Research Article

Insular cortex processes aversive somatosensory information and is crucial for threat learning

See allHide authors and affiliations

Science  31 May 2019:
Vol. 364, Issue 6443, eaaw0474
DOI: 10.1126/science.aaw0474

This article has been retracted. Please see:

Fear behavior and the insular cortex

Threat learning is important to avoid dangers in the environment. The amygdala is a brain structure involved in the formation of threat memories. How unconditioned stimulus information reaches the amygdala remains unclear. Berret et al. found that independent subpopulations of neurons in the insular cortex project either to the lateral or the central amygdala and drive threat learning and fear-associated responses, respectively. Neurons in the posterior insula adapted their response patterns to conditioned stimulus presentation during fear conditioning and maintained this pattern at retrieval. Multisensory integration in the insula thus contributes to the storage and retrieval of threat memories.

Science, this issue p. eaaw0474

Structured Abstract


Animals and humans need to learn about potential dangers in the environment to guarantee survival. This threat learning can be studied in laboratory animals with fear conditioning paradigms, in which an innocuous sensory event such as a tone [the conditioned stimulus (CS)] comes to predict a potentially harmful event such as a footshock [the unconditioned stimulus (US)]. During threat learning, an association between the sensory representations of the CS and US is formed in a brain area called the amygdala, especially in the lateral amygdala (LA). However, the synaptic pathways that carry information about a footshock to the LA are unknown. More generally, it has not been addressed which brain area(s) upstream of the amygdala process aversive somatosensory events and conduct this information to the amygdala.


To address this question, we took advantage of optogenetic approaches in behaving mice. We concentrated on the insular cortex, which is known to send axons to the amygdala and to respond to tones and somatosensory stimulation. Mice were trained to acquire a fear response, assessed as immobility (freezing), when a tone was paired with a mild footshock. Our principal approach was to use an inhibitory optogenetic protein to suppress action potential (AP) firing in insula neurons or in the axons connecting the insula with the amygdala. We hypothesized that if the insula sends information about the US to the amygdala, then this manipulation should suppress threat learning.


Silencing AP activity in the posterior insular cortex suppressed acute fear behavior and strongly impaired the formation of threat memories 1 day later, when tones were applied alone. Anatomical tracing and ex vivo electrophysiological experiments then showed that two largely separate neuron populations in the insular cortex form strong excitatory synapses with neurons in the LA or the central amygdala (CeA). Silencing the projection from the insular cortex to the CeA during the US reduced acute fear behavior, whereas silencing the projection to the LA impaired the formation of a threat memory 1 day later but left acute fear behavior unchanged. Complementary experiments with an excitatory optogenetic protein showed that activation of the insular neurons that target the CeA (CeA projectors) rapidly initiated immobility, but this manipulation did not result in an aversive memory. Conversely, optogenetic activation of the LA projectors paired with a tone led to strong aversive behaviors and, on the next day, to escape-like behaviors when the tone was presented alone, showing that stimulation of LA projectors creates an aversive memory. In vivo recordings showed that about one-quarter of the neurons in the posterior insular cortex responded to footshocks (the US). A similar but not completely overlapping neuron population acquired a response to the tones when these were reinforced by footshocks during the threat learning paradigm. Finally, silencing the posterior insular cortex during tone presentation on the retrieval day revealed a contribution of the insula to threat memory retrieval.


The insular cortex is intricately involved in processing aversive somatosensory information. Silencing the posterior insular cortex largely removes the aversive quality of footshock stimulation, thus suppressing an essential drive for learning about such harmful events. The insular cortex routes information to specific amygdalar subdivisions and can thus drive temporally separate components of fear behavior. Furthermore, the insula forms associations between innocuous and harmful sensory events and, together with the LA, is necessary for the retrieval of threat memories. Taken together, the posterior insula processes aversive somatosensory events and contributes to elaborate their negative valence.

Optogenetic studies to determine aversive signaling in brain areas upstream of the amygdala.

Optogenetic silencing of the insula during the footshock leads to reduced threat learning (left). Largely separate neuron populations of the insular cortex project to the LA (red neurons) and the CeA (green neurons) (middle). Selective silencing of the projection to either the CeA or LA impairs different phases of threat learning (right).


Learning about threats is essential for survival. During threat learning, an innocuous sensory percept such as a tone acquires an emotional meaning when paired with an aversive stimulus such as a mild footshock. The amygdala is critical for threat memory formation, but little is known about upstream brain areas that process aversive somatosensory information. Using optogenetic techniques in mice, we found that silencing of the posterior insula during footshock reduced acute fear behavior and impaired 1-day threat memory. Insular cortex neurons respond to footshocks, acquire responses to tones during threat learning, and project to distinct amygdala divisions to drive acute fear versus threat memory formation. Thus, the posterior insula conveys aversive footshock information to the amygdala and is crucial for learning about potential dangers in the environment.

Animals and humans need to learn about potential dangers in the environment. Strong mechanisms of threat learning are therefore present in most animal species (1, 2). In classical fear conditioning (3), animals learn to associate an innocuous sensory event such as a tone [the conditioned stimulus (CS)] with an inherently aversive event such as a mild electrical footshock [the unconditioned stimulus (US)]. After threat learning, the CS acquires an emotional value and then induces a defensive response such as freezing behavior (immobility) (4, 5). The lateral amygdala (LA) has been identified as a brain structure that is critical for threat learning and in which an association between the CS and the US takes place (510). Auditory information reaches the LA from the auditory thalamus and cortex (5, 11, 12). However, the brain areas upstream of the amygdala that process information about aversive somatosensory events and transfer this information to the amygdala have remained elusive (1315).

The insular cortex processes multisensory information, including visceral (16), gustatory (17, 18), somatosensory (19), and auditory modalities (1921). It provides output to the autonomic nervous system (22), and the gustatory area of the insular cortex mediates approach or avoidance behavior in response to different tastants (18, 2325). Functional imaging in humans suggests that the insula is involved in multimodal sensory processing and some cognitive functions (26). The human insular cortex also becomes activated by painful stimuli as part of a larger pain network (27), and patients with stroke-induced lesions that include the insular cortex suffer from pain asymbolia, the absence of emotional response to painful stimuli (28). Nevertheless, whether the insular cortex provides information about aversive somatosensory events to the amygdala during threat learning has remained unclear (13, 29).

Silencing the posterior insular cortex suppresses threat learning

We used in vivo optogenetic methods in behaving mice to test whether the insular cortex might process footshock information relevant for threat learning. Mice were exposed to six tone blocks on day 1 in a habituation session (Fig. 1A). On day 2 (training day), a 1-s footshock was given after each tone block, which induced freezing (4). On day 3, four tone blocks were given without a footshock in a different context, to test the retrieval of the auditory-cued threat memory (5) (Fig. 1A). Because we hypothesized that the insular cortex might code for footshock information, we aimed to inactivate the insular cortex selectively during each footshock. For this, we used an adeno-associated virus (AAV) driving the expression of the light-sensitive Cl transporter halorhodopsin (eNpHR3.0) (30). The AAV was injected bilaterally in the posterior insular cortices 4 weeks before behavioral testing, and an optical fiber was implanted over each injection site (Fig. 1B and fig. S1). On the training day, yellow light was applied for 3 s beginning 1 s before the footshock (Fig. 1A); mice expressing enhanced green fluorescent protein (eGFP) served as control animals (see the Materials and methods section). Control experiments showed that yellow light strongly hyperpolarized halorhodopsin-expressing neurons in the insular cortex, whereas in naïve insula neurons without halorhodopsin expression, yellow light did not influence the electrical signaling (fig. S2).

Fig. 1 The posterior insular cortex is necessary for threat learning and acute fear behavior.

(A) Threat learning protocol (left) and timing of halorhodopsin-mediated silencing of insular cortex neurons during the US (right). (B) Scheme of bilateral injection of an AAV1 vector that drives expression of halorhodopsin or a control construct, followed by bilateral placement of optical fibers over the injection sites in the insular cortex. pInsCx, posterior insular cortex. (C) Time course of the freezing level across the 3 days of threat learning for the control group (black) and the halorhodopsin (Halo)–expressing group (red). Significance was tested by ANOVA and Bonferroni post hoc analysis (Materials and methods). n, number of mice. (D) Average freezing levels during the training day in a pooled analysis (Fear; freezing levels pooled from the fifth and sixth pairings on day 2) and during retrieval on day 3 (Retrieval) for both the control group and the halorhodopsin group (black and red, respectively). Statistical significance was computed by Student’s t test. (E) Side-view map of the mouse brain according to (49), displaying the location of the insular cortex and its adjacent cortical areas. The bottom panel is a zoomed-in view of the posterior insular cortex with granular (GI), dysgranular (DI), and agranular (AIP) areas. The locations of optical fibers in the halorhodopsin-expressing mice are indicated by dots, and the estimated areas of silenced brain tissue (55) are depicted by the green shaded areas (see also fig. S1). AI: agranular insular cortex; AID and AIV: dorsal and ventral parts of the AI; S1 and S2: primary and secondary somatosensory cortex; Au1: primary auditory cortex; AuV and AuD: secondary auditory cortex, ventral and dorsal parts; V1 and V2: primary and secondary visual cortex; Ect: ectorhinal cortex; PRh: perirhinal cortex.

In control mice, an increasing freezing response with each tone presentation was observed during the training (day 2). During threat memory retrieval (day 3), the mice showed pronounced freezing in response to the tone (Fig. 1C, black data). In contrast, in the halorhodopsin-expressing group, freezing was significantly reduced [analysis of variance (ANOVA); P < 0.001; F1,176 = 140.4]. A post hoc Bonferroni test showed that freezing was significantly reduced in the halorhodopsin group compared with the control group during the third to sixth tone–footshock pairing on the training day and during all four tone presentations on the retrieval day (Fig. 1C; P < 0.05 and 0.001). Control mice displayed prolonged freezing during the fifth and sixth tone–footshock pairing on the training day, whereas the halorhodopsin-expressing mice moved normally and continued to explore the environment at the corresponding times, with only brief bouts of freezing (movie S1; pooled analysis of the last two CS–US pairings, P < 0.001, t test; Fig. 1D, left). Similarly, the average freezing level in response to the n = 4 CS presentations on the retrieval day was significantly reduced in the halorhodopsin group compared with the control group (Fig. 1D, P < 0.001, t test; see fig. S3 for freezing levels for individual mice). On the other hand, mice in the halorhodopsin group displayed unchanged acute jumping in response to the footshock (fig. S3), indicating that not all aspects of the painful stimulus were blocked by silencing the posterior insular cortex. Post hoc mapping of the injection sites and of the optical fiber placement sites confirmed that a region in the granular part of the posterior insular cortex was silenced (Fig. 1E).

Insular cortex makes functional connections with LA and CeA

Given the role of insular cortex activity for threat learning shown in Fig. 1, we mapped the potential output connections from the insular cortex to the amygdala (3133), both anatomically and functionally (Fig. 2). Anterograde tracing using AAV8:hSyn:Chronos-eGFP injected into the posterior insular cortex (see Materials and methods) showed fibers in the anterior part of the LA and the basal amygdala (BA), in the anterior part of the central amygdala (CeA), as well as in the posterior part of the BA (Fig. 2, A and B). To validate that these output fibers are functional and to assess the neurotransmitter employed at these putative connections, we used an ex vivo optogenetic approach for mapping long-range connections (34). Specifically, we made recordings in amygdala slices 4 weeks after the expression of Chronos-eGFP in the insular cortex (fig. S4). In all neurons recorded in the anterior LA and anterior CeA, local stimulation by blue light caused excitatory postsynaptic currents (EPSCs) with a fast, NBQX (2,3-dihydroxy-6-nitro-7-sulfamoylbenzo[f]quinoxaline)–sensitive component at negative holding potentials. At positive potentials, an additional slow APV (d,l-2-amino-5-phosphonovaleric acid)–sensitive component was observed, which identifies the connections from the posterior insular cortex to the LA and CeA as glutamatergic (Fig. 2, C and D). Control experiments at the insula-to-LA connection showed that EPSCs were blocked by tetrodotoxin (1 μM) and partially recovered when the K+ channel blocker 4-aminopyridine (1 mM) was coapplied with tetrodotoxin, showing that this connection is monosynaptic (fig. S4). EPSC amplitudes increased gradually with stimulus strength, indicative of compound EPSCs composed of smaller unitary EPSCs (fig. S4). In the LA, postsynaptic neurons included CaMKII-positive pyramidal cells, which were targeted for recording by the use of CaMKIICre x tdTomato mice. Furthermore, EPSCs with similar amplitudes were observed when Chronos-eGFP expression was targeted to CaMKII-positive neurons in the insular cortex or when Chronos-eGFP was expressed nonselectively (Fig. 2C), suggesting that many presynaptic neurons in the insula were CaMKII-positive. In the CeA, a brain structure rich in inhibitory neurons (35), EPSCs with approximately similar amplitudes were observed in somatostatin-positive (SOM+) and somatostatin-negative (SOM) neurons, as revealed by SOMCre x tdTomato mice (fig. S4), in agreement with a recent study (24). Taken together, neurons in the posterior insular cortex make strong glutamatergic output synapses in both the CeA and the LA.

Fig. 2 The posterior insular cortex makes robust excitatory projections to both the LA and CeA by largely separate projector populations.

(A) Coronal slice of a mouse brain showing the expression of Chronos-eGFP in the posterior insula 3 weeks after injection with a AAV8:hSyn:Chronos-eGFP virus. Scale bar, 500 μm. (B) Images of eGFP-positive fibers in the anterior (left) and posterior (right) amygdala complex. Scale bars, 250 μm. CeL, CeM, and CeC: lateral, medial, and capsular parts of the CeA; LAv, LAd, and LAm: ventral, dorsal, and medial parts of the LA; BAl and BAm: lateral and medial parts of the BA. (C) Scheme of experimental approach (left) and optogenetically evoked EPSC (middle) in a CaMKII-positive LA neuron under control conditions (black) and after application of 50 μM APV (red) at two different holding potentials (−70 and +50 mV). The bar graph on the right shows the average EPSC amplitude (AMPA component, −70 mV), separated for recordings after Cre-dependent expression of Chronos in CaMKII-positive neurons (n = 24 recordings) or after Cre-independent expression of Chronos in the insula (n = 18 recordings). (D) Same as in (C) but for a recording of a SOM+ neuron in the CeA. The fast component of the EPSC was blocked by NBQX (5 μM; green), and the remaining slow EPSC at +40 mV was sensitive to APV (blue). (E) Brain section at bregma −0.94 mm (right) showing the injection of the retrograde labels CTB-647 (red) and CTB-488 (green) centered in the CeA and LA, as shown in the scheme (left). (F) Coronal mouse brain section on the level of the posterior insular cortex (bregma −0.9 mm) with projection neurons labeled by CTB-647 (red, LA projectors) and CTB-488 (green, CeA projectors). Yellow rectangle, ROI used for the cell quantification shown in (G). (G) Quantification of LA projectors (red), CeA projectors (green), and neurons labeled by both retrograde labels (yellow) over the thickness of the insular cortex (x axis) and for four separate regions (S2, GI, DI, and AIP). The inset bar graphs show quantifications over the entire cortical depth (average ± SEM of neuron numbers from n = 3 subsequent sections). Data in (C) and (D) are average ± SD.

We next analyzed the degree of overlap of the neuron populations in the posterior insula that project to the LA and to the CeA. We employed double-retrograde labeling experiments with injections of green- or red-labeled choleratoxin B (CTB-A488 or CTB-A647) into the CeA and LA (Fig. 2E). This revealed two largely segregated populations of LA and CeA projectors (Fig. 2, F and G, and figs. S5 and S6; n = 3 mice). LA projectors were found in the secondary somatosensory cortex (S2, ventral part) and in the granular and dysgranular part of the posterior insular cortex. On the other hand, CeA projectors were localized in various layers of the dysgranular and agranular posterior insular cortex but could also be sparsely found in the deeper layers of the granular posterior insular cortex (Fig. 2, F and G, and fig. S6). The overlap between the two populations was weak, especially in S2 and the granular posterior insular cortex (<1% for S2; <1, 5, and 13% for granular, dysgranular, and agranular posterior insular cortex, respectively) (Fig. 2G).

Insula–amygdalar connections route information for acute fear versus threat memory

We next investigated whether the output connections of insular cortex neurons to either the CeA or the LA have different functions during acute fear behavior versus 1-day threat memory. We silenced the output fibers of the insula specifically in either the CeA or the LA at the time of footshock presentation during training (Fig. 3A), after virally expressing halorhodopsin in the posterior insular cortex (see fig. S7 for the histological validation of fiber placements). Silencing of the insula output fibers in the CeA (Fig. 3B) caused significantly reduced freezing in the halorhodopsin group compared with the control group (Fig. 3C; ANOVA, P < 0.001, F1,240 = 25.66; fig. S8). A post hoc Bonferroni test showed a significant reduction of acute fear behavior on the training day (P < 0.01 and 0.001 for the fifth and sixth pairing) (Fig. 3C; see also Fig. 3D, left, pooled analysis, P < 0.001; t test). However, threat memory retrieval was not significantly different in the halorhodopsin and control groups (Fig. 3C; P > 0.05, Bonferroni post hoc test; Fig. 3D, right, pooled data, P = 0.71, t test).

Fig. 3 Differential roles of the insular cortex projections to the CeA and LA in acute fear behavior versus threat memory.

(A) We expressed halorhodopsin bilaterally in the posterior insular cortex and placed optical fibers over the output axons in either the CeA [(B) to (D)] or the LA [(E) to (G)]. We silenced the output fibers during the footshock presentation on the training day (A). (B to D) Freezing levels throughout the 3 days of behavior (C) and pooled analysis of freezing for both day 2 and day 3 [(D), left and right, respectively] for the experiments in which the posterior insula CeA synapse was silenced. Note the reduced acute fear behavior especially for the fifth and sixth pairing on day 2, whereas freezing during the retrieval day was not significantly reduced. (E to G) Results for the posterior insula LA synapse silencing. Acute fear behavior was not significantly altered, whereas threat memory retrieval on day 3 was significantly reduced [(F) and (G), right]. Data in (C), (D), (F), and (G) are mean ± SD.

When we placed the optical fibers bilaterally over the LA (Fig. 3, E to G, and fig. S9), there was also a significant difference in freezing in the halorhodopsin group compared with the control group (Fig. 3F; ANOVA, P < 0.001, F1,240 = 45.3; fig. S10). However, the post hoc Bonferroni test indicated no significant difference in acute fear behavior on day 2 (Fig. 3F, P > 0.05 for each of the six pairings; Fig. 3G, left; P > 0.05). Conversely, 1-day threat memory was significantly reduced in the halorhodopsin group (Fig. 3F, P < 0.01 or 0.001, Bonferroni post hoc test), as was also revealed by the pooled analysis for the retrieval day (Fig. 3G, right, P < 0.001, t test). Reduced threat memory retrieval but unchanged acute fear behavior were also observed when LA neurons themselves were silenced during the footshocks on the training day (fig. S11) (36). Therefore, LA projectors of the insular cortex probably drive LA neuron depolarization during the footshock, an activity necessary for forming an associative memory in the LA. However, the activity of LA neurons during the footshock was not necessary for acute fear behavior on the training day.

Separate insula neuron pools drive freezing versus aversive adaptive behaviors

To further investigate the roles of CeA and LA projectors in the posterior insula, we activated each projector neuron population separately. We injected an AAV vector that is taken up retrogradely and that drives Chronos-eGFP expression (AAVretro:hSyn:Chronos-eGFP) (37), bilaterally into the CeA or into the LA, and placed optical fibers bilaterally over the posterior insular cortex (fig. S12). Four weeks later, mice underwent a modified threat learning paradigm in which the footshocks on the training day were replaced by optogenetic stimulation of either CeA or LA projectors (1-ms stimuli at 40 Hz for 10 s) (Fig. 4A). Stimulation of the CeA projectors stopped any movement of the mice (Fig. 4, B and D, and movie S2). After cessation of blue-light stimulation, the mice swiftly returned to their normal exploratory behavior. Presenting the mice with tones on the next day evoked neither head shaking nor freezing (Fig. 4, C and D; fig. S13; P < 0.05 and 0.01, respectively; n = 4 mice), and the time spent grooming did not differ between the 2 days (Fig. 4D; P = 0.94). Thus, CeA projectors in the posterior insular cortex initiate freezing behavior, but their activity does not leave an aversive memory trace.

Fig. 4 Separate insula projectors drive freezing versus aversive adaptive behaviors.

(A) Protocol of the optogenetic stimulation used to activate either the CeA or LA projectors in the posterior insular cortex. On day 2, the optogenetic stimulation (a 10-s train of 40-Hz light stimuli of 1-ms duration) coterminated with 30-s blocks of tone stimulation (left). On day 3, the tones were given alone (right). (B) (Left) Scheme of the targeting of CeA projectors. (Right) Scored behavior of one example mouse. Mice exhibited stereotypic freezing behavior selectively during the 10 s of optogenetic stimulus. (C) During the third day, tone stimuli alone did not evoke a notable behavioral response. (D) Quantification of the behaviors observed in four mice. During day 3, the tone evoked significantly less head shaking and freezing than on day 2 (P < 0.05 and 0.01, respectively; t test), whereas the mice spent significantly more time in exploration (P < 0.001; t test). (E) Optogenetic activation of LA projectors resulted in a range of pronounced defensive behaviors (see key) that outlasted the optogenetic stimulation. (F) During the third day, presentation of the first tone (in the absence of optogenetic stimulation) induced escape-like behavior (dark gray), which outlasted the entire experiment session. (G) Quantification of the behaviors observed in five mice. Note the significant shift from various aversive behaviors on day 2 toward escape-like exploration on day 3 (gray bar; P < 0.001). Most other behaviors also differed significantly from day 2 to day 3 (*P < 0.05; **P < 0.01; ***P < 0.001).

We then stimulated LA projectors in an analogous approach (Fig. 4, E to G). During and after the 10-s optogenetic stimulation blocks, the mice displayed a sequence of strongly aversive behaviors, including head shaking and backward moving, followed by rising on the hindpaws, movement of forepaws in the air with closed eyes, and flattened ears (Fig. 4E and movie S3). After cessation of light stimulation, mice slowly recovered and often showed freezing, excessive grooming, or escape-like exploratory behavior (Fig. 4E and fig. S14; n = 5 mice). Varying the light intensity showed that aversive adaptive behaviors started to appear at ~10 to 30% of maximal blue-light intensity, but no qualitatively different behaviors were observed at lower light intensities (fig. S14; n = 3 mice). We hypothesize that the strong intensity of aversive behaviors is caused by the stimulation of a large number of LA projectors, which might be equivalent to a strong internal representation of various painful sensations. One day later, exposing mice to the first tone block in a different context caused escape-like exploratory behavior, with occasional bouts of other defensive behaviors that continued over the entire retrieval session, regardless of the timing of subsequent tone blocks (Fig. 4, F and G, and fig. S14; n = 5 mice).

Insular neuron activity reflects threat learning of an auditory CS

The posterior insular cortex plays a role in acute fear learning and in 1-day threat memory (Figs. 1 to 3). We therefore sought to record the response of posterior insular cortex neurons in vivo over the 3 successive days of the threat learning protocol. We used four tetrode electrodes placed around a central optical fiber (Fig. 5A) (38). We injected an AAVretro driving the expression of mCherry-IRES-Cre into the LA of channelrhodopsin-2 (ChR2) reporter mice (Fig. 5, A and B), which resulted in the expression of ChR2 in LA projectors of the posterior insular cortex (Fig. 5B). This, in turn, allowed us to make recordings from optogenetically identified LA projectors (fig. S15) (38), as well as from nonidentified units in the posterior insular cortex.

Fig. 5 Insula neurons respond to US stimulation and acquire CS responsiveness.

(A) Schematic showing the optrode placement in the posterior insular cortex and the injection of an AAVretro vector driving the expression of Cre-recombinase and mCherry in a channelrhodopsin reporter mouse. The scheme on the right shows the optrode with four tetrodes (T1 to T4). (B) Expression of mCherry in LA (left) and the retrogradely expressing neurons in the insular cortex (right). Note the position of the optrode in the granular part (GI) of the posterior insular cortex (right). Scale bars, 250 μm. ec, external capsule. (C) Freezing level across the 3 days of fear conditioning training for the example mouse. (D) Responses of a unit to tone (CS) presentations on the training day [top; average (av) z-score response to the n = 30 0.1-s tones during each tone block] and to the footshock coterminating with each tone block (bottom). PSTH, peristimulus time histogram. The response on the far right is the average over n = 6 footshocks. (E) Response of the same unit as in (D) to n = 4 tone blocks (CS) during retrieval on day 3. Note the response to footshocks, the entrainment to tones on day 2, and the retained tone responsiveness on day 3. Correspondingly, this unit was classified as a US+ CS-entrained unit. (F) Distribution of units with different combinations of US responsiveness and CS entrainment, averaged over all units of the sample (78 units from six mice). (G and H) z-score responses of all units measured in the sample mouse from (B) to (E) in response to the tone blocks (G) and to the footshocks on day 2 (H). The data was separated for non–CS-entrained units (gray) versus CS-entrained units (pink), and the average response of each group is shown superimposed (red and black; average ± SEM). (I) Grand average of responses to tone of non–CS-entrained units (black) and CS entrained units (red) for all units from six mice. In the figure, red values of n indicate CS-entrained units, whereas black values denote units that were nonresponsive to the CS.

Overall, we found n = 20 units in the posterior insular cortex that responded to the footshock on day 2, out of a total of n = 78 units that could be followed over all 3 days (n = 6 mice). Figure 5, D and E, shows a unit that responded to the footshock. This unit was initially unresponsive to tones but developed a response to the tone (CS) on the training day and retained this response during threat memory retrieval on day 3 (Fig. 5, D and E). Such units were called US-responsive (US+) and CS-entrained units. In addition, we found other combinations of responses, including units that were nonresponsive to footshocks and did not show CS entrainment; units that showed CS entrainment but had no footshock response; and units that responded to footshocks but did not show CS entrainment (see fig. S16 for examples of the response types and fig. S17 for examples of spike waveforms). Thus, there was some dissociation between US-responding neurons and the plasticity of the CS response, as found recently in in vivo imaging of BA/LA neurons (10). Within the population of optogenetically identified LA projectors, the summed categories of all US-responsive units were enriched (Fig. 5F; sum of red and pink categories, P = 0.026, Fisher’s exact test). On the other hand, the fraction of CS entrained units was not significantly different for unidentified units (22%) and for the putative LA projectors (39%) (Fig. 5F; sum of red and dark blue categories; P = 0.16; Fisher’s exact test).

We averaged the responses to tones and footshocks over all 3 days for both CS-entrained and non–CS-entrained units within mice (Fig. 5, G and H, and fig. S18) and then calculated the grand average across mice (Fig. 5I). This showed that CS-entrained units developed a robust response to the tone presentation on the training day and retained this response during the retrieval day (Fig. 5I).

The insular cortex contributes to threat memory retrieval

The observation of tone-entrained units indicates that associative plasticity takes place in the insular cortex. Therefore, activity in the insula might also be necessary for the retrieval of threat memory. To test this possibility, we silenced the posterior insular cortex during each tone pip on the retrieval day after bilateral expression of eNpHR3.0-eYFP in the insula (Fig. 6A and fig. S19). This resulted in significantly reduced freezing in the halorhodopsin group compared with the control group (Fig. 6B; ANOVA, P < 0.01, F1,100 = 8.59). Analysis with a post hoc Bonferroni test showed significantly reduced freezing during two tone blocks on the retrieval day (Fig. 6B; P < 0.05). Similarly, pooled analysis showed significantly reduced freezing during retrieval (Fig. 6C; P < 0.001) but no change during training as expected (Fig. 6C, left, P = 0.8). Post hoc mapping of the fiber positions confirmed a position in the granular posterior insular cortex (Fig. 6D). Thus, tone-driven activity of the posterior insular cortex contributes to the retrieval of auditory-cued threat memory.

Fig. 6 Tone-driven insular cortex activity is necessary for threat memory retrieval.

(A) Protocol to optogenetically silence the response of insula neurons to each tone on the retrieval day. (B) Time course of the freezing level across the 3 training days for control mice (black) and the halorhodopsin-expressing mice (red). (C) Freezing level during the training day (Fear; freezing averaged for the fifth and sixth tone–footshock pairing) and the retrieval day (all four presentations grouped) for the control and the halorhodopsin-expressing animals. ns, not significant. (D) Side-view map displaying the position of each optical fiber in halorhodopsin-expressing mice. Data in (B) and (C) are mean ± SD.


The posterior insular cortex plays an important role in threat learning. Silencing of neurons in the posterior insular cortex during the footshock largely removed the aversive qualities of the US and subsequently prevented the formation of 1-day threat memory (Fig. 1). Because a teaching signal about an aversive event is thought to drive associative threat learning at CS-coding synapses in the LA (5, 11, 15) and possibly in other areas (39), suppression of a relevant representation of the aversive sensory event should impair the formation of threat memory, as we observed (Figs. 1 and 3). Furthermore, because optogenetic stimulation of LA projectors in the insula was strongly aversive for mice (Fig. 4), and because silencing of this connection during the footshock impaired threat memory formation (Fig. 3F), the posterior insula likely carries US information to the LA during associative threat learning. It has recently been shown that a neuron population in the BA codes for the aversive qualities of pain (40). Because we found that insular cortex activity is necessary for the aversive properties of footshocks, it is possible that insular cortex–amygdala networks cooperate in attributing aversive meaning to painful stimuli. Affective pain information can also be conveyed from the parabrachial nucleus to the CeA (41), but the latter pathway cannot explain associative plasticity in the LA.

Beyond providing information about a US to the amygdala, our study shows that the insular cortex itself substantially contributes to the formation of a 1-day threat memory. Many neurons in the posterior insular cortex, including LA projectors, responded to the US. A partially overlapping population of neurons relative to these US responders developed a response to the tone (CS) during threat learning and retained this response during retrieval 1 day later (Fig. 5). Consistent with the insular cortex’s role (besides LA) for storing threat memories, silencing the insula during the tone stimulation on the retrieval day reduced threat memory (Fig. 6). Thus, 1-day auditory cued threat memories might be stored in both the LA and the insular cortex, whereas longer-lasting threat memories become dependent on secondary sensory cortices (42). A functional magnetic resonance imaging study of fear conditioning in rats using a visual CS has shown activation of the LA, granular insular cortex, and hypothalamus upon replay of the CS after fear conditioning (43), consistent with our finding that the insular cortex contributes to threat memory retrieval. Because the posterior insular cortex is immediately upstream to the LA and makes strong glutamatergic synapses in the LA (Fig. 2), it is possible that the increased activity of insular cortex neurons to the CS (Fig. 5) contributes to the increased tone responsiveness of LA neurons after threat learning (7, 10). The serial arrangement of the posterior insular cortex and LA evokes a model of memory storage in which associative plasticity takes place serially in connected brain structures, a mechanism that might amplify the response to the CS of the more downstream brain areas.

The posterior insular cortex is also essential for acute fear behavior during the training day (Fig. 1), with a clear contribution of the insular cortex to the CeA pathway (Fig. 3). In agreement with this role, optogenetic stimulation of CeA projectors in the insula caused strong freezing (Fig. 4), consistent with the previous demonstration that optogenetic stimulation of CeA SOM+ neurons, or more nonselective stimulation in the medial CeA, drives freezing (35, 39). Amygdalar connections from the gustatory part of the insular cortex, which is located anterior to the posterior insular cortex studied here (16, 44, 45), have been shown to cause conditioned place aversion in response to bitter tastants (2325), suggesting roles for both the somatosensory insular cortex and the gustatory insular cortex in aversive valence coding. Our finding that CeA projectors in the insular cortex drive acute freezing behavior whereas LA neurons do not (Fig. 3 and fig. S11) reinforces previous evidence that brain structures different from the canonical LA and BA can access the CeA in fear behavior (46).

In human stroke patients with lesions that include the posterior insular cortex, an absence of emotional and avoidance responses to pain or threatening gestures was observed, a condition termed pain asymbolia (28). This condition is analogous to the behavioral deficits described here in which mild painful stimuli lose their aversive meaning in mice with a silenced posterior insular cortex (Fig. 1). As we show here in mice, the rich interconnectivity between the insular cortex and amygdala (Fig. 2) (3133) was responsible for these deficits. Thus, the insular cortex has a fundamental role in evaluating aversive events and signaling them to the LA, in preparing motor actions in response to danger via its connections to the CeA and possibly to other structures, and in storing associative memories about threats. Given the crucial functions of the insular cortex in threat learning, and given that insular cortex activity is altered in several psychiatric diseases (47, 48), a more detailed understanding of plasticity and circuit mechanisms in this brain area will be of high relevance.

Materials and methods


All experimental procedures on laboratory animals (Mus musculus) were conducted in accordance with the veterinary office of the canton of Vaud, Switzerland (authorizations 2885.0, 3274.0). For behavior experiments we used male C57BL6/J mice (6 to 8 weeks old), purchased from Charles River Laboratories (France) and maintained in a breeding colony in the EPFL-SV animal facility. For some experiments (Fig. 5), ChR2 reporter mice of the same age were used (Rosa26:lsl:ChR2-eYFP, Jackson Lab 024109, Ai32). For the ex vivo experiments, SOMcre x tdTomato mice (Somatostatin-ires-Cre mice; Jackson Lab 013044, crossed with Ai9 reporter line Rosa26:lsl:tdTomato; Jackson Lab 007909) and CaMKIIcre x tdTomato (CaMKII-Cre mice; Jackson Lab 005359, crossed with Ai9 reporter line) were used. The mice were maintained on a normal 12-hour light/dark cycle, and provided with food and water ad libitum. Mice were randomly assigned to different experimental groups.

Stereotaxic surgery for virus injection and optical fiber implantation

Mice were anesthetized by inhalation of a 3% isoflurane mix in O2 gas (produced by the CombiVet animal gas anesthesia system; Rothacher Medical, Switzerland) and maintained under inhalation of 1 to 1.5% isoflurane in O2. Mice were then placed in a stereotactic apparatus (Kopf Instruments, model 940, USA) on a heating pad (Harvard Instruments, USA) and local subcutaneous analgesia using 2% lidocaine was applied. The coordinates of the craniotomy were determined by a mouse brain atlas (49). Following exposure of the skull, a hole was drilled on each side with a 0.5-mm bore (Komet Dental, Germany) using a Tech2000 drill handpiece (Ram Products Inc, USA). Viral suspension was microinjected using 10- to 20-μm tip diameter glass pipettes (PCR capillaries, Drummond Scientific, USA; pulled on a P-97 puller, Sutter Instruments, USA) and an oil hydraulic micromanipulator (MO-10, Narishige, Japan) at the following coordinates (AP, posterior from the bregma/ML, ± lateral from the midline/DV, ventral from the skull surface at bregma, in mm): for insula: −0.9/±3.9/−3.8; for LA: −1.15/±3.45/−4.45; for CeA: −1.22/±2.65/−4.70. For in vivo optogenetic manipulations, we implanted optical fibers bilaterally in the skull as described by (50) using dental UV curing cement (Tetric EvoFlow; Ivoclar Vivadent, Liechtenstein). The implants were inserted at the following stereotaxic coordinates: posterior insula (AP: −0.95 mm; ML: ±3.80 mm; DV: −3.40 mm); LA (AP: −1.15 mm; ML: ±3.48 mm; DV: −3.60 mm); CeA (AP: −1.22 mm; ML: ±2.60 mm; DV: −4.17 mm). To prevent interference with the implants by other animals, the mice were singly housed after the surgery. Behavioral experiments were performed 3 to 5 weeks later, to allow sufficient time for the injected AAV to express the opsins.

Viral vectors

To silence either the posterior insula neurons (Figs. 1 and 6), or the terminal areas of posterior insula projections to the CeA or the LA (Fig. 3), or else LA neurons (fig. S11), we used an AAV1:hSyn:eNpHR3.0-eYFP (Addgene). As controls in these experiments, we used an AAV1:hSyn:eGFP (Addgene). For the ex vivo optogenetic recordings (Fig. 2), an AAV8:hSyn:Chronos-eGFP (University of North Carolina vector core; UNC) was used to drive the expression of Chronos [a channelrhodopsin variant; see (51)]. In some of these experiments (see Fig. 2C), we used an AAV8:EF1α:FLEX:Chronos-eGFP (UNC) and CaMKIICre mice (see above), to achieve expression of Chronos selectively in CaMKII-positive neurons of the posterior insula cortex. To activate CeA or LA projectors in the posterior insular cortex (Fig. 4), we used a retrograde AAVretro:hSyn:Chronos-eGFP, and as control an AAVretro:hSyn:eGFP (both from Addgene). Usually, 200 nl of viral suspension was injected per site; the viral titer was adjusted to ~3.5·1012 ml−1. For optrode recordings, an injection of 200 nl of retrograde AAVretro:Ef1α:mCherry-IRES-Cre (Addgene) was made in the LA, at a titer of 1.3·1013 ml−1.

Optrode recordings

For in vivo recording of extracellular AP activity of optogenetically identified neurons (Fig. 5) (38), we custom-built optrodes using four tetrodes placed around a central optical fiber (200-μm/230-μm core/outer diameter, NA 0.5; Thorlabs, USA; see Fig. 5A). Tetrodes were twisted from insulated Pt/Ir wires (17-μm diameter; California Fine Wire, USA) and glued onto the external surface of the optical fiber that was glued inside the ceramic ferrule (1.25-mm outer diameter; Thorlabs) and inserted into the movable part of a microdrive (Axona Ltd, UK). The free ends of the tetrode wires were fixed to the pins of a NPD-18-VV-GS micro-connector (Omnetics Corp., USA) that was rigidly attached to the movable part of the microdrive. To minimize the impedance of the recording and reference channels to values typically <100 kilohms at 1 kHz, platinum was deposited on the tips of the tetrode wires using platinum black plating solution (Neuralynx, UK) according to manufacturer’s manual, and an iontophoretic amplifier (MVCS-01, NPI Electronic, Germany). The impedance for each channel was measured in the phosphate-buffered saline solution (PBS) using the lock-in function of an EPC-10 patch-clamp amplifier (HEKA Elektronik, Germany). The optrode implantation surgery into the left insular cortex was similar to the procedure used to implant optical fibers; in addition, a grounding microscrew (Antrin Miniature Specialties, USA) was implanted in the skull, to which a copper wire connected to a ground pin was presoldered.

For the optrode recordings in freely behaving mice, the fear conditioning chamber was surrounded with a custom-built Faraday cage. Spiking activity was acquired with a 16-channel amplifier ME16-FAI-μPA under control of the MC_Rack software (both from Multi Channel Systems) at 40-kHz sampling frequency; unsorted spike events were visualized online after band-pass filtering (0.6 to 6 kHz) and amplitude-based detection (threshold of −3 to −3.5 SD) in the MC_Rack software (see also fig. S15A). One day before the start of the threat learning protocol, the optrode was repositioned in the ventral direction by ~100 to 300 μm using a microdrive (performed under ketamine anesthesia) until the target depth of 3.9 mm was reached. We then applied light pulses (1 to 3 ms, 0.8 to 3 mW at the fiber output, 2-Hz repetition rate) produced by a 473-nm diode pumped solid state (DPSS) laser (MBL-FN-473-150 mW; CNI Lasers, China) that was triggered with a Master-8 stimulator (A.M.P.I., Israel). The opto-tagged units could be detected by apparent light-evoked spiking activity within the expected latency window of 2 to 8 ms after the light pulse onset (fig. S15B). During the 3 days of fear conditioning protocol (Fig. 1A), the spiking activity as well as the synchronization triggers indicating the timing of the sound CS and the footshock US stimuli (generated by the VideoFreeze software; Med Associates Inc, USA) were continuously sampled with the MC_Rack software. After each behavior session, light-evoked spikes were collected by applying ~2000 light pulses at 2 Hz to identify the opto-tagged cells post hoc during the data analysis.

The raw data was converted from MC_Rack into HDF5 format using Multi Channel Data Manager software (MultiChannel Systems). The data was then processed (except the spike clustering) using custom-written routines in IGOR Pro 7 (WaveMetrics, USA). The voltage traces were band-pass filtered (0.6 to 6 kHz; 4th-order Butterworth filter), and footshock stimulation artefacts (only from recordings on day 2) were detected and blanked by zeroing under manual control (see fig. S15A). In some mice, we observed a slow light-evoked artefact likely related to the local field potentials that could not be filtered out completely and thus prevented proper spike detection. To minimize these artefacts in such cases, we subtracted the per-channel average traces calculated over ~200 subsequent light pulses. Negative amplitude spikes were then detected using threshold method (typically set at −3.2 SD) on each channel, and the precise spike locations within each tetrode were dictated by the timestamp of the largest amplitude event. The light-evoked spikes recorded after behavior were sampled in a time window 2 to 8 ms after the light pulse onset (see fig. S15B) and were clustered separately from the spikes recorded during behavior. Individual spike cutouts (filtered for clustering with a wider band-pass 4th-order Butterworth filter, 0.4 to 6 kHz) were then exported into MATLAB (MathWorks, USA). The MClust toolbox (Dr. David Redish; University of Minnesota, USA) was used to cluster the tetrode spikes by an unsupervised clustering algorithm (KlustaKwik) (52), using the spike valley and the principal components PCA1-PCA3 as the clustering parameters. The quality of clustering was manually controlled by checking the average spike waveform similarity and visualizing the cluster projections, and occasionally some of the clusters were fused together if they were not well separated. Following this, the quality of the spike cluster isolation was analyzed by computing the isolation distance and L-ratio, to control for type I and type II errors (53). Only clusters with isolation distance larger than 24 and L-ratio lower than 0.5 were retained for further analysis; this reduced the total number of clusters in the sample from n = 179 to 78 (sum of unidentified units and LA projectors; n = 6 mice). However, qualitatively similar results as the ones shown in Fig. 5 were obtained when this quality control step was omitted.

The clustered timestamp data were re-imported into IGOR Pro for subsequent analyses such as waveform matching, alignments to the stimuli and z-score calculations, averaging and displaying. For each recording day in every animal, identification of the opto-tagged units was done by matching the average waveforms of the clusters recorded during the behavior session with those collected upon delivery of the light pulses. Similarly, this average waveform matching method was used to follow the units across 3 experimental days (fig. S15C). Waveform matching was facilitated by calculation of a metric for each pair of units, which was the product of the summed inverted root mean squared value of the point-to-point difference between the waveforms, and of the inverse of unity complement of the average Pearson’s coefficient calculated between the waveforms for four channels of a tetrode. The experimenter was blind to the spiking pattern of individual units during the behavior sessions until the identity of units (opto-tagged or unidentified, and across days) was determined.

The recorded units were classified into four categories, based on their responses to sensory stimuli, according to the following criteria. In general, a z-score of 2 (1 SD above baseline noise) was considered a response. (i) The unit was considered US-responsive if the z-score during the 1-s-long footshock, calculated from the average of the n = 6 footshock presentations on day 2, exceeded a value of 2. (ii) For the classification of CS responses, average z-scores over 100-ms-long tones were calculated from 30 aligned and averaged tone response (on day 2, only 29 CS repetitions were averaged to avoid the influence by the US response that immediately followed the 30th tone). For the animal FT9742, the z-score was calculated over the first 40 ms after the tone onset because of the brief response characteristics in this mouse. A given unit was considered CS entrained if, first, the z-score did not exceed 2 for at least five out of six CS presentations on day 1 (criterion that the unit was not innately responsive to tones), and, second, if out of the last three tone presentations on day 2, at least two resulted in a summed z-score larger than 4 (criterion for increasing tone response on day 2) or if at least two tone presentations on day 3 resulted in a summed z-score larger than 4 (criterion for the maintenance of the tone response on day 3).

Fear conditioning

Before the start of fear conditioning, the mice were habituated to head tethering with the fiber-optic patch cord. Each mouse underwent one habituation phase daily for a total of 5 to 6 days before starting the behavioral training.

Fear conditioning was conducted on 3 consecutive days in a rectangular conditioning chamber placed in a sound- and light-attenuated box (63.5 cm wide, 35.5 cm high, 76 cm deep; NIR-022MD, Med Associates) equipped with a speaker in a side wall, and a CMOS video camera (30 frames/s) with a near-infrared filter for continuous video recordings of mouse behavior.

On day 1, mice were subjected to a habituation session in the fear conditioning chamber (context A), in which six tone blocks (CS) of 7 kHz at 80 dB were presented (each block composed of 100-ms pips repeated 30 times at 1 Hz). The freezing level of the mice during CS presentation was analyzed as an estimate of baseline behavior.

On the training day (day 2), mice were placed in the conditioning chamber (context A) with a stainless-steel grid floor connected to a shock generator (ENV-414S, Med Associates). Mice were submitted to a fear conditioning paradigm by pairing the conditioning sound stimulus (CS) with a mild electric footshock (US) (AC 0.6 mA, 1-s duration). The onset of the US coincided with the offset of the last tone pip in the CS. The CS-US pairing was repeated six times at a variable intertrial interval of 60 to 90 s.

Threat memory retrieval consisted of the tone recall on day 3. The mice were placed in the conditioning chamber with a new environment (context B) and were reexposed to four presentations of the CS tone without footshock. The freezing behavior was monitored during CS to evaluate the retrieval of cued fear memory. The context B was modified from the original conditioning chamber to minimize context generalization.

During optogenetic behavioral experiments, yellow (561 nm) or blue (473 nm) light was delivered from a DPSS laser (MGL-FN-561-AOM-100mW or MBL-FN-473-150mW, respectively; CNI Lasers). To avoid resting leak of light through an AOM module of the yellow laser, an additional mechanical shutter (SHB05T; Thorlabs) was introduced at the output before the fiber. The shutter was closed at all times except when the laser light had to be applied, for silencing during fear conditioning day 2: 3-s duration, starting 1 s before the footshock; for silencing during retrieval day 3: 250-msec duration, starting 50 msec before each pip; for activation during fear conditioning day 2: 10-s duration, coterminating with the tone. The laser power was adjusted for each animal in accordance with the attenuation value of the given optical fiber implant so that the total light output power was 10 mW at each fiber tip.

Freezing was analyzed using Video Freeze software (Med Associates) and expressed as the percentage of time mice spent freezing during the CS presentation. We found that the optical fiber connected to the mouse head on the day of optogenetic silencing (day 2 in Figs. 1C, 3C, 3F; day 3 in Fig. 6B) caused movement artefacts in the quantification of freezing with the Video Freeze software. To compensate for this, we used different threshold levels on day 2 (versus days 1 and 3) for the data in Figs. 1C, 3C, 3F; for this reason, the absolute values of freezing are not directly comparable between days 2 and 3 in these datasets. For the dataset in Fig. 6B, in which the cable artefact interferes with the correct quantification of the freezing behavior during retrieval, we hand-scored the behavior of all mice in the control and halorhodopsin group on day 2 and day 3, similar as shown in Fig. 4. During hand-scoring, the analyzing person was blinded to the identity of the mouse with respect to control versus halorhodopsin group. The fraction of time spent freezing during the tone blocks was then computed from the scored behavior data.

Ex vivo electrophysiological recordings and optogenetic circuit mapping

For ex vivo electrophysiology, adult mice of the same age range as used for behavior were employed. Precautions were made to obtain viable slice preparations suitable for recordings, including the use of low-Na+ dissection buffer (see below) (54). Mice used for electrophysiological experiments were anaesthetized with isoflurane, decapitated, and their brains were quickly removed and placed in ice-cold dissection buffer (in mM): 110 NMDG (N-methyl-d-glutamine), 110 HCl, 2.5 KCl, 1.2 NaH2PO4, 20 HEPES, 25 Glucose, 5 sodium ascorbate, 2 Thiourea, 3 sodium pyruvate, 10 MgCl2, 0.5 CaCl2, saturated with carbogen gas (95% O2 and 5% CO2). Coronal slices (300 μm) containing the amygdala (with LA or CeA) complex were cut in dissection buffer using a Microtome VT1200S (Leica Microsystems, Germany), and were subsequently transferred to a storage chamber containing a HEPES based storage solution (in mM): 92 NaCl, 2.5 KCl, 30 NaHCO3, 1.2 NaH2PO4, 20 HEPES, 25 glucose, 5 sodium ascorbate, 2 Thiourea, 3 sodium pyruvate, 2 MgCl2 and 2 CaCl2, at 34C°, pH7.4, saturated with carbogen, and allowed to cool down to room temperature. After at least 40 min of recovery time, slices were transferred to the recording chamber and constantly perfused with ACSF (in mM): 125 NaCl, 2.5 KCl, 25 NaHCO3, 1.2 NaH2PO4, 25 glucose, 0.4 sodium ascorbate, 3 Myo-Inositol, 2 sodium pyruvate, 1 MgCl2 and 2 CaCl2, pH7.4, saturated with carbogen. Recordings were done either at room temperature (20 to 22°C; Fig. 2 and fig. S4, F to H) or at 35°C using a bath heater (Warner Instruments; model TC-334B; for the data shown in fig. S2).

To test the functional connectivity between posterior insular cortex and CeA or LA structures, an AAV8:hsyn:Chronos-eGFP virus was injected into posterior insula of SOMCre x tdT mice and CaMKIICre x tdT mice. After brain slicing, whole-cell patch clamp recording was performed in genetically identified cells based on their tdTomato fluorescence, as well as in tdTomato-negative cell in case of CeA recordings. Whole-cell currents were recorded using EPC-10 patch-clamp amplifier (HEKA Elektronik) under control of PatchMaster software (HEKA Elektronik). The patch pipette solution contained (in mM): 140 Cs-gluconate, 10 HEPES, 8 TEA-Cl, 5 Na-phosphocreatine, 4 Mg-ATP, 0.3 Na-GTP, 5 EGTA, pH 7.2 adjusted with CsOH.

For optogenetic activation of Chronos-expressing insula projections to LA or CeA, 1-ms-long blue-light pulses were applied from a high-power LED (CREE XP-E2, emission spectrum centered around λ = 480 nm; Cree Inc, USA) driven by a LED controller (Cyclops LED Driver, The LED light source was coupled into the epifluorescence port (TILLphotonics, Germany) of a BX51WI microscope (Olympus, Japan) equipped with the 60X/0.9 NA objective, thus focusing the light onto a region of interest. Single-light pulses were delivered at 20-s intervals and repeated at least 10 times.

To identify the currents evoked by the stimulation of posterior insula fiber terminals, we applied gabazine (GABAA receptor inhibitor; 5 μM; at −70-mV holding), NBQX (AMPA receptor inhibitor; 5 μM; at −70-mV holding) and AP-5 (2-amino-5-phosphonopentanoic acid; NMDA receptor inhibitor; 50 μM; at +40-mV holding voltage) (all from BIOTREND, Switzerland).

Immunohistochemistry and immunofluorescence microscopy

After virus injection or behavior, mice were perfused transcardially with 4% paraformaldehyde (PFA) solution in PBS. Afterward, the brains were extracted, kept in 4% PFA overnight for postfixation and then dehydrated in 30% sucrose solution for 2 days. Frozen brains were cut into 40-μm-thick coronal slices with a sliding freezing microtome Hyrax S30 or HM450 (Carl Zeiss, Germany). Sections were washed with PBS and mounted on slides in Dako fluorescence mounting medium (Dako, Switzerland).

For assessing the viral expression and the placement of optical fibers, images of serially mounted slices were taken with a slide scanner microscope VS120-L100 (Olympus) with 10X/0.4 NA air objective or an upright LSM 700 confocal microscope (Carl Zeiss) with 20X/0.8 NA air and 40X/1.3 NA oil-immersion objectives (Bioimaging and Optics Platform, BIOP, EPFL). Cell quantification following CTB injection in LA and CeA (Fig. 2 and figs. S5 and S6) was performed using an automated ImageJ plugin (developed by Mr. Olivier Burri, BIOP, EPFL). For this analysis, the cells were counted in 20-μm bins within regions of interest (ROIs) of 250×1100 μm oriented perpendicularly to the pia.

Verification of the optical fiber placement from series of histological sections (40 μm) was done with a custom-written image analysis routine in Igor Pro 7 (available upon request). This routine allowed manual alignment of a model optical fiber with a fiber tract in 3D, thus taking into account the tilt introduced during trimming and sectioning. After the alignment, a light cone volume [calculated according to (55)] was mapped onto the tissue below the fiber to assess the brain areas that were illuminated in vivo (see figs. S1, S7, S9, and S18 for the mapped fiber tips and light cones shown in red and blue, respectively).

The histological sections were aligned to the reference mouse brain atlas (49). The bregma position of the anatomical plates that correspond to a given section is indicated on the histological sections. The abbreviations of names of brain areas follow the reference brain atlas (49) (see Figs. 2 and 5 and figs. S1, S4, S6, S7, S9, S12, and S19). A side-view map of the mouse brain (Figs. 1E and 6D) was constructed by projecting the edges of the individual cortical subareas from each coronal section of the reference mouse brain atlas (49), centering on the insular cortex and its dorsal and posterior neighboring cortical areas.

Statistical analysis

Data are expressed as mean values ± standard deviation of the mean (mean ± SD). Error bars in the figures represent SD, unless specified. Statistical analysis was performed using PRISM statistical software (GraphPad, USA). Differences in the freezing time courses between the experimental groups (optogenetic actuators vs. control) were analyzed using a two-way ANOVA (referred to as “ANOVA” in the main text), followed by a Bonferroni post hoc test to compare replicate means by row (group effect), assuming the data were normally distributed for each time point. We additionally performed a pooled analysis of the freezing data by pooling the freezing levels of the 5th and 6th tone–footshock stimulation for the training day, and the n = 4 tone blocks from the retrieval day (Figs. 1D, 3D, 3G, and 6C). The statistical significance of this pooled analysis was tested with a Student’s t test (referred to as “t test” in the main text). This was done either after a Shapiro-Wilk normality test suggested normality of the respective distributions, or else after assuming the normal distribution if the sample size was not large enough for the normality test.

We also tested the homogeneity of variance with the F-test ( for the behavioral data in Figs. 1D, 3D, 3G, and 6C on days 2 and days 3. For a total of 40 time points in this data [resulting from 4× (6 + 4) time points in these datasets], we found that n = 33 time points had a homogeneous variance. Violation of the variance homogeneity in the remaining n = 7 time points should be tolerable because of almost identical sample numbers in the control group versus the halorhodopsin group.

P values <0.05 were considered to be significant. * indicates P < 0.05, ** indicates P < 0.01, and *** indicates P < 0.001.

Supplementary Materials

Figs. S1 to S19


Movies S1 to S3

References and Notes

Acknowledgments: We thank H. Murray, B. Lecrinier, and T. Baticle for technical assistance and help with virus constructs; Y. Sych (University of Zürich, Switzerland) for advice on the construction of optrodes; and O. Burri for custom codes for image analysis. Image acquisition was carried out in the Bioimaging and Optics Platform of EPFL (BIOP). Funding: This work was supported by the Swiss National Science Foundation (SNSF) (31003A_176332/1 to R.S.), the SNSF National Competence Center for Research Synapsy “The Synaptic Bases of Mental Diseases” (project 28 to R.S.), the German Research Foundation (DFG Priority Program 1608/SCHN 451/5-2 to R.S.), and an EMBO fellowship (ALTF 224-2015 to M.K.). Author contributions: E.B., M.K., O.K., and R.S. designed the research. E.B., W.T., and M.K. performed in vivo optogenetic experiments. S.P. and E.B. performed ex vivo optogenetic mapping experiments with slice electrophysiology. M.K. performed anatomical experiments and image analysis. O.K. designed measurement and analysis routines for in vivo optrode recordings; O.K. and D.O. performed in vivo optrode recordings. R.S., M.K., E.B., and O.K. wrote the paper. Competing interests: The authors declare no competing interests. Data and materials availability: All data needed to evaluate the conclusions of the study are present in the paper or the supplementary materials. AAV vectors were obtained under material transfer agreements with Stanford University, the Massachusetts Institute of Technology, and the University of North Carolina.
View Abstract

Stay Connected to Science

Navigate This Article