C ambient temperature and 50% humidity). Mice were maintained in a 12-hour light/dark cycle (light on at 7 a.m.) with free access to ad libitum chow food and water, and were allocated randomly to control and experimental groups. All experiments were performed on 8-week male mice to avoid potential sex as well as age differences. All experimental data were obtained by individuals blind to the experimental condition.
Acoustic startle reflex (ASR) test. After a 30 min period of acclimation to the testing room, mice were placed in a chamber within the darkened startle apparatus (MED Associate) for the ASR test. Broadband white noise were emitted from a high-frequency speaker mounted above the apparatus. A load cell platform under the chamber digitize the pressure generated by startled mice, and ASR amplitude was analyzed using Startle Reflex Software (MED Associate). After 5 min habituation period inside the startle chamber, mice received 30 startle trials (20-ms white noise stimuli with intensities of 85, 90, or 95 dB). Each of the three white noise stimuli was applied 10 times in random order at random inter-trial intervals (ITIs) ranging from 58-63 s. The resulting startle reflexes of the mice in the startle chamber were recorded during 600 ms after acoustic stimuli onset. ASR amplitude of each trial was defined as the largest peak-to-trough response that occurred within 200 ms after the onset of the startle stimulus. ASR amplitude of each stimulus intensity was calculated as the mean amplitude of 10 startle trials.
Open field test. After a 30 min period of acclimation to the testing room, mice were individually placed into the open field area (45 × 45 cm 2 ) and their spontaneous motor activities were measured for 10 min. The area was cleaned using 75% ethanol between each trial. Total distance together with time spent in the center zone were quantified automatically using ANY-maze system (Global Biotech Inc.).
Rotarod test. Before the training session started, mice were placed on the stationary rod for 5-min acclimation. Mice were trained three sessions per day for 2 days, with a 30 min inter-session interval. In each training session, mice were placed on the rotating rod accelerated from 4 rpm to 60 rpm in 300 s. Total time was recorded (XR1514, Xinruan) until the mice fell off from the rod or gripped and circled around two times. Mice staying more than 300 s were taken manually from the rod, with total time recorded as 300 s. After finishing the session, the apparatus was cleaned using 75% ethanol.
Tail-flick test. Tail-flick latency was measured using a tail-flick test apparatus (Ugo Basile). For each test, a radiant light beam was focused on a ventral portion of the mouse's tail (1.5-2 cm from its top). Light exposure was ended automatically once the mouse exhibit tail-flick motion. A 30 s cut-off time was employed to avoid skin injury. At the end of each test, the apparatus was cleaned with 75% ethanol alcohol.
Gait analysis. Gait analysis was conducted using the DigiGait imaging system (Mouse Specifics Inc.). Mice were placed individually on a transparent treadmill enclosed by a plexiglass compartment and have 3-5 min to freely explore the compartment while the treadmill remained still. After brief acclimation, the belt was started at an initial speed of 10 cm/s and gradually increased to a constant speed of 25 cm/s. Digital videos of paw movement were acquired using a camera mounted under the treadmill. For each mouse, approximately 5 s of video were selected based on continuous movement. Videos were then analyzed using DigiGait software to automatically calculate gait parameters including stride-related variables as well as stance width.
Pupil monitoring. Mice were surgically prepared for head-fixed monitoring of pupil diameter. After a 3-day recovery period, mice were placed individually in a custom-made restrainer and were aroused using air-puff stimuli generated by a nitrogen tank connected to a solenoid-operated piston pump at random ITIs ranging from 60-80 s. A small tube was placed behind the mice's head to deliver air puff. Pupil sizes were recorded using an infrared camera-based eye-tracking system (ISCAN systems, USA), filtered (CyberAmp, USA), and digitized (CED, UK) for offline analysis. %changes of pupil diameter during air puff were calculated using custom-made MATLAB routines (The MathWorks, Natick, MA, USA).
Stereotaxic injection. Mice were anesthetized with an intraperitoneal injection of pentobarbital sodium (0.5%, 0.01 ml/g) and mounted on a stereotaxic apparatus (68015, RWD Life Science). Ophthalmic ointment was applied to prevent the eyes from drying. The skull was exposed by a midline scalp incision, and a craniotomy was drilled unilaterally or bilaterally above each cannula or optical fiber implantation or injection site. Viruses were delivered at a rate of 50 nl/min using glass pipettes connected to a pump (Legato130, KD Scientific). After completion of the injection, the glass pipettes were held in place for 10 min before being withdrawn to allow for diffusion of viruses. The implanted cannula or optical fiber was stabilized to the skull with dental cement. The animals were allowed to recover from anesthesia on a heating pad and housed for 2-4 weeks after surgery. Coordinates used for RtTg injection were as followed: bregma −3.85 mm, lateral ± 0.47 mm, and dura −4.75 mm (the actual infected RtTg region corresponds to AP~−4.72 mm in the Mouse Brain Atlas, MBA). Coordinates used for CN injection were as followed: bregma −5.35 mm, lateral ± 2.45 mm, and dura −3.50 mm (the actual infected RtTg region corresponds to AP~−5.88 mm in the MBA).
Unless otherwise specified, mice were group-housed for 3 weeks after viruses injection before further conducting behaviour tests or/and immunostaining. For all experiments after surgery, the mice with incorrect virus injections and fiber/ cannula implantation sites were excluded from analysis.
C-fos detection. To perform c-fos staining of the startle reflex, mice were placed in a darkened startle chamber for the ASR test. After a 5 min habituation period, mice received 10 startle trials (different fixed white noise intensities or air puffs were used in specific experiments). Each of the trials was applied at random inter-trial intervals (ITIs) ranging from 58-63 s. Mice of the control group were placed in the chamber for 15 min without white noise stimulation. 1.5 h after being placed in the startle chamber, mice were sacrificed for c-fos immunostaining procedure.
Electromyography (EMG) recording. To perform EMG recording of neck and hindlimb muscles during optogenetic activation of RtTg neurons, AAV-hSyn-ChR2-EGFP or AAV-hSyn -EGFP were unilaterally injected into the RtTg, followed by optical fiber implantation. Three weeks later, mice were mildly anesthetized with isoflurane (1-3%), with the neck and hindlimbs were shaved off. The ends of the electrode wires were bared for 1-2 mm and threaded through surgical needles which were then inserted into the neck and hindlimb muscles. EMG signals were then amplified and filtered (Biotex Kyoto, Japan), digitized (Power 1401), and recorded (Spike2 Software) at a sampling rate of 1000-Hz. The Spike2 data were further analyzed using custom-made MATLAB routines.
In vivo electrophysiological recordings. Mice were head-fixed in a stereotaxic frame in which the hollow ear bars were coupled to a white noise delivery system. After craniotomy and brain surface exposure, the linear silicon electrode (NeuroNexus, USA) was inserted into the RtTg to sample the neuronal firing rate of multiple units elicited by acoustic stimuli. Mice received startle trials with varied noise intensities (60, 75, 80, 85, 90, and 95 dB). Each of the white noise stimuli was applied in random order at random inter-trial intervals (ITIs) ranging from 58-63 s. Signals were amplified, digitized (Blackrock, USA), and recorded (Nex Technologies, USA) at a sampling rate of 30 kHz. RtTg neuronal firing rate as well as firing latency after the onset of white noise were analyzed using NeuroExplore software (Plexon Inc.).
In vivo fiber photometry. Mice were housed individually for two weeks after AAV injection and optical fiber (200 μm diameter, 0.33 numerical aperture (NA), AniLab) implantation. Fluorescence signals were output to an amplifier (C7319, Hamamatsu) with a 0.01 s moving average filter (MF525-39, Thorlabs). The signals were then digitized (Power 1401) and recorded (Spike2 Software) at a 300 Hz sampling rate. The laser power at the tip of optical fiber was 15 μW for RtTg. For monitoring the neuronal activity of RtTg during startle, we recorded white noise stimuli as well as airpuffs induced startle reflexes and Ca 2+ transients or EGFP signals simultaneously. The raw data was then analyzed using MATLAB. Ca 2+ fluorescence changes (ΔF/F) during startle reflex and conditioned fear were aligned to individual bouts and were calculated as (F−F 0 )/F 0 , where F 0 was the average fluorescence signals from 1 s preceding stimuli (95 dB stimuli or the CS) to stimuli onset. (ΔF/F) values were presented as heatmaps or plotted with shaded area indicating s.e.m.
In vivo optogenetic manipulation. Mice were housed individually for three weeks after AAV injection and optical fiber (200 μm diameter, 0.33 NA, AniLab) implantation. each optical fiber was attached to a 473-nm laser generator (AniLab-Opto), and light pulses were delivered using Master-8 (A.M.P.I.). For in vivo optogenetic activation of RtTg neurons in general or RtTg glutamatergic neurons, mice were placed in a startle chamber and received blue light stimulation (473-nm, 15 ms pulse, 5 mW, 10 mW, 15 mW, and 20 mW). For activation of CN-RtTg projection, Vglut2-Cre mice were placed in the startle chamber and received blue light stimulation (473 nm, 15 ms pulse, 15 mW). Each mouse received 10 stimuli in total with an inter-trial interval of 30 s, and opto-evoked startle amplitude was calculated as the mean amplitude of 10 stimuli trials.
To determine the excitability of target neurons, whole-cell current clamp mode (I = 0 pA) was obtained with pipettes (5-7 MΩ) filled with internal solution which contained K-gluconate 145 mM, MgCl 2 4 mM, HEPES 10 mM, EGTA 10 mM, Mg-ATP 5 mM, and Na-GTP 0.3 mM (pH 7.2, 285-290 mOsm). Depolarizing current step (1 s, −20 to 140 pA) was injected into the target neurons. CNO was added to the HEPES aCSF (continuously bubbled with 5% CO 2 and 95% O 2 ) for at least 10 min to act on hM3Dq + /hM4Di + cell bodies or terminals. For membrane potential changes recording, CNO was bath applied for 10 min after 5 min stable baseline.
The electrophysiological signals were recorded via a MultiClamp 700B amplifier (Molecular Devices) equipped with Digidata 1550 A (Axon Instruments), digitized at 10 kHz, and filtered at 3 kHz. Data with a series resistance >25 MΩ or changed by >20% were discarded before analyzed using Clampfit 10.3 (Molecular devices, USA) and Minianalysis (Synaptosoft).
For chemogenetic inhibition of CN-RtTg projections, mice were housed individually for 1-4 weeks after AAV injection or/and cannula implantation. Internal injectors attached to 10 μL syringes were inserted into guide cannula (RWD) for local CNO (0.2 μg/μl, 500 nl/side, Sigma) infusion at a rate of 200 nl/ min using an infusion pump (KDS Scientific). After the infusion, injectors were hold still for 1 min to allow for full drug diffusion into the target region. Behaviour tests were conducted 15 min after CNO infusion.
Statistics and reproducibility. Statistical analysis was performed using MATLAB (MathWorks) and GraphPad Prism 8.0 (GraphPad Software). Comparisons between control and treatment groups were conducted using paired/unpaired t-tests or two-way ANOVA combined with FDR corrections, as indicated in the figure legends. Data are presented as mean ± s.e.m, with significant difference defined as P < 0.05. No statistical methods were used to predetermine sample sizes, but our sample sizes are similar to those reported in previous publications. Unless otherwise noted, all representative experiments were repeated at least three times independently with similar results.
Reporting summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
The in situ hybridization database from Allen Mouse Brain Atlas (https://mouse.brainmap.org/experiment/show/73818754) (https://mouse.brain-map.org/experiment/show/ 72081554) were used. The data that support the findings of this study are provided in the article and its Supplementary information files, and are available from the corresponding author upon reasonable request. Source data are provided with this paper.

Code availability
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/ licenses/by/4.0/.