Intrinsic Excitability in Layer IV–VI Anterior Insula to Basolateral Amygdala Projection Neurons Correlates with the Confidence of Taste Valence Encoding

Visual Abstract


Introduction
In the natural setting, animals approach novel taste stimuli tentatively, as to closely examine them according to a genetic plan, as well as in relation to associated visceral consequences (Schier and Spector, 2019). Bitter and sour tastes are innately aversive, acting as warning signals for the presence of toxins (Bachmanov et al., 1996). Conversely, neophobia to innately appetitive sweet and moderately salty tastants dissipates over time (Lin et al., 2012). Importantly, animals can learn to avoid innately appetitive tastants [e.g., saccharin-, or NaCl-water; the conditioned stimulus (CS)], through conditioned taste aversion (CTA; Garcia et al., 1955;Nachman and Ashe, 1973). This single-trial associative learning paradigm results in robust aversion following the pairing of the CS with a malaise-inducing agent [the unconditioned stimulus (US)], such as LiCl (Bures et al., 1998). CTA memories are robust, but can be extinguished through unreinforced CS re-exposures, and subsequently reinstated through US re-exposure (Schachtman et al., 1985;Mickley et al., 2004). Unlike other forms of classical conditioning, the interstimulus interval (ISI) between taste experience (CS) and visceral outcome (US), extends to several hours (Adaikkan and Rosenblum, 2015). How CTA learning enables this long-trace associative process, within timeframes that deviate from classical Hebbian plasticity mechanisms is currently unknown (Chinnakkaruppan et al., 2014;Adaikkan and Rosenblum, 2015).
The primary taste cortex, the anterior insula (aIC), along with the basolateral amygdala (BLA), govern the encoding and retrieval of taste information (Piette et al., 2012;Bales et al., 2015). Gustatory processing in IC neurons encompasses thalamocortical and corticocortical inputs that relay taste-, as well as palatability-related inputs from the BLA, that reflect the emotional valence associated with taste stimuli (Stone et al., 2020). Neuronal taste responses at the IC and BLA are plastic and spatially dispersed, using temporal information to encode multiple types of information relating to stimulus identity and palatability (Grossman et al., 2008;Sadacca et al., 2012;Arieli et al., 2020;Vincis et al., 2020). Both synaptic plasticity and neuronal intrinsic properties are proposed to serve as cellular mechanisms underlying learning and memory (Citri and Malenka, 2008;Sehgal et al., 2013). CTA learning promotes LTP induction in the BLA-IC pathway (Jones et al., 1999;Juárez-Muñoz et al., 2017), and strengthens cell type-specific functional connectivity along the projection (Haley et al., 2016). Intrinsic excitability is the tendency of neurons to fire action potentials (APs) when exposed to inputs, reflecting changes in the suit and properties of specific ion channels (Disterhoft et al., 2004;Song and Moyer, 2018). Although independent mechanisms are involved, recent evidence indicates learning and memory necessitates the that coupling of intrinsic and synaptic plasticity (Turrigiano, 2011;Greenhill et al., 2015;C. H. Wu et al., 2021).
The IC is an integration hub tuned for the encoding of both exteroceptive as interoceptive information (Gogolla et al., 2014;Haley and Maffei, 2018;Livneh et al., 2020;Koren et al., 2021). By virtue of its extensive network of connectivity, this elongated cortical structure has been shown to integrate sensory, emotional, motivational, and cognitive brain centers through distinct mechanisms. For example, deletions of either Fos or Stk11 in BLA-aIC neurons, alter intrinsic properties at the aIC, and impair CTA acquisition (Levitan et al., 2020). Furthermore, approach behaviors in social decision-making are modulated by subjective and sex-specific affective states that regulate cell-type-specific changes in intrinsic properties at IC projections to the nucleus accumbens (Rogers-Carter et al., 2018Rieger et al., 2022). The posterior IC (pIC) integrates visceral-sensory signals of current physiological states with hypothalamus-gated amygdala anticipatory inputs relating to food or water ingestion, to predict future physiological states (Livneh et al., 2017(Livneh et al., , 2020. Conversely, aversive visceral stimuli such as LiCl, activate CaMKII neurons projecting to the lateral hypothalamus in right-, but not the left IC, whose optogenetic activation or inhibition can bidirectionally regulate food consumption (Y. Wu et al., 2020). We have previously shown that the aIC-BLA projection is necessary and sufficient for CTA acquisition and retrieval (Lavi et al., 2018;Kayyal et al., 2019), while CTA Retrieval requires activation of the projection concomitant with parvalbumin (PV) interneurons . Moreover, artificial activation of aIC-BLA projecting neurons is sufficient to induce CTA for appetitive taste (Kayyal et al., 2019). Here, using retrograde viral tracing, behavioral analysis, and whole-cell patch-clamp slice electrophysiology, we assessed two hypotheses: (1) that the intrinsic properties of the aIC-BLA projection change as a function of certainty of taste valence prediction, but not percept; and (2) that predictive valence-specific changes in intrinsic properties would be reflected through excitability, being low when taste outcome is highly predictive (i.e., following CTA Retrieval or unreinforced familiarization), and high when taste valence is uncertain (i.e., following novelty or extinction). Our data demonstrate for the first time that the intrinsic properties of LIV-VI aIC-BLA neurons are differentially regulated by innate and learned drives, reflecting the confidence of currently perceived taste valence.

Animals
Animals used were 8-to 12-week-old C57BL/6j [wild type (WT)] adult male mice. Mice were kept in the local animal resource unit at the University of Haifa on a 12/12 h light/dark cycle. Water and chow pellets were available ad libitum, while ambient temperature was tightly regulated. All procedures conducted were approved by the University of Haifa Animal Care and Use Committee (Ethics License 554/18), as prescribed by the Israeli National Law for the Protection of Animals-Experiments with Animals (1994).

Animal surgery and viral injections
Following surgery and stereotactic injection of viral vectors, behavioral paradigms were performed, as previously described . Briefly, mice were treated with norocarp (0.5 mg/kg), before being anesthetized (M3000 NBT Israel/Scivena Scientific) and transferred to a Model 963 Kopf stereotactic device. Upon confirming the lack of pain responses, the skull was surgically exposed and drilled to bilaterally inject 0.25 ml of ssAAV_retro2-hSyn1-chi-mCherry-WPRE-SV40p(A) (physical titer 8.7 Â 10E12 vg/ml), at the BLA (AP À1.58; ML 63.375; DV À4.80). Viral delivery was performed using a Hamilton micro-syringe (0.1 ml/min), while the sculp was cleaned and closed using Vetbond. Animals were then administered with 0.5 mg/kg norocarp and 0.5 mg/kg of Baytril (enrofloxacin), and then transferred to a clean and heat-adjusted enclosure for 2 h. Upon inspection, mice were returned to fresh cages along with similarly treated cage-mates. Weight-adjusted doses of the Norocarp and Baytril were administered for an additional 3 d. All AAV constructs used in this study were obtained from the Viral Vector Facility of the University of Zurich (http://www. vvf.uzh.ch/).
Electrophysiological studies of the influence of innate taste identity, novelty, and valence on aIC-BLA excitability WT mice treated with viral constructs labeling aIC-BLA projecting neurons were used for electrophysiological studies. Upon recovery, mice were randomly assigned into treatment groups (Fig. 1). Following 24 h of water deprivation, animals were water restricted for 3 d, receiving water in pipettes ad libitum for 20 min/d (Kayyal et al., 2019;Yiannakas et al., 2021). This regime has been extensively used by our lab as it allows rodents to reliably learn to drink from water pipettes with minimal weight loss. Mean total drinking was recorded on the third day of water restriction. Novel taste consumption groups were presented with 1.0 ml of either 0.5% saccharin (Saccharin 1x), or Quinine 0.014% (Quinine 1x). One hour following the final taste presentation, animals were subjected to patch-clamp electrophysiology . The Water group underwent the same behavioral procedure without novel taste presentations were killed for electrophysiological investigations 1 h following water presentation. To dissociate between taste identity and familiarity-related changes in electrophysiological properties, a cohort of mice treated to label the aIC-BLA projection were similarly water deprived following familiarization with saccharin (Saccharin 5x). Following the initial water restriction, Saccharin 5x animals were allowed access to 0.5% saccharin, in 20min sessions for 4 d. On the fifth day, mice were provided with 1.0 ml of the tastant, 1 h before killing for electrophysiological recordings. Additionally, WT animals injected with the same viral vector, were allowed a month to recover, following which they were killed for electrophysiological investigations without any behavioral manipulation (Cage Controls). Retrieval of appetitive and novel taste increases excitability in LIV-VI aIC-BLA projection neurons. A, Diagrammatic representation of experimental procedures. Following surgery and stereotaxic delivery of ssAAV_retro2-hSyn1-chi-mCherry-WPRE-SV40p(A) into the BLA, mice were allowed four weeks of recovery. Animals were subsequently assigned to treatment groups and trained to drink from pipettes (see Materials and Methods). We compared the intrinsic properties of LIV-VI aIC-BLA neurons among the Water (n = 6 animals, 23 cells), Saccharin 1x (n = 5 animals, 20 cells), Saccharin 1x (4 h) (n = 4 animals, 17 cells), Saccharin 5x (n = 6 animals, 18 cells), and Quinine 1x groups (n = 4 animals, 19 cells), as well as a Cage Control group (n = 4 animals, 19 cells) that underwent surgery and stereotaxic delivery of ssAAV_retro2-hSyn1-chi-mCherry-WPRE-SV40p(A) at the BLA without water restriction. B, Graph showing the water consumption before treatment (mean 6 SD). There was no significant difference between water intakes between the groups before the treatment. One-way ANOVA, p = 0.9766. C, Representative traces of LIV-VI aIC-BLA projecting neurons from the six treatment groups. Scale bars: 20 mV vertical and 50 ms horizontal from 300-pA step. D, The dependence of firing rate on current step magnitude in LIV-VI aIC-BLA neurons was significantly different among the treatment groups. Excitability in the Saccharin 1x was increased compared with all other groups. Two-way repeated measures ANOVA, Electrophysiological studies of the influence of learned aversive taste memory retrieval on aIC-BLA excitability WT mice were treated with viral constructs labeling aIC-BLA projecting neurons to assess the electrophysiological properties of the projection during aversive or appetitive taste memory retrieval. Upon recovery, mice in CTA Retrieval group were trained in CTA for saccharin (LiCl 0.14 M, 1.5% body weight), while the appetitive saccharin retrieval group (Saccharin 2x) received a matching body weight adjusted injection of saline . Three days following conditioning, both groups underwent a memory retrieval task, receiving 1.0 ml of the conditioned tastant 1 h before killing (Figs. 2, 3 and 4). Brain tissue was extracted and prepared for electrophysiological recording, as above.
Electrophysiological studies of the influence of learned aversive taste memory extinction and reinstatement on aIC-BLA excitability Electrophysiological studies of CTA extinction and reinstatement were conducted in a cohort of WT male mice . Following surgery, recovery and water restriction, animals were randomly assigned to the extinction and reinstatement groups (Figs. 3, 4). The aversion index for the extinction and reinstatement groups were calculated by the formula: Aversion index ¼ Volume of water volume of ðwater þ tastantÞ Ã 100: Adult male mice used to study extinction and reinstatement were trained in CTA for saccharin following extinction, the reinstatement group received an identical intraperitoneal dose to the original unconditioned stimulus (LiCl 0.14 M, 1.5% body weight), 24 h before retrieval. Conversely, the extinction group received a similarly weight-adjusted dose of saline. During the final retrieval session, both groups of mice were allowed access to 1.0 ml of the CS, 1 h before killing under deep anesthesia and slice preparation for electrophysiology.
The recordings were made from the soma of insula pyramidal cells, particularly from layer 2/3 and layer 5/6. Liquid junction potential (10 mV) was not corrected online. All current clamp recordings were low pass filtered at 10 kHz and sampled at 50 kHz. Pipette capacitance and series resistance were compensated and only cells with series resistance smaller than 20 MX were included in the dataset. Data quantification was done with Clampfit (Molecular Devices) and subsequently analyzed using GraphPad Prism. The method for measuring active intrinsic properties was based on a modified version of previous protocols (Kaphzan et al., 2013;Chakraborty et al., 2017;Sharma et al., 2018).

Recording parameters
Resting membrane potential (RMP) was measured 10 s after the beginning of whole-cell recording (rupture of the membrane under the recording pipette). The dependence of firing rate on the injected current was obtained by injection of current steps (of 500-ms duration from 0 to 400 pA in 50-pA increments). Input resistance was calculated from the voltage response to a hyperpolarizing current pulse (À150 pA). SAG ratio was calculated from voltage response À150 pA. The SAG ratio during the hyperpolarizing steps was calculated as [(1 -DV SS /DV max ) Â 100%] as previously reported by (Song et al., 2015). The membrane time constant was determined using a single exponential fit in the first 100 ms of the raising phase of cell response to a 1 s, À150-pA hyperpolarization step.
For measurements of a single action potential (AP), after initial assessment of the current required to induce an AP at 15 ms from the start of the current injection with large steps (50 pA), a series of brief depolarizing currents were injected for 10 ms in steps of 10-pA increments. The first AP that appeared on the 5-ms time point was analyzed. A curve of dV/dt was created for that trace and the 30 V/s point in the rising slope of the AP was considered as threshold (Chakraborty et al., 2017). AP amplitude was measured from the equipotential point of the threshold to the spike peak, whereas AP duration was measured at the point of half-amplitude of the spike. The medium after-hyperpolarization (mAHP) was measured using prolonged (3 s), high-amplitude (3 nA) somatic current injections to initiate time-locked AP trains of 50 Hz frequency and duration (10 -50 Hz, 1 or 3 s) in pyramidal cells. These AP trains generated prolonged (20 s) AHPs, the amplitudes and integrals of which increased with the number of APs in the spike train. AHP was measured from the equipotential point of the threshold to the anti-peak of the same spike (Gulledge et al., 2013). Fast (fAHP), and slow AHP (sAHP) measurements were identified as previously described (Andrade et al., 2012;Song and Moyer, 2018). Series resistance, Rin, and membrane capacitance were monitored during the entire experiment. Changes of at least 30% in these parameters were criteria for exclusion of data.

Classification of burst and regular spiking neurons
At the end of recordings, neurons were classified as either burst (BS) or regular spiking (RS) as reported previously (Kim et al., 2015;Song et al., 2015). Briefly, neurons that fired two or more action potentials (doublets or triplets) potential toward a depolarizing current step above the spike threshold current were defined as burst spiking (BS). Regular spiking (RS) neurons on the other hand, were defined as neurons that fired single action potential in response to a depolarizing current step above spike threshold (Extended Data Fig. 1-2A).

Statistical analysis of individual intrinsic properties across treatments
Group size was based on previously published results using similar methods Kayyal et al., 2021), as well as through conducting power analysis calculations, to obtain power !0.8 with a = 0.05 (https:// www.stat.ubc.ca/;rollin/stats/ssize/n2.html). Individual intrinsic properties of aIC-BLA projecting neurons in the respective treatment groups (Figs. 1-4) were analyzed using appropriate statistical tests (one-way or two-way ANOVA, GraphPad Prism), as defined in the Table 5, statistics table. Two-way repeated measurements of ANOVA (RM-ANOVA) followed by Sídák's (for two groups) or Tukey's (for more than two groups) post hoc multiple comparison test was performed for firing properties. The intrinsic properties were determined with two-tailed unpaired t tests, and oneway ANOVA followed by Tukey's or Dunn's multiple comparisons test were used. For all tests, *p , 0.05 was considered significant. D'Agostino and Pearson test used for the identifying the normal distribution of the data. Multiple comparisons were corrected post hoc with Tukey's for one-way/two-way ANOVA and Dunn's for Kruskal-Wallis test.
Following spike-sorting, the ratio of BS:RS aIC-BLA projecting neurons in the sampled population was compared across our treatments (Mann-Whitney test, GraphPad Prism). Similarly, individual intrinsic properties in BS and RS aIC-BLA projecting neurons were analyzed following spike-sorting (one-way or two-way ANOVA, GraphPad Prism). All data reported as mean 6 SEM.

Immunohistochemistry
From each electrophysiological recording, three 300mm-thick mouse brain slices were obtained starting from continued group was increased compared with Saccharin 2x (49.14 6 1.568 mV), p = 0.0005, Mann-Whitney test. H, Input resistance in the CTA Retrieval group (136.4 6 9.064 MX) was significantly decreased compared with Saccharin 2x (181.1 6 11.7 MX). p = 0.0036, Unpaired t test. I, SAG ratio following CTA Retrieval (13.41 6 1.31) was significantly enhanced compared with Saccharin 2x (7.815 6 1.176). p = 0.0037, Unpaired t test. Data are shown as mean 6 SEM *p , 0.05, **p , 0.01, ***p , 0.001, ****p , 0.0001. Learned aversive taste memory retrieval decreases the excitability of burst spiking LIV-VI aIC-BLA neurons (see Extended Data Fig. 2-1). Figure 3. Extinction of CTA enhances, whereas reinstatement decreases, the excitability of LIV-VI aIC-BLA projecting neurons. A, Experimental design of behavioral procedures conducted to compare the intrinsic properties of LIV-VI aIC-BLA neurons following CTA Extinction (n = 5, animals, 14 cells) and Reinstatement (n = 3 animals, 15 cells). B, The graph showing the reduced aversion following the successful extinction in both treatment groups. C, Data showing the saccharin consumption on the test day following successful extinction and Reinstatement of CTA. CTA reinstated mice showed significantly reduced saccharin consumption bregma coordinates 1.78, 1.54, and 1.18, respectively. Slices were washed with PBS and fixed using 4% paraformaldehyde in PBS at 4°C for 24 h. Slices were then transferred to 30% sucrose/PBS solution for 48 h and mounted on glass slides using Vectashield mounting medium with DAPI (H-1200). Slides were then visualized using a vertical light microscope at 10Â and 20Â magnification (Olympus CellSens Dimension). Images were processed using Image-Pro Plus V-7 (Media Cybernetics). The localization of labeled mCherry1 neurons in the agranular aIC, where recordings were obtained from, was quantified manually across three bregma-matched slices, for each animal. Quantification was done using randomly assigned IDs for individual animals, regardless of treatment. Representative images were additionally processed using the Olympus CellSens 2-D deconvolution function.  Fig.  5A,B). The adequacy of the sample was evaluated using the Bartlett's test and the Kaiser-Meyer-Olkin (KMO) measure was applied. The degrees of freedom (df) were calculated using the following formula: df ¼ ] variables À 1: The number of principal components was chosen according to the percentage of variance explained (.75%). The parallel analysis evaluated the optimal number of components and selected three PCs, explaining 62.47% of the variance. Oblique factor rotation (par) of the first three PCA components, using a standard "rotatefactors" routine from MATLAB Statistics Toolbox. This approach maximizes the varimax criterion using an orthogonal rotation. To optimize variance, oblique factor rotation (paramax) was used, and the threshold chosen to define a variable as a significant contributor was a variance !0.7 given the small sample size. The correlation matrix was adequate as the null hypothesis of all zero correlation was rejected (x 66 2 = 387.444, p , 0.001), and KMO exceeded 0.5 (KMO = 0.580).
To calculate the proportion of the variance of each variable that the principal components can explain, communalities were calculated and ranged from 0.426 to 0.897 (Extended Data Fig. 5-1A-C). The communalities scores were calculated using the following for- ; where m is the number of selected PCs. The threshold chosen was Comm ! 60%. Because of the imbalance in sample sizes between groups, the PCA space is biased in favor of the group with bigger sample size. The BS neurons in the six behavioral groups previously mentioned were resampled to ensure that sample sizes were balanced across groups datasets (Fig.  5A,B). Particularly, we reduced the number of Saccharin 1x and 2x observations by using random sampling ("randn" function in MATLAB); for Saccharin 1x, we chose 10 of the 17 total elements, and for Saccharin 2x, we selected 10 of the 13 total elements.

K-means clustering
Unbiased clustering analysis based on the K-means algorithm was conducted to search for an optimal division of samples into a predetermined number of clusters within an unlabeled multidimensional data set (MacQueen, 1967). In the present study we chose K-means clustering methods to cluster the similar groups into a high and low predictive valance outcome following the memory, based on their intrinsic properties. As a popular method for a cluster analysis, K-means clustering aims to partition n observations into k-clusters in which each observation belongs to the cluster with the nearest mean, serving as a prototype of a cluster. In this classification a set of clustering results with different number of clusters can be calculated by setting a different k. The final step is to determine the optimal dimension which gives the best classification. To assess the distribution of bursting neurons in multidimensional space, we performed a k-means cluster analysis in MATLAB for the principal components (k = 2 clusters, maximum continued compared with extinguished mice. p = 0.0179, Mann-Whitney test. D, Representative traces of LIV-VI aIC-BLA projection neurons firing from the two treatment groups. Scale bars: 20 mV and 50 ms horizontal from 300-pA step. E, Excitability in LIV-VI aIC-BLA neurons was significantly different among the treatment groups. Two-way repeated measures ANOVA, Current Â Treatment: p , 0.0001. F, Action potential threshold in the Reinstatement group (À29.43 6 1.731 mV) was enhanced compared with Extinction (À36.06 6 1.481 mV). p = 0.0076, Unpaired t test. G, The membrane time constant following Reinstatement (25.48 6 1.58 ms) was significantly enhanced compared with Extinction (17.55 6 2.684 ms, p = 0.047). p = 0.0153, Unpaired t test. For panels D-F: *p , 0.05, **p , 0.01, ***p , 0.001, ****p , 0.0001. All data are shown as mean 6 SEM. Extinction of CTA enhances the excitability of burst spiking LIV-VI aIC-BLA projecting neurons compared with CTA reinstatement (see Extended Data Fig. 3-1).

Data availability
All data generated or analyzed during this study are included in the manuscript and supporting files. Source data files have been provided for all figures as Extended Data 1.

Results
To prove or refute our hypotheses, we conducted a series of electrophysiological recordings in slice preparation from the mouse aIC, in which we labeled aIC-BLA projecting neurons using retrograde adeno-associated viral tracingretro AAV (see Materials and Methods). Electrophysiological recordings were obtained from the aIC-BLA projecting neurons in LI-III (Table 1) and LIV-VI (Table 2), following novel appetitive or aversive taste stimuli ( Fig. 1), following appetitive or aversive taste memory retrieval (Fig. 2), as well as following extinction and reinstatement (Figs. 3,4). We focused on deep-lying LIV-VI aIC-BLA projecting neurons (Table 2), as intrinsic properties in superficial LI-III aIC-BLA projecting neurons were unaffected by taste identity, familiarity, or valence (Table 1). We measured action potential (AP) firing frequency in response to incrementally increasing depolarizing current injections, as well as 11 distinct of intrinsic properties (Tables 2-Table 4 and 5, statistics table): resting membrane potential (RMP); slow, medium, and fast after-hyperpolarization (sAHP, mAHP, fAHP); input resistance (IR), SAG ratio; the amplitude, half-width and threshold for APs; the time taken for a change in potential to reach 63% of its final value (membrane time constant, t ), as well as the minimum current necessary for AP generation (Rheobase). Statistical analysis was conducted using repeated measures one-way or two-way ANOVA (see Table 5, statistics table).

When taste is both appetitive and novel, excitability in LIV-VI aIC-BLA projection neurons is increased
To delineate the mechanisms through which novelty is encoded on the LIV/VI aIC-BLA projection, we labeled the projection (see Materials and Methods), and compared the intrinsic properties across neutral, innately aversive, and innately appetitive taste stimuli. Following surgery recovery mice were randomly assigned to the following behavioral groups: Water (Control for procedure, Water; n = 6 animals, 23 cells), 0.5% Saccharin for the first (Novel innate appetitive, Saccharin 1x; n = 5 animals, 20 cells), or fifth time (Familiar appetitive, Saccharin 5x; n = 6 animals, 18 cells), 0.04% Quinine (Novel aversive, Quinine 1x; n = 4 animals, 19 cells), or Cage Controls that did not undergo water-restriction (Baseline control, Cage control; n = 4 animals, 19 cells). This approach allowed us to examine excitability changes that relate to the innate aversive/appetitive nature and novelty/familiarity associated with tastants, while accounting for the effects of acute drinking, as well as the water restriction regime itself. Guided by evidence regarding the induction of plasticity cascades, the expression of immediate early genes, as well as the timeframes involved in LTP and LTD in IC neurons (Rosenblum et al., 1997;Hanamori et al., 1998;Jones et al., 1999;Escobar and Bermúdez-Rattoni, 2000), the five treatment groups were killed 1 h following taste consumption. Although changes in activity can be observed within seconds to minutes, depending on their novelty, salience, and valence (Barot et al., 2008;Lavi et al., 2018;Y. Wu et al., 2020), sensory experiences can modulate the function of IC neurons for hours (Juárez-Muñoz et al., 2017;Rodríguez-Durán et al., 2017;Haley et al., 2020;Yiannakas et al., 2021). We had previously identified a CaMKII-dependent short-term memory trace at the IC that last for the first 3 h following taste experiences, regardless of their valence (Adaikkan and Rosenblum, 2015). To address whether similar time-dependency of the physiological correlations engaged by the IC during novel taste learning, a sixth group was killed 4 h following novel saccharin exposure [ Fig. 1, Saccharin 1x (4 h)].
Daily water intake before the final taste exposure and was not different among the five groups that underwent water restriction (Fig. 1B, one-way ANOVA, p = 0.4424, F = 0.9766, R 2 = 0.1634). However, excitability in response to incremental depolarizing currents was significantly different between the six groups ( Fig. 1D; two-way ANOVA, p , 0.0001, F (8,880) = 1269). Exposure to saccharin for the first time (i.e., novel appetitive), at the 1-h time point, resulted in enhanced excitability on the aIC-BLA projection compared with all other groups ( Fig. 1D; see Table 2). Conversely, fAHP ( Fig. 1H; one-way ANOVA, p , 0.0001, F = 8.380, R 2 = 0.2758) in the Quinine 1x group was increased compared with all other groups, in contrast to Saccharin 1x where it was most decreased (Table 2). In fact, fAHP in the Saccharin groups recorded at 1 h (p , 0.0001, z = 5.150) or 4 h (p = 0.0099, z = 3.406) following novel taste consumption was decreased compared with innately aversive Quinine 1x (Fig. 1H). Although fAHP in the Saccharin 1x group was decreased compared continued in the Quinine 1x group (57.11 6 1.376 mV), and CTA Retrieval (56.21 6 0.9978 mV), was significantly increased compared with Saccharin 2x (49.14 6 1.568 mV, p = 0.0175 and 0.0229, respectively). Conversely, action potential attitude in the Saccharin 1x (52.03 6 1.308 mV), Extinction (55.09 6 2.122 mV), and Reinstatement (53.1 6 2.906 mV) groups was similar. One-way ANOVA, p = 0.0061. F, Action potential half-width following Extinction (0.7386 6 0.03145 ms) and Reinstatement (0.8187 6 0.06929 ms) was elevated compared with Saccharin 1x (0.6005 6 0.03260 ms), Saccharin 2x (0.5780 6 0.02994 ms) as well as CTA Retrieval (0.5959 6 0.02080 ms, but no with Quinine 1x (0.6300 6 0.03555 ms). One-way ANOVA, p = 0.0002. G, The membrane time constant in the Saccharin 1x (14.82 6 1.485 ms) group was significantly decreased compared with Reinstatement (25.48 6 1.58 ms, p = 0.0043) groups was. Differences between CTA Retrieval (20.96 6 1.724 ms, p = 0.0189), Quinine 1x (21.55 6 1.638 ms), Saccharin 2x (19.28 6 1.837 ms), and Extinction (17.55 6 2.684 ms) groups failed to reach significance. One-way ANOVA, p = 0.0047. Figure 5. The intrinsic properties of burst spiking LIV-VI aIC-BLA projecting neurons represent taste experience and the probability for further learning. A, Data across all intrinsic properties from BS LIV-VI aIC-BLA neurons of the Saccharin 1x, Saccharin 2x, and Extinction groups were combined and assigned to the Low predictive following memory group (32 BS cells). Conversely, the intrinsic properties of BS LIV-VI aIC-BLA neurons from animals having undergone CTA Retrieval, 5Â Saccharin, and Reinstatement were combined and assigned to the High predictive following memory group (31 BS cells). The resultant three-dimensional scatter representation of the two groups with both the Cage Control (p = 0.0136, z = 3.318) and Water (p = 0.0177, z = 3.243) groups, this was not the case for the Saccharin 1x (4 h) group (p . 0.9999 for both; Table 2). Importantly, fAHP (Fig. 1H) was nearly identical in treatment groups where the tastant could be deemed as highly familiar and safe, such as the Cage Control group (that did not undergo water restriction), as well as animals in the Water or Saccharin 5x groups (that had undergone water restriction).
These results demonstrate that in the context of taste novelty, innately appetitive saccharin drove increases in excitability and decreases in fAHP of LIV-VI aIC-BLA projecting neurons, compared with innately aversive quinine (Fig. 1D,H). Compared with the Cage Control and Water groups, fAHP on the projection was significantly enhanced by innately aversive quinine and was decreased by innately appetitive novel saccharin (Fig. 1H). However, the effect of appetitive taste novelty on firing frequency was time dependent, as it was observed at 1 h, but not 4 h, following novel taste exposure (Fig. 1D). Furthermore, following familiarity acquisition for saccharin (Saccharin 5x), excitability was decreased compared with Saccharin 1x, matching the Cage Control, Water, and Quinine 1x groups (Fig. 1D). This led us to consider whether increased excitability is not related to taste identity or palatability (Wang et al., 2018), but the perceived salience of taste experiences, which encompasses both novelty and valence (Ventura et al., 2007;Kargl et al., 2020). Previous studies have suggested that the induction of plasticity signaling cascades and IEGs in pyramidal neurons of the aIC (commonly used as surrogates for changes in excitability), is a crucial step for the association of taste and visceral information during CTA learning (Adaikkan and Rosenblum, 2015;Soto et al., 2017;Y. Wu et al., 2020). Activation of the aIC-BLA projection is indeed necessary for the expression of neophobia toward saccharin , as well as for CTA learning and retrieval (Kayyal et al., 2019). Yet, its chemogenetic inhibition does not affect the attenuation of neophobia, nor the expression of aversion toward innately aversive quinine (Kayyal et al., 2019). Furthermore, aversive taste memory retrieval necessitates increases in presynaptic inhibitory input on the projection . Bearing this in mind, we hypothesized that increases in excitability on the projection could be indicative of a labile state of the taste trace at the aIC, which manifests when taste cues are not (yet) highly predictive of the visceral outcome of the sensory experience (Bekisz et al., 2010;Galliano et al., 2021). In such a scenario, taste memory retrieval following strong single-trial aversive learning would be expected to result in decreased excitability compared with control animals. To assess this hypothesis, we next examined intrinsic excitability in mice retrieving an appetitive (Saccharin 2x, CTA Retrieval control) or learned aversive memory (CTA Retrieval) for saccharin.
Learned aversive taste memory retrieval decreases the excitability of LIV-VI aIC-BLA projecting neurons Following recovery from rAAV injection, mice in the CTA Retrieval group underwent water restriction and CTA conditioning for 0.5% saccharin (see Materials and Methods; Fig. 2A). Electrophysiological recordings were obtained from aIC-BLA neurons 3 d later, 1 h following retrieval (n = 8 animals, 27 cells). Mice in the Saccharin 2x group on the other hand, were familiarized with saccharin without conditioning, and recordings were obtained within the same period, following retrieval (n = 5 animals, 20 cells).   (17) Values are expressed as mean 6 SEM. The number of cells is in parentheses. Statistical analysis was performed by Student's t test. RMP, resting membrane potential; mAHP, medium after hyperpolarization potentials; AP Thresh, action potential threshold; AP Amp, action potential amplitude; AP half-width, action potential half-width.
Through this approach we aimed to examine the hypothesis that like innately aversive and highly familiar appetitive responses ( Fig. 1), learned aversive taste memory retrieval would be correlated with suppression of the intrinsic excitability on the projection. As expected, CTA Retrieval mice, exhibited decreased consumption of the conditioned tastant compared with control animals that were only familiarized with saccharin ( Fig.  2B, Mann-Whitney test, p = 0.0085; Sum of ranks: 52.50, 38.50; Mann-Whitney U = 2.500). Intrinsic excitability in LIV-VI aIC-BLA projecting neurons was increased in response to depolarizing current injections ( Fig. 2D; p , 0.0001, F (8,360) = 483.3), and was significantly different between the two treatments (p = 0.0014, F (1,45) = 11.60). Excitability was enhanced in the Saccharin 2x group compared with CTA Retrieval, while a significant interaction was identified between the treatment and current injection factors (p , 0.0001, F (8,360) = 9.398). Fast AHP (fAHP) on LIV-VI aIC-BLA projecting neurons tended to be increased in the CTA Retrieval group (Table 2), however differences compared with Saccharin 2x failed to reach significance (unpaired t test; p = 0.0527, t = 1.990, df = 45).
Conversely, AP amplitude in the Saccharin 2x group was significantly decreased compared with CTA Retrieval ( Fig. 2G; Unpaired t test; p = 0.0002, t = 3.983, df = 45). In addition, the CTA Retrieval group exhibited significantly decreased IR ( Fig. 2H; unpaired t test; p = 0.0036, t = 3.072, df = 45) and significantly enhanced SAG ratio ( Fig. 2I; unpaired t test; p = 0.0037, t = 3.060, df = 45), compared with Saccharin 2x. In accord with our hypothesis, excitability on LIV-VI aIC-BLA projecting neurons was decreased by aversive taste memory retrieval. We have previously shown that compared with CTA Retrieval and Reinstatement, appetitive memory retrieval and extinction were associated with (1) an enhancement of IEG induction (c-fos and Npas4) at the aIC, and (2) decreased frequency of presynaptic inhibition on the aIC-BLA . In accord, other published work investigating the induction of IEG in the rodent IC, found that consistent with a reduction in spiking activity (Grossman et al., 2008), the induction of c-fos in IC neurons was decreased by aversive taste memory retrieval (Haley et al., 2020). Earlier studies have also reported increases in c-fos following the extinction of cyclosporine A-induced CTA (Hadamitzky et al.,  Values are expressed in mean 6 SEM. The number of cells is in parentheses. Statistical analysis was performed by one-way ANOVA post hoc Tukey's and Dunn's multiple comparisons. Student's t test was performed for the comparison between two groups. RMP, resting membrane potential; fAHP, mAHP and sAHP, fast, medium, and slow after hyperpolarization potentials, respectively; AP Thresh, action potential threshold; AP Amp, action potential amplitude; AP halfwidth, action potential half-width. Data are shown as mean 6 SEM *p , 0.05, **p , 0.01, ***p , 0.001, ****p , 0.0001, with respect to the corresponding symbols. $Vs. Cage Control *Vs. Saccharin 1x^Vs. Saccharin 2x ;Vs. Reinstatement. %Vs. Water #Vs. Quinine 1x &Vs. Saccharin 5x. 2015). We thus hypothesized that if excitability in these cells serves as key node for a change in valence prediction, extinction, which constitutes a form of appetitive re-learning, would be associated with enhanced excitability compared with CTA Retrieval and reinstatement (Berman, 2003;Suzuki et al., 2004;Morrison et al., 2016;Slouzkey and Maroun, 2016). In addition, through these extinction and reinstatement studies, we were able to examine the real-life relevance of these changes on intrinsic excitability, in a context where behavioral performance reflects the balance between contrasting memories and the availability of retrieval cues (Fig. 3).
The predictability of the valence arising from taste experiences determines the profile of intrinsic properties of LIV-VI aIC-BLA projecting neurons Using similar approaches, electrophysiological recordings were obtained from LIV-VI aIC-BLA projecting neurons from mice having undergone unreinforced CTA extinction (Extinction; n = 5, 14 cells), or US-mediated CTA reinstatement (Reinstatement; n = 3 animals, 15 cells). Behaviorally, the two groups of animals were similar in terms of their aversion profile over 9 unreinforced extinction sessions ( Fig. 3B; two-way ANOVA; Extinction: p , 0.0001, F (8,54) = 13.44; Treatment: p = 0.0681, F (1,54) = 3.466; Interaction: p = 0.9697, F (8,54) = 0.2803). As expected, saccharin consumption during the test day in the Reinstatement group was decreased compared with Extinction ( Fig. 3C; Mann-Whitney test; p = 0.0179; Sum or ranks: 30, 6; Mann-Whitney U = 0). Consistent with our findings in Figure 2, aversive taste memory retrieval in the Reinstatement group was associated with decreased excitability compared with the Extinction group ( Fig. 3E; two-way ANOVA, Current injection: p , 0.0001, Unlike animals that underwent familiarization with the tastant without conditioning (Fig. 1), excitability on the projection in the Extinction group was not decreased  Values are expressed in mean 6 SEM. The number of cells is in parentheses. Statistical analysis was performed by one-way ANOVA post hoc Tukey's and Dunn's multiple comparisons. Student's t test was performed for the comparison between two groups. RMP, resting membrane potential; fAHP, mAHP, and sAHP, fast, medium, and slow after hyperpolarization potentials, respectively; AP Thresh, action potential threshold; AP Amp, action potential amplitude; AP halfwidth, action potential half-width. *BS LV/VI aIC-BLA Saccharin 2x vs. CTA Retrieval, *p , 0.05, **p , 0.01, ***p , 0.001. ;BS LV/VI aIC-BLA Extinction vs. Reinstatement, ;p , 0.05, ;;p , 0.01, ;;;p , 0.001. #BS LV/VI aIC-BLA Saccharin 1x vs. Quinine 1x, Saccharin 5x and Saccharin 1x (4hr), #p , 0.05, ##p , 0.01. by familiarization (Fig. 3). Conversely, although the intrinsic mechanisms employed would appear to differ, aversive taste memory retrieval regardless of prior experience, was associated with baseline excitability of the aIC-BLA projection (Fig. 3). Our findings in this section (Fig. 3), revealed that during taste memory retrieval, excitability on the projection is not solely dependent on the relevant novelty or appetitive nature of tastants, and does not subserve the persistence of CTA memories (Fig. 2). Instead, excitability on the aIC-BLA projection is indeed shaped by prior experience but is best predicted by the probability for further aversive (re)learning.
The predictability of taste valence intrinsic is primarily reflected on the excitability of burst spiking, but not regular spiking LIV-VI aIC-BLA projecting neurons Our initial analysis of individual intrinsic properties (Figs. 1-3) highlighted that excitability is enhanced following appetitive experiences in which the internal representation is still labile and is associated with the possibility for further aversive learning (novelty or extinction). Conversely, following extensive familiarization, aversive conditioning, or reinstatement, whereby taste exposure leads to memory retrieval of specific valence, excitability on LIV-VI aIC-BLA projecting neurons was similar to baseline (Figs. 1,  4). While the precise mechanism through which sensory input is encoded at the cortex (and other key regions), is still a matter of ongoing research, studies indicate that bursting in cortical Layer V pyramidal neurons can encode oscillating currents into a pattern that can be reliably transmitted to distant postsynaptic terminals (Kepecs and Lisman, 2003;Samengo et al., 2013;Zeldenrust et al., 2018). Spike burst is defined as the occurrence of three or more spikes from a single neuron with ,8-ms intervals (Ranck, 1973;Connors et al., 1982). In brain slices from naive mice, half of the neurons of a given structure exhibit burst firing, while the distribution of burst spiking (BS) to regular spiking (RS) neurons, changes along the anteriorposterior axis of the subiculum (Staff et al., 2000;Jarsky et al., 2008). Importantly, the two cell types fine-tune the output of brain structures by virtue of differences in synaptic plasticity, as well as intrinsic excitability mechanisms (Graves et al., 2012;Song et al., 2012). Furthermore, there are changes in the ratio of BS:RS neurons in individual brain structures, as well as differences in the recruitment of signaling events, ion channels and metabotropic receptors among the two cell types (Wozny et al., 2008;Shor et al., 2009). Correspondingly, complex region and task-specific rules govern the molecular and electrophysiological mechanisms through which information encoding and retrieval takes place in the two cell types (Dunn et al., 2018;Dunn and Kaczorowski, 2019). Little is currently known regarding the influence of cell identity in the repertoire of plasticity mechanisms employed by the IC to facilitate taste-guided behaviors (Maffei et al., 2012;Haley and Maffei, 2018).
Our post hoc spike sorting analysis allowed us to distinguish between BS and RS LIV-VI aIC-BLA projecting neurons, and thus their relative contribution to behaviorally driven changes in the suit of intrinsic properties (Extended Data Figs. 1-3, 2-1, 3-1). Through this comparison, we uncovered that Saccharin 1x differed to other groups in terms of excitability and fAHP in BS LIV-VI aIC-BLA neurons (Extended Data Fig. 1-3; Table 3), while no such changes were observed in RS neurons (see summary of RS intrinsic properties; Table 4). Similarly, excitability in the Saccharin 2x group was significantly enhanced compared with CTA Retrieval in BS, but not in RS, LIV-VI aIC-BLA neurons (Extended Data Fig. 2-1A,F). Significant differences in IR, SAG ratio and AP amplitude between CTA Retrieval and Saccharin 2x were primarily driven by BS LIV-VI aIC-BLA neurons (Extended Data Fig. 2-1B-D, G-I). Conversely, significant differences in AP half-width between the aversive and appetitive memory retrieval groups were only observed in RS neurons (Extended Data Fig. 2-1J). Correspondingly, excitability in the Extinction group was enhanced compared with Reinstatement in BS, and not RS, LIV-VI aIC-BLA neurons (Extended Data Fig.  3-1A,H). Indeed, excitability on BS LIV-VI aIC-BLA neurons following extinction and reinstatement, reflected the subjective predictability of taste memory retrieval, being high following extinction compared with reinstatement (Extended Data Fig. 3-1). However, this effect was mediated through alternative mechanisms compared with single-trial learning and memory retrieval (Extended Data Figs. 1-3, 2-1, 3-1). Significant differences between the Extinction and Reinstatement groups, were observed in terms of the sAHP, AP threshold, SAG ratio, and t in BS but not in RS LIV-VI aIC-BLA neurons (Extended Data Fig. 3-1B-F).
Encouraged by these findings, we focused on the Saccharin 1x, Saccharin 2x, Saccharin 5x, CTA Retrieval, Extinction, and Reinstatement groups, as to isolate the contribution of BS LIV-VI aIC-BLA neurons in encoding the subjective predictability of taste experience during taste learning, re-learning, and memory retrieval (Fig. 5). Consistent with studies in the hippocampus (Graves et al., 2016), we found that the percentage of BS LIV-VI aIC-BLA projecting neurons in the sampled population was highest in the context of novel taste learning (Extended Data Fig. 1-2B; Saccharin 1x, 85%), and subsided following progressive familiarization (Extended Data Fig. 1-2B; Saccharin 2x, 65%, Mann-Whitney test: p = 0.0562; Sum of ranks: 303.5, 161.5; Mann-Whitney U = 70.50; Saccharin 5x, 55.56%, Mann-Whitney test: p = 0.0034; Sum of ranks: 291, 87; Mann-Whitney U = 32). Interestingly, animals retrieving CTA, exhibited the lowest proportion of BS neurons among the six treatments (Extended Data Fig. 1-2B; CTA Retrieval, 44.44% BS), and significant differences were observed compared with control animals (Extended Data Fig. 1-2B; Saccharin 2x, 65% BS; Mann-Whitney test: p = 0.0102; Sum of ranks: 257, 208; Mann-Whitney U = 55). Thus, the ratio of BS:RS LIV-VI aIC-BLA projecting neurons was plastic in relation to experience and was highest in response to appetitive novelty, in accord with studies investigating the intrinsic excitability of subiculum output neurons in relation to contextual novelty and valence encoding (Dunn et al., 2018). Indeed, the ratio of BS:RS LIV-VI aIC-BLA neurons was progressively decreased by familiarity acquisition (Saccharin 1x . 2x . 5x), as well as following aversive taste memory recall (CTA Retrieval), compared with both appetitive learning (Extended Data Fig. 1-2B; Saccharin 1x, Mann-Whitney test: p , 0.0001; Sum of ranks: 407, 188; Mann-Whitney U = 35) and relearning (Extended Data Fig. 1-2B; Extinction, Mann-Whitney test: p = 0.0007; Sum of ranks: 182, 224; Mann-Whitney U = 29). However, our comparison failed to account for the influence of complex experiences over time, as differences between Extinction and Reinstatement failed to reach significance (Extended Data Fig. 1-2B; Mann-Whitney test: p = 0.3870, Sum of ranks: 133.5, 97.50, Mann-Whitney U = 42.50), while perplexingly, the ratio of BS:RS aIC-BLA neurons in these groups was differentially increased compared with Saccharin 5x (Extended Data Fig. 1-2B; Extinction, Mann-Whitney test: p = 0.0300; Sum of ranks: 81, 150; Mann-Whitney U = 26; Reinstatement, Mann-Whitney test: p = 0.3498; Sum of ranks: 90, 120; Mann-Whitney U = 35), but not Saccharin 2x (Extended Data Fig. 1-2B Changes in the intrinsic properties of neuronal ensembles have recently been suggested to contribute to homeostatic mechanisms integrating both cellular and synaptic information (C.H. Wu et al., 2021). In our current study, we randomly sampled from neuroanatomically defined LIV-VI aIC-BLA projecting neurons, and even following spike sorting (Extended Data Fig. 1-2), the probability of recording from engram cells (10% of neurons within a region) would be extremely low (Tonegawa et al., 2015). Importantly, the correlative nature does not exclude the possibility that these changes are the consequence of representational drift (Driscoll et al., 2017). We thus set out to examine the hypothesis that applying linear dimension reduction method on the complement of intrinsic properties recorded in BS LIV-VI aIC-BLA neurons would allow us to distinguish between taste experiences that differ in terms of their perceived predictability (or the associated probability for further aversive learning).
Principal component analysis of the profile of intrinsic properties in BS LIV-VI aIC-BLA projecting neurons separates treatment groups in relation to the perceived predictability of taste valence for saccharin We assigned six of our treatment groups into highly predictive scenarios (Saccharin 5x, CTA Retrieval and Reinstatement), and low predictive scenarios (Saccharin 1x, Saccharin 2x, Extinction). We used parallel analysis to select the components across the complement of intrinsic properties in each treatment group, with the first three principal components (PC1-3) explaining 30.84%, 17.88%, and 13.75% of the total variance, respectively, and 62.47% of the variance, collectively (Extended Data Fig. 5-2). PC1 ( Fig. 5B; Extended Data Fig. 5-2) was characterized by strong negative loadings for Rheobase (À0.88304), sAHP (À0.85985), and mAHP (À0.82421), while a positive correlation was identified for IR (0.694764). The direction of PC2 ( Fig. 5B; Extended Data Fig. 5-2) was positively correlated with fAHP (0.682681) and AP halfwidth (0.614103) and was negatively correlated with Excitability at 350 pA (À0.69 587). PC3 ( Fig. 5B; Extended Data Fig. 5-2) positively correlated with SAG ratio (0.889949) and RMP (0.682681), whereas a significant negative correlation with mAHP (À0.6095) was also identified. Unlike aversive or appetitive taste memory retrieval (i.e., highly predictive), appetitive novelty or extinction learning (i.e., low predictive), was associated with increased input resistance, faster action potential generation and decreased afterhyperpolarization on BS aIC-BLA neurons (Fig. 5). Importantly, PCA of the intrinsic properties of LIV-VI aIC-BLA projecting neurons regardless of cell type (BS and RS together; Extended Data Fig. 5-1), failed to segregate the two groups of treatments. This cell-type-specific profile of intrinsic properties could provide the framework through which BS LIV-VI aIC-BLA projecting neurons are able to inspect the gastrointestinal consequences associated with tastants over prolonged timescales, when these consequences are not accurately predicted by sensory experience or memory retrieval alone (Adaikkan and Rosenblum, 2015;Lavi et al., 2018;Kayyal et al., 2019).

Discussion
Learning and memory are subserved by plasticity in both synapse strength and neuronal intrinsic properties (Citri and Malenka, 2008;Sehgal et al., 2013). While Hebbian rules can explain associative learning paradigms where seconds separate the CS and US (Krabbe et al., 2018), additional cellular-level mechanisms are needed to explain how learning operates in paradigms where the time between CS and US extends to hours (Adaikkan and Rosenblum, 2015;C.H. Wu et al., 2021). In this study, we demonstrate that following taste experiences, taste percept and prior experience are integrated in the intrinsic properties of the aIC in a time-dependent and cell typespecific manner. We further show that regardless of the identity or prior history associated with taste, the intrinsic properties of BS LIV-VI aIC-BLA projecting neurons encodes the perceived confidence of taste valence attribution.
We focused on the aIC-BLA projection; a circuit causally involved in the acquisition and retrieval of CTA memories (Kayyal et al., 2019. We examined the hypothesis that excitability in aIC-BLA neurons can serve as a taste valence updating mechanism enabling the prolonged ISI between CS and US in CTA learning (Adaikkan and Rosenblum, 2015), and/or contributes to anticipatory valence attribution (Barrett and Simmons, 2015). Our basic proposition diverged from Hebb's famous postulate that cells with increased excitability over hours can potentially wire together with cells conveying incoming modified valence information (Hebb, 1961).
The confidence with which taste valence is encoded is the product of the subjectively perceived (1) appetitive or aversive nature of tastants and (2) novelty or familiarity associated with tastants (Russell, 1980;Kahnt and Tobler, 2017). We first examined each axis separately and later in tandem as to better simulate real-life scenarios. We measured the intrinsic properties of aIC-BLA neurons 1 h following taste experience, a previously established suitable time point (Jones et al., 1999;Haley et al., 2020).
To dissociate novelty-related changes from those involving hydration, taste identity and familiarity; we compared the intrinsic properties of aIC-BLA neurons following Water, a neutral and familiar tastant, Saccharin, an innately appetitive tastant, in the context of novelty (1Â) or familiarity (5Â), and Quinine, an innately aversive novel tastant (Fig. 1). Excitability was high following novel saccharin exposure, but not in response to Quinine (Fig. 1D). Indeed, concerted activity at the aIC and BLA encodes the presence, identity, and palatability of taste experiences within the 2 s preceding swallowing (Katz et al., 2001;Grossman et al., 2008;Fontanini et al., 2009). Palatability can be enhanced as a function of experience (Austen et al., 2016) but can also be decreased by sensory satiety and alliesthesia (Rolls et al., 1981;Yeomans, 1998;Siemian et al., 2021). However, excitability on the projection was enhanced in response to novelty and was decreased following familiarization (Fig. 1). Further inconsistent with palatability encoding; changes in excitability captured 1 h following novel saccharin exposure subsided 4 h later ( Fig. 1), while excitability remained plastic even following longer periods of water restriction, that could be considered monotonous (Fig. 5). Deciphering whether and how aIC-BLA neurons contribute to palatability processing would require in vivo recordings to capture taste-evoked changes, within timescales that are beyond the scope and means of our current study (Vincis and Fontanini, 2016).
The correlation identified between excitability and innate current taste valence, encouraged us to examine the predictability of future outcomes following aversive taste memory retrieval. Bearing in mind our previous findings using transcription-dependent activity markers at the aIC , we hypothesized that aversive taste memory retrieval (CTA Retrieval or Reinstatement) would be associated with decreased excitability compared with stimulus-matched and familiarity-matched controls (Saccharin 2x and Extinction). Indeed, excitability on aIC-BLA projecting neurons following CTA Retrieval was decreased compared with Saccharin 2x (Fig. 2B), while Reinstatement was also associated with decreased excitability compared with Extinction (Fig. 3E). Hence, regardless of the complexity of taste memory retrieval, excitability in aIC-BLA neurons was best predicted by the subjective predictability of taste valence, increasing in response to innately appetitive taste experiences in which the perceived possibility for avoidance learning was high/ taste valence predictability was low (Fig. 4). Conversely, when the subjective confidence with which taste valence was predicted was high, excitability on the projection remained unchanged (Figs. 1, 4).
Innately and learned aversive tastants were both associated with decreased excitability on aIC-BLA projecting neurons compared with appetitive controls; however, these effects were mediated through alternative mechanisms (Figs. 2-4). Quinine increased fAHP on the projection compared with saccharin, regardless of familiarity or perceived valence (Figs. 1, 4). Postspike after-hyperpolarization (AHP) has a key function in transducing the summed result of processed synaptic input, directly impacting neuronal excitability in relation to both learning and aging (Oh and Disterhoft, 2020). In pyramidal cells of the hippocampus and cortex, differences in fAHP are mediated by the Ca 21 -dependent and voltagedependent BK currents that promote repolarization at the beginning APs trains (Shao et al., 1999). Interestingly, studies in the prefrontal cortex (PFC), have shown that fear conditioning decreases excitability, whereas extinction training enhances excitability and decreases medium and slow AHP (Santini et al., 2008;Maglio et al., 2021). Chronic ethanol consumption has shown to suppress excitability and to increase AHP in IC neurons (Luo et al., 2021). Conversely, oxytocin-dependent signaling has been shown to promote social affective behaviors, via increases in excitability and decreases in sAHP in IC neurons (Rogers-Carter et al., 2018). Further studies would be necessary to fully address this, but our findings could indicate that enhanced fAHP is induced by innately aversive tastants or quinine specifically.
Unlike Quinine, CTA memory retrieval was associated with increased AP amplitude and SAG ratio, as well as decreased IR in BS LIV-VI aIC-BLA projecting neurons, compared with control animals ( Fig. 2; Extended Data Fig.  2-1). On the other hand, the decreased excitability in the Reinstatement group compared with Extinction was characterized by decreased AP threshold, increased t , and decreased sAHP (Figs. 3, 4; Extended Data Fig. 3-1). The hyperpolarization-activated, cyclic nucleotide-gated current (I h ) regulates membrane depolarization following hyperpolarization (Hogan and Poroli, 2008;Shah, 2014). The opening of HCN channels generates an inward current, that modulates AHP, RMP and IR in cortical pyramidal and PV interneurons (Yang et al., 2018). However, conductance through I h channels, regulates synaptic integration at the soma of pyramidal neurons, by suppressing excitability, decreasing IR, and increasing t (Hogan and Poroli, 2008). Evidence indicates that this dichotomous impact of HCN channels on neuronal excitability, is mediated by A-type K channels at the dendrites (Mishra and Narayanan, 2015), and M-type channels at the soma (Hu et al., 2007). Notably, AP halfwidth was significantly increased in the Extinction and Reinstatement groups that had undergone extinction training, compared with all other saccharin-treated groups (Fig. 4F). Mechanistically, this effect could reflect changes in the distribution and/or the properties of voltage-gated or calcium-gated ion channels (Faber and Sah, 2002;Grubb and Burrone, 2010;Kuba et al., 2010). Such broadening of spike width has also been reported in infralimbic PFC neurons projecting to the amygdala in response to extinction training (Senn et al., 2014). PV-dependent restriction of excitability and burst firing, is instrumental in experience-dependent plasticity in the amygdala (Morrison et al., 2016), the hippocampus (Donato et al., 2013;Xia et al., 2017), and visual cortex (Yazaki-Sugiyama et al., 2009;Kuhlman et al., 2013). Conversely, in the striatum, PV interneurons restrict bursting, calcium influx, and synaptic plasticity to appropriate temporal windows that facilitate learning, but not retrieval (Owen et al., 2018). Elegant recent studies report that rapid eye movement sleep is associated with a PVdependent somatodendritic decoupling in pyramidal neurons of the PFC (Aime et al., 2022). At the IC, the maturation of GABAergic PV circuits is key for multisensory integration and pruning of cross-modal input to coordinate the detection of relevant information (Gogolla et al., 2014). Activation of IC PV disrupts the expression of taste-guided goal-directed behavior (Vincis et al., 2020) and enhances tasteguided aversive responses . Our findings could be indicative of a prediction-dependent decoupling mechanism at the IC, whereby the restriction of bursting activity in LIV-VI aIC-BLA neurons impinges on innate drives toward the tastant and further learning, depending on prior experience.
We further probed our results and hypotheses using PCA and attempted to segregate behaviors based on the perceived ability of the CS to predict the consequences of sensory experience, and the probability for further learning (Fig. 5). We focused on BS LIV-VI aIC-BLA projecting neurons since bursting has been implicated in coincidence detection by deep-layer neurons (Boudewijns et al., 2013;Shai et al., 2015), as well as the encoding of novelty and valence relating to different modalities (Song et al., 2015;Dunn et al., 2018;Yousuf et al., 2019). Our PCA of intrinsic properties in BS LVI-VI aIC-BLA projecting neurons demonstrated that distinct plasticity rules are at play depending on the balance between the probability for associative learning and the certainty with which taste predicts the valence of experience during retrieval (Extended Data Figs. 5-1, 5-2). We propose that increased excitability and reduced fAHP on BS LIV-VI aIC-BLA projecting neurons might represent a transient neuronal state in the absence of adequate predictive cues for the outcome of taste experiences (Adaikkan and Rosenblum, 2015).
The IC has long been considered crucial for interoception, which is increasingly understood to be supported by distinct direct or indirect functional bidirectional connectivity. Indeed, interoceptive inputs relating to the processing, or anticipation of physiological states of hydration and satiety manifest at the pIC (Livneh et al., 2017(Livneh et al., , 2020Livneh and Andermann, 2021). However, this is rarely the case when it comes to physiological hydration or satiety inputs and the aIC, that has is primarily involved in interoceptive processes in the context food poisoning or CTA Y. Wu et al., 2020). As other studies currently in press demonstrate, hydration correlates and requires decreased activity in aIC-BLA and increases in pIC-BLA CB1 receptor-expressing neurons (Zhao et al., 2022). In fact, and in further accord with our previous findings (Lavi et al., 2018;Kayyal et al., 2019), optogenetic activation of the aIC-BLA projection was found to be anxiogenic, while physiological conditions that are associated with negative valence and anxiety were encoded by changes in activity on the projection (Nicolas et al., 2022). Under uncertain conditions that are associated with greater potential significance, recruitment of the aIC is thought to contribute to attention, effort, and accurate processing (Lovero et al., 2009), as to identify better response options (Preuschoff et al., 2008). In agreement with earlier computational models of the cortical connectivity (Mumford, 1991(Mumford, , 1992, recent work indicates that the aIC facilitates prediction-related encoding driven by hedonics, rather than homeostatic needs (Darevsky et al., 2019;Price et al., 2019;Darevsky and Hopf, 2020). Our results, propose a cellular framework for such an emotional predictive function at the aIC. Indeed, sex can be an important biological variable when examining brain circuits in relation to behavior (Rogers-Carter et al., 2018). Although we cannot exclude the possibility of sex-specific differences, in previous studies where we manipulated activity at the IC in a cell type-specific manner, we found no differences between male and female mice (Kayyal et al., 2019;Yiannakas et al., 2021). Future studies will further explore whether and how the interplay between such distinct mechanisms at the aIC, enables its complex role in learning, memory, and decision-making.