Direct auditory cortical input to the lateral periaqueductal gray controls sound-driven defensive behavior

Threatening sounds can elicit a series of defensive behavioral reactions in animals for survival, but the underlying neural substrates are not fully understood. Here, we demonstrate a previously unexplored neural pathway in mice that projects directly from the auditory cortex (ACx) to the lateral periaqueductal gray (lPAG) and controls noise-evoked defensive behaviors. Electrophysiological recordings showed that the lPAG could be excited by a loud noise that induced an escape-like behavior. Trans-synaptic viral tracing showed that a great number of glutamatergic neurons, rather than GABAergic neurons, in the lPAG were directly innervated by those in layer V of the ACx. Activation of this pathway by optogenetic manipulations produced a behavior in mice that mimicked the noise-evoked escape, whereas inhibition of the pathway reduced this behavior. Therefore, our newly identified descending pathway is a novel neural substrate for noise-evoked escape and is involved in controlling the threat-related behavior.


Introduction
Mammals have evolved several strategies to deal with a dangerous situation that rely on behavioral responses such as freezing, escaping, and fighting [1,2], depending on the distance from where threat stimuli occur [3]. Sound is one of the natural threatening stimuli that elicit defensive behaviors. Different auditory stimuli elicit different behaviors, which may be harbored in distinct neural substrates [4]. For example, the superior colliculus (SC)-dorsolateral periaqueductal gray (dlPAG) circuit is recruited in frequency upsweeps-elicited escape [5], whereas the

Involvement of Glu lPAG neurons in noise-evoked escape behavior
Previous work has shown that mice exhibit defensive behaviors in response to a loud sound [22,24]. We employed a similar paradigm to detect sound-evoked defensive responses. As shown in Fig 1A, the mice were allowed to habituate to an environment consisting of two chambers. Then, a sound was suddenly delivered in the chamber where the mouse was located. Following a loud white noise stimulus (80-dB sound pressure level [SPL], 5-s duration), the mouse immediately escaped toward the opposite chamber ( Fig 1A-1C and S1 Video). To confirm this result, we trained head-fixed mice to freely run above a rotatable plate. Running speeds were measured with a rotatory encoder and recorded by a computer (Fig 1D). In response to the noise, the mice displayed immediate running (Fig 1E and 1F, S2 Video). In contrast, a low-level noise  with the same duration failed to induce escape behavior during the first trial ( Fig 1E, and S1A  Fig). The 80-dB SPL noise-evoked escape displayed clear adaptation from the fifth trial of sound stimulation (S1B-S1D Fig). Because the 80-dB SPL noise could reliably induce escape behavior without impairing hearing sensitivity [25], this stimulus was used for the rest of the study. To reveal specific brain regions involved in the noise-elicited escape behavior, we assessed expression of the c-Fos protein in the brain following the noise stimulation [26,27]. Compared with control mice that were not exposed to the noise, mice subjected to the noise showed massive c-Fos expression in the limbic and auditory-related areas, including the ACx and lPAG (Fig 1G and 1H, S2 Fig). Furthermore, we performed extracellular recording in freely moving mice [28] and found that the noise at 80-dB SPL, rather than at 40-dB SPL, strongly increased neuronal firing rates in the lPAG (Fig 1I and 1J). In addition, we observed an adaption of noise-evoked neuronal firings in lPAG neurons but not in ACx neurons (S3 Fig). These results suggest that the excitability of lPAG neurons is required for the noise-evoked escape behavior.
Glu lPAG neurons have been implicated in promoting flight behavior [9] and were activated by the escape-evoking noise. To test the role of Glu lPAG neurons in noise-driven escape no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

Competing interests:
The authors have declared that no competing interests exist.

Activation of the Glu ACx !Glu lPAG pathway evoked an escape behavior
Given increased excitability of lPAG neurons in the presence of noise, we investigated whether activation of the Glu ACx !Glu lPAG pathway produces escape behaviors. As expected, after optical activation of ChR2-containing Glu ACx terminals (5-8 mW) in the lPAG (Fig 3A and 3B), mice displayed a series of defensive behaviors, including escaping toward and spending more time in the opposite chamber, running on the rotatable plate, and wall rearing (Fig 3C-3H, S4 and S5 Videos). None of these behaviors was observed in the control group. In addition, escape behavior evoked by optical activation of the Glu ACx !Glu lPAG pathway showed an adaptation similar to that evoked by noise (S8A and S8B Fig). Interestingly, co-application of subthreshold noise (40-50-dB SPL) and subthreshold light (4-5 mW) caused the mice to escape but failed to significantly change behavior when presented alone (S8C Fig). These results indicate that activation of the Glu ACx !Glu lPAG pathway is sufficient to drive noise-evoked defensive behavior.
https://doi.org/10.1371/journal.pbio.3000417.g002 noise-evoked running for head-fixed mice (Fig 4I and 4J). The light stimuli did not change these behavioral responses at a statistical level in the control group (Fig 4G, 4H, and 4J). In addition, we found that the locomotion during light stimulation was not changed in the openfield test (S9 Fig).
We also used the chemogenetic method to silence ACx axon terminals in the lPAG by ACx infusion of AAV-DIO-human Gi-coupled M4 muscarinic receptor (hM4Di)-mCherry and lPAG infusion of clozapine-N-oxide (CNO). We found that light-evoked EPSCs in lPAG neurons through optically stimulating ChR2-and hM4Di-expressing ACx fibers were abolished by bath-applied CNO (Fig 4K-4M). Similar noise-evoked defensive behaviors were observed after chemogenetic inhibition of the ACx!lPAG pathway in CaMKII-Cre mice (Fig 4N-4Q). These results indicate that the Glu ACx !Glu lPAG pathway is at least one of the underlying circuits that govern noise-evoked defensive behavior.

Action of Glu ACx !Glu lPAG pathway as a defense circuitry
Previous work has shown that the ICx receives a descending projection from the ACx [22,29]. Thus, it is possible that action potentials can back-propagate to cell bodies of lPAG-projecting ACx neurons to activate ICx neurons when ACx!lPAG projection fibers are excited [30]. If this is the case, then behavioral effects following activation of the Glu ACx !Glu lPAG pathway may result from sequential activation of the ICx!dlPAG pathway. To address this issue, we first co-injected respectively CTB-488 and CTB-555 into the lPAG and ICx to outline the  ACx!lPAG and ACx!ICx pathways (Fig 5A and 5B). Seven days after the injection, we observed a minority of ACx neurons (37.16% ± 3.50%) that send collateral axons to both the lPAG and ICx (Fig 5C and 5D). These results are consistent with previous studies [22,24,31,32].
To determine whether behavioral responses evoked by activation of the ACx!lPAG pathway depend on the ICx, we used Cre-dependent expression of AAV-DIO-hM4Di-mCherry in the ICx and intraperitoneal injection of CNO to selectively inhibit ICx glutamatergic neurons in CaMKII-Cre mice (Fig 5E-5H). This is based on our finding that most retrogradely traced neurons from the lPAG were positive for glutamate in the ICx (S10A-S10C Fig). Three weeks after injection of AAV-DIO-hM4Di-mCherry into the ICx and that of AAV-DIO-ChR2-eYFP into the ACx, behavioral testing was conducted. We found that 50 minutes after CNO injection (3 mg/kg), optical activation of ChR2-containing Glu ACx terminals in the lPAG still evoked escaping, running, and wall rearing (Fig 5I-5N). The role of ACx!lPAG projection in eliciting defensive reactions was corroborated in mice with ICx inactivation by AAV-human synapsin (hSyn)-hM4Di-mCherry (S11 Fig). In addition, after injection of AAV-eNpHR-eYFP into the ICx, we found most eNpHR-containing fibers in the dlPAG, rather than in the lPAG (S10D-S10F Fig). Optical inhibition of ICx terminals in the lPAG did not affect noiseevoked escape (S10G-S10I Fig).
To characterize the collateral pathway from lPAG-projecting ACx neurons, we employed a combinational viral strategy by lPAG infusion of AAV-Retro-Cre and ACx infusion of Credependent AAV-DIO-ChR2, respectively (Fig 6A and 6B). The Cre recombinase-containing neurons were visible in the ACx and mCherry-containing projection fibers were observable in the lPAG, the inferior colliculus (IC), the medial SC (mSC), and the amygdala (Fig 6C). To rule out the involvement of these collateral pathways during optical stimulations, the ACx was silenced with muscimol before behavioral testing ( Fig 6D). We found that the escape and running behaviors were reliably evoked by optical activation of ACx terminals in the lPAG after ACx inactivation (Fig 6E-6H). These results indicate that the Glu ACx !Glu lPAG pathway functions as a defense circuit independent of the collateral pathway.

Escape mediated by the mSC!dlPAG pathway
It has been reported that the mSC conveying cortical inputs to the dlPAG also mediates frequency upsweeps-evoked escape [5,24]. Our anterograde tracing experiment has shown that there were ACx fibers in the mSC (Fig 6C). In order to understand the role of this alternative pathway, we compared the escape behavior evoked by the optically activated ACx!lPAG pathway with that evoked by the optically activated mSC!dlPAG pathway. After infusion of AAV-DIO-ChR2-mCherry into the mSC, the fibers containing mCherry were visible in the dlPAG (Fig 7A-7C). The AAV-ChR2 protocol was verified by successful recordings of light-evoked action potentials in mSC neurons and EPSCs in dlPAG neurons (Fig 7D-7G). Upon optical activation of these terminals, the mice exhibited defensive behaviors, including escape toward the opposite chamber and running on the turntable (Fig 7H-7K). The running evoked by optical activation of the mSC!dlPAG pathway was faster in speed (Fig 7I and 7J) and shorter in the peak latency (Fig 7I and 7K) than that evoked by optical activation of the ACx!lPAG pathway. the speed of noise-evoked running in the absence or presence of CNO. The underlying data for this figure can be found in S1 Data. Values are means ± SEM ( �� P < 0.01; ��� P < 0.001). Two-way ANOVA with Bonferroni post hoc analysis for (G), (H), (J), (O), (P), and (Q). AAV, adeno-associated virus; ACSF, artificial cerebrospinal fluid; ACx, auditory cortex; Aq, aqueduct; CaMKII, Ca 2+ /calmodulin-dependent protein kinase II; ChR2, channelrhodopsin-2; CNO, clozapine-N-oxide; Cre, cyclization recombination; DIO, double-floxed inverted orientation; eNpHR, enhanced natronomonas pharaonis halorhodopsin; EPSC, excitatory postsynaptic current; eYFP, enhanced yellow fluorescent protein; Glu, glutamatergic; hM4Di, human Gi-coupled M4 muscarinic receptor; lPAG, lateral periaqueductal gray. https://doi.org/10.1371/journal.pbio.3000417.g004 In addition, light-evoked EPSCs on dlPAG neurons by optical activation of mSC terminals were more pronounced than those associated with lPAG neurons subjected to optical activation of ACx terminals (Fig 7G and Fig 2P). We then examined whether the responses of these two circuits are dependent on the nature of auditory stimuli. Our results showed that escape behaviors evoked by noise were reduced when mSC!dlPAG projections were optically inhibited (S12C Fig). Similarly, escape behaviors evoked by frequency upsweeps were reduced when ACx!l-PAG projections were optically inhibited (S12D Fig). These results indicate that the two circuits differentially contribute to the auditory-related defensive behaviors.

Discussion
In this study, we have discovered a previously unexplored pathway for controlling noiseevoked escape behaviors that is cell specific and directly descending from the ACx to the lPAG (the Glu ACx !Glu lPAG pathway). The robust supporting evidence for the existence of this pathway came from our viral tracing experiment and optogenetic experiment. Specifically, Transsynaptic viral tracing revealed that a great number of glutamatergic neurons, rather than GABAergic neurons, in the lPAG are directly innervated by those in layer V of the ACx (Fig  2). Activation of this pathway by optogenetic manipulation mimicked the noise-evoked escape, whereas inhibition of the pathway reduced the escape (Fig 3 and Fig 4). Therefore, we have successfully identified a new pathway that is an important neural substrate for noise-evoked escape and is among the multiple neural circuits controlling threat-related behavior. Fig 8  shows proposed neural networks, including our newly identified one, that are involved in noise-evoked defensive behavior.
The PAG receives inputs from multiple regions, such as the amygdala, the hypothalamus, and the prefrontal cortex, which are involved in regulation of defensive behaviors [9,23,33,34]. However, our knowledge of the precise cell type-specific projections is limited. In this study, we found that glutamatergic neurons in the layer V of the ACx preferentially innervate lPAG glutamatergic neurons. This finding is interesting because glutamatergic neurons of both ACx and lPAG are flight promoting in auditory defensive behaviors [9,22].
https://doi.org/10.1371/journal.pbio.3000417.g007 vlPAG [9], whereas sound-evoked escape largely relies on the dl/lPAG [22]. Our cell-specific viral tracing showed that layer V Glu ACx neurons preferentially project to the lPAG but not the vlPAG. We also observed an adaptation of noise-evoked firings in the lPAG neurons but not in the ACx neurons, which might account for the adaption of noise-evoked escape behavior. These findings suggest a subregion-specific functional role of the lPAG in sound-driven defensive behaviors. Moreover, our extracellular recordings showed that noise stimulation strongly increased the neuronal firing rate in the lPAG. Given the spatial proximity of the lPAG to the dlPAG, it is necessary to discriminate lPAG and dlPAG neurons by the optotagging approach based on the input specificity in the future study.
Because corticofugal projections have been found to drive defensive behaviors [22,24], this raises the question of the nature of direct auditory cortical inputs to the lPAG, as well as their role in producing defensive behavior. Approximately 40% of ACx neurons send axon collateral inputs to both lPAG and ICx, resembling the projection pattern of the ventromedial hypothalamus to both the anterior hypothalamic nucleus and PAG [33]. In this study, we found that afferent inputs on PAG neurons differ greatly between these two pathways. ACx descending inputs preferentially target glutamatergic neurons, whereas ICx projecting fibers contact both glutamatergic and GABAergic neurons. Notably, optical activation of the Glu ACx !Glu lPAG pathway elicited defensive behavior when the ICx was deactivated. These findings suggest that the newly identified Glu ACx !Glu lPAG pathway controls noise-evoked escape by bypassing the ICx.
The mSC receives descending cortical inputs, and the mSC!dlPAG pathway contributes to sound-driven escape [5,24]. Whether the ACx!lPAG pathway functions differentially from the mSC!dlPAG pathway to trigger sound-elicited defensive behaviors requires further investigation. Nevertheless, we found that the running evoked by optical activation of the mSC!dlPAG pathway was faster in speed and shorter in the peak latency than that evoked by optical activation of the ACx!lPAG pathway. In addition, light-evoked EPSCs on dlPAG neurons by optical activation of mSC terminals were more pronounced than those associated with lPAG neurons subjected to optical activation of ACx terminals. These suggest the two circuits differentially contribute to the auditory-related defensive behaviors.
When acoustic information reaches the ACx, corticofugal projections send divergent inputs to lPAG neurons, to dlPAG-projecting ICx neurons, and to dlPAG-projecting mSC neurons (Fig 8). The activation of the PAG by monosynaptic ACx!lPAG projections may precede that by ACx!ICx (mSC)!dlPAG projections, suggesting that ACx!lPAG projections might rapidly mobilize the PAG, whereas the mSC and the ICx could amplify descending cortical controls [5,24].
Some fear conditioning studies showed that the ACx is indispensable for complex soundcued fear but not for tone-cued fear [38][39][40]. Our present study also indicates an indispensable role of the ACx in the escape response to a white noise, but we are not sure whether the ACx is required for an escape response when a pure tone is used. It should be noted that there exist at least three major discrepancies between our behavioral paradigms and those in fear conditioning. (1) The innately elicited behaviors in the present study do not require, but learned fear behaviors do require, a training stage of associating a conditioned stimulus and unconditioned stimulus [41]. (2) The neural circuits recruited by these behaviors might also be quite different [2,4]. (3) We used a white noise for innately elicited escape, whereas others used complex sound, such as frequency-modulated sweep sound, for threat conditioning. Given those discrepancies, it is hard to predict how important the role of the ACx is in the escape behavior elicited by a complex sound. Further study is required to elucidate this issue.
The ACx evaluates the nature of threatening acoustic signals [22]. The cognitive control served by cortical processing could generate flexible defense behavior that is unlike the stereotyped behavior served by the subcortical circuit [42][43][44][45]. It has been reported that the ACx [46,47] and prefrontal cortex-zona incerta-lPAG circuit [20] may account for the extinction of learned fear. In this way, the neural network of the ACx might contribute to multiple auditory threat-related behaviors, such as noise-evoked escape behavior and its adaptation in the current study.

Ethics statement
All animal protocols were approved by the Animal Care and Use Committee of the University of Science and Technology of China (USTCACUC1402021) and in accordance with the National Institutes of Health Guide for the Care and Use of Laboratory Animals.

Animals
In all experiments, C57BL/6J, CaMKII-Cre, Gad2-Cre, and Ai14 (RCL tdT) male mice (purchased from Charles River or Jackson Laboratories) at 8-10 weeks of age were used. Until the cannula surgery, the mice were housed five per cage in a colony with ad libitum access to water and food (standard mouse chow). They were maintained under a 12-hour light/dark cycle (lights on from 7:00 AM to 7:00 PM) at a stable temperature (23-25˚C).

Virus injection
Prior to surgery, the mice were fixed in a stereotactic frame (RWD, Shenzhen, China) under a combination of xylazine (10 mg/kg) anesthesia and ketamine (100 mg/kg) analgesia. A heating pad was used to maintain the core body temperature of the animals at 36˚C. A volume of 100-300 nL virus (depending on the expression strength and viral titer) was injected using calibrated glass microelectrodes connected to an infusion pump (micro 4, WPI, Sarasota, FL) at a rate of 30 nL/minute. The coordinates were defined as dorsoventral (DV) from the brain surface, anterior-posterior (AP) from bregma, and mediolateral (ML) from the midline (in mm) [48,49].

Optogenetic manipulations in vivo
An optical fiber was initially implanted into the lPAG, in the brain of an anesthetized mouse that had been immobilized in a stereotaxic apparatus. The implant was secured to the animal's skull with dental cement. Chronically implantable fibers (diameter, 200 μm, Newdoon, Hangzhou) were connected to a laser generator using optic fiber sleeves. The delivery of blue light (473 nm, 5-8 mW, 20 Hz, 10-millisecond pulses) or yellow light (594 nm, 5-8 mW, constant) was controlled by a Master-8 pulse stimulator (A.M.P.I., Jerusalem, Israel). The same stimulus protocol was applied to the mice in the control group. The mice were allowed at least 10 days for recovery before injections to minimize stress during the behavioral assays. The location of the fibers was examined in all mice at the conclusion of the experiments, and data obtained from mice in which the fibers were located outside of the desired brain region were discarded. Behavioral assays were performed immediately after light stimulation.

Local drug infusion
An internal stainless steel injector attached to a 10-μL syringe (Hamilton, Reno, NV) and an infusion pump was inserted into the guide cannula (I.D. 0.34 mm, RWD, Shenzhen, China) and used to infuse muscimol (0.2 μL, 0.5 mg/mL) into the left ACx or mSC [10] absent of virus injection or CNO (0.1 μL, 1 mg/mL) into the right lPAG at a flow rate of 100 nL per minute. The injector was slowly withdrawn 2 minutes after the infusion, and the behavioral assays were performed approximately 30 minutes after the infusion.

In vivo electrophysiological recording
Animals were prepared for surgery as described above [28,51]. For chronic extracellular recordings (Fig 1I and 1J, S3 Fig), a custom-made four movable tetrode array was implanted into the lPAG (AP, 4.65 mm; ML, 0.6 mm; DV, 1.3 mm) and the ACx (AP, 2.45 mm; ML, 3.8 mm; DV, 0.9 mm). Each tetrode was made of four twisted fine platinum/iridium wires (12.5μm diameter, California Fine Wire, Grover Beach, CA). The screw-based microdrive scaffolds for lowering the electrodes were cemented onto the skull. The mice were allowed to recover for at least 3 days before recordings were made. The recording sites were verified by passing an electrical current (20 μA, 15-20 seconds) to lesion the brain tissue at the end of all experiments. For head-fixed recordings (S11E and S11F Fig), a screw for head fixation was cemented on top of the skull. An array of two electrodes, one as recording electrode (approximately 1.0 MΩ, FHC, Bowdoin, ME) and the other tip-stripped electrode as reference, were positioned with a stepping-motor microdriver. Auditory stimuli were generated digitally using a computer-controlled Auditory Workstation from Tucker-Davis Technologies (TDT, Alachua, FL) and delivered through an open-field magnetic speaker (MF1, TDT) with an interval of 30 seconds. SPL was calibrated with a condenser microphone (Center Technology, Taiwan). Recording electrodes were attached to a 16-channel headstage, and neuronal signals were amplified, filtered at a bandwidth of 300-5,000 Hz, and stored using TDT software (OpenEX, TDT). Spike sorting was performed with a sorting method involving a T-Dis E-M algorithm built in Offline Sorter 4 (Plexon, USA). The firing rates of sorted units were calculated using Neuroexplorer 5 (Nex Technologies, USA). Peristimulus histograms (PSTHs) of firing rates were computed over a bin width of 10 milliseconds for each unit between −1 and 6 seconds, and in this time window the mean and SD of firing rates across all bins were calculated. The units with firing rates during noise stimulation between 99% confidence interval (mean ± 2.576 SD) are classified as not responsive ones, and those higher or lower than the confidence interval are sound-promoting or sound-inhibiting ones, respectively.
Whole-cell patch-clamp recordings. Neurons in the slice were visualized using a 40× water-immersion objective on an upright microscope (BX51WI, Olympus, Japan) equipped with interference contrast (IR/DIC) and an infrared camera connected to the video monitor. Whole-cell patch-clamp recordings were obtained from visually identified ACx layer V, ICx, mSC, or PAG cells. Patch pipettes (3-5 MO) were pulled from borosilicate glass capillaries (VitalSense Scientific Instruments Co., Wuhan, China) with an outer diameter of 1.5 mm on a four-stage horizontal puller (P1000, Sutter Instruments, Novato, CA) and filled with intracellular solution that contained (in mM) 130 K-gluconate, 2 MgCl 2 , 5 KCl, 0.6 EGTA, 10 HEPES, 2 Mg-ATP, 0.3 Na-GTP (pH, 7.3-7.2; osmolarity, 285-290 mOsm/kg). The neurons were held at −70 mV in voltage-clamp mode to record the membrane currents and at 0 pA in current-clamp mode to record the membrane voltages. The signals were acquired via a Multiclamp 700B amplifier (Molecular Devices, Sunnyvale, CA), low-pass filtered at 2.8 kHz, digitized at 10 kHz, and analyzed with Clampfit 10.7 software (Molecular Devices). If the series resistance changed more than 20% during the recording, the experimental recording was immediately terminated.
Light-evoked responses. Optical stimulation was delivered using a laser (Shanghai Fiblaser Technology Co., China) through an optical fiber 200 μm in diameter positioned 0.2 mm from the surface of the brain slice. To test the functional characteristics of AAV-DIO-ChR2, fluorescently labeled neurons expressing ChR2 in CaMKII-Cre mice 3-4 weeks after virus injection were visualized and stimulated with a blue light (473 nm, 5-10 mW) using 5-Hz, 10-Hz, or 20-Hz stimulation protocols with a pulse width of 10 milliseconds. Similarly, the function of AAV-DIO-eNpHR3.0 was assessed in fluorescently labeled neurons expressing eNpHR by applying sustained yellow light stimulation (594 nm, 5-10 mW, 200 milliseconds). For electrophysiological recording of monosynaptic postsynaptic currents, 1 μM tetrodotoxin (TTX) and 1 mM 4-aminopyridine (4-AP) were added to the bath solution to eliminate the polysynaptic components, and blue light (473 nm, 10-millisecond pulse) was delivered to the lPAG or dlPAG of CaMKII-Cre mice in which the ACx or mSC had been injected with AAV--DIO-ChR2-mCherry. DNQX (10 μM) was used to block glutamate receptors. Unless otherwise stated, all drugs were purchased from Sigma-Aldrich (St. Louis, MO). TTX was obtained from Hebei Aquatic Science and Technology Development Company, China.

Immunohistochemistry
The mice were deeply anesthetized with pentobarbital sodium (50 mg/kg, i.p.) and sequentially perfused with saline and 4% (w/v) PFA. The brains were subsequently removed and postfixed in 4% PFA at 4˚C overnight. After cryoprotection of the brains with 30% (w/v) sucrose, coronal sections (40 μm) were cut on a cryostat (Leica CM1860, Germany) and used for immunofluorescence. The sections were incubated in 0.3% (v/v) Triton X-100 for 0.5 hour, blocked with 10% donkey serum for 1 hour at room temperature, and incubated with primary antibodies, including anti-c-Fos (1:500, rabbit, Santa Cruz Biotechnology, Dallas, TX), anti-glutamate (1:500, rabbit, Sigma-Aldrich), and anti-GABA (1:500, rabbit, Sigma Aldrich) at 4˚C for 24 hours, followed by the corresponding fluorophore conjugated secondary antibodies for 2 hours at room temperature. Fluorescence signals were visualized using Leica DM2500 and Zeiss LSM710 microscopes and analyzed using ImageJ 1.4 (NIH). For counting immunoreactive cells, the 8-bit grayscale image was background subtracted before applying a threshold to all images. The threshold was adjusted within 10% of the average intensity, and cells at or above the threshold are considered immunopositive.

Behaviors
All behavioral tests were conducted within a soundproof chamber, and mice were habituated for 3 days prior to testing. Auditory stimuli of white noise or frequency upsweeps (frequencymodulated upsweep from 17 to 20 kHz over 3 seconds) were generated through a RZ6 Multi I/ O Processor (TDT, Alachua, FL) and delivered by MF1 open-field magnetic speakers. SPL was calibrated carefully. During each testing session, behavior was recorded using an infrared camera. Blue light was generally delivered at 20 Hz for 10-20 seconds with the exception of 5 minutes for testing wall rearing. The duration of the yellow light was identical to that of sound. The experimental area was cleaned with 75% ethanol after each test to remove olfactory cues from the apparatus. To avoid behavioral adaptation, the mouse for the optogenetic experiment (Fig 3, Fig 5, Fig 6, and Fig 7) was not exposed to sound unless the synergetic effect of light and noise on escape behavior was evaluated (S8 Fig). Escape behavior test. Mice were placed in a behavioral box (40 × 25 × 25 cm) consisting of two chambers and a middle plate. Two speakers were placed on the walls of each chamber. Each mouse was allowed to freely explore the surroundings and cross the opening in the plate toward the opposite side. Sound was delivered from the chamber where the mouse was located. Time spent in the opposite chamber was determined within 30 seconds from the end of sound or light delivery. The probability and time spent in the opposite chamber for each mouse were averaged across five trials.
Running behavior test for head-fixed mice. Mice with implanted optical fibers were used in this set of experiments. Each mouse was clamped to a fixing bar on the optic fiber sleeve and allowed to adapt to head fixation. Then, the mouse could run freely on a Plexiglass circular plate (diameter, 30 cm) that was connected to a rotatory encoder used to record running speed. The data were digitized and stored on the computer for offline analysis. Sounds were delivered from a speaker placed 10 cm from the ear of each head-fixed mouse.
Wall rearing behavior test. Each mouse was placed in a single box (50 × 50 × 60 cm) to observe blue light-evoked wall rearing behavior. Offline inspection of the video was performed to determine the number and total duration of wall rearing within a time window of 5 minutes before and during light stimulation.
Open-field test. Mice were placed in one corner of an open-field apparatus that consisted of a square area (25 × 25 cm) and a marginal area (50 × 50 × 60 cm); the mice were allowed to freely explore their surroundings. The animals' movement trajectories were recorded for 5 minutes using EthoVision XT software 8.5 (Noldus Information Technology).