|Home | About | Journals | Submit | Contact Us | Français|
Through learning processes, cues associated with emotionally salient reinforcing outcomes can come to act as substitutes for the reinforcer itself. According to one account of this phenomenon, the predictive cue associatively elicits a representation of the expected outcome by reactivating cells responsible for encoding features of the primary reinforcer. We tested this hypothesis by examining the role of neural ensembles in gustatory cortex (GC) during receipt of gustatory stimuli (sucrose and water) and cues associated with those stimuli using the immediate early genes (IEGs) Arc and Homer1a. Because these plasticity-related IEGs are expressed in the neuronal nucleus 5 and 30 mins, respectively, after salient events, we examined how individual neurons encoded these stimuli in two separate behavioral epochs. In Experiment 1, we showed that tasting identical sucrose solutions, but not tasteless water, in the two epochs increased both IEG activity and the degree of overlap between neural ensembles in GC. In Experiment 2, odor cues associated with sucrose, but not water, evoked potentiation of IEG activity in GC similar to sucrose itself. Surprisingly, lesions of the basolateral amygdala had minimal effects on associative encoding in GC. Finally, these associatively-driven representations of sucrose appeared to be outcome specific, as neural ensembles that were activated by the sucrose-associated cue were also activated by sucrose itself. This degree of overlap between associative and primary taste activity at the ensemble level suggests that GC neurons encode important information about anticipated outcomes. Such representations may provide outcome-specific information for guiding goal-directed behavior.
Animals learn and remember information about foods they encounter. Predictive cues such as odors can be used to select food, while flavors and post-ingestive features of foods can guide future selection and consumption. Much research has considered the role of gustatory cortex (GC) in acquiring and retaining such information. GC, a region of the insula receiving dense ascending fibers from the gustatory thalamus, has been shown to be driven by tastants, particularly in the dysgranular layer (Kosar et al., 1986). In addition to this sensory role, GC is also important for acquiring new information about tastes on the basis of experience (Kiefer et al., 1984). Recent work has shown that such learning requires NMDA-dependent plasticity (Escobar et al., 1998) and protein synthesis (Bermudez-Rattoni et al., 2004; Elkobi et al., 2008) in GC.
Although encoding properties of GC neurons in response to tastants have been examined (Katz et al., 2001; Lemon and Katz, 2007), less is known about GC representation of tastes in response to non-gustatory cues that predict those tastes. Imaging studies in rodents using immediate early genes (Dardou et al., 2006, 2007; Kerfoot et al., 2007) and in humans using fMRI (Veldhuizen et al., 2007; Small et al., 2008) show increased activity in GC in the presence of taste-predictive cues. Similarly, GC function appears necessary for performance thought to be mediated by such associatively-activated representations. For example, after selectively sating rats with one tastant, rats with GC lesions are impaired in their ability to spontaneously reduce pressing a lever that previously produced that food (Balleine and Dickinson, 2000).
These findings suggest that GC is involved in forming or using associative representations of taste outcomes. Parallel findings indicate that the basolateral amygdala, (BLA) also participates in rapid encoding of cue-outcome associations (Allen et al., 1991; Hatfield et al., 1996; Balleine et al., 2003; Pickens et al., 2003). As such, connections between BLA and GC may be important for encoding features of representations in GC (Barot et al., 2008; Grossman et al., 2008; Fontanini et al., 2009).
Here we tracked how individual neurons in GC were recruited to encode such information using two immediate early genes (IEGs) Arc/Arg3.1 (Arc) and Homer1a (H1a) (Link et al., 1995; Lyford et al., 1995; Brakeman et al., 1997). These IEGs are NMDA-dependent markers for plasticity and show differential temporal expression patterns in activated neurons (Guzowski et al., 2001). We exploited this temporal difference to examine patterns of neural activity across two different behavioral epochs, allowing us to map ensemble activity for both tastes (experiment 1) and taste-associated cues (experiment 2). We found that sucrose increased IEG expression in GC compared to tasteless water, and that repeated exposure to identical solutions activated overlapping populations of neurons. Furthermore, taste-associated cues increased IEG expression in GC, even in the absence of tastants, and neurons activated by sucrose-associated cues were reactivated by sucrose alone. BLA lesions failed to disrupt the content of these associative ensembles in GC, suggesting that limbic input is not necessary for GC associative taste representations.
Thirty-one Long-Evans rats (Charles River Laboratories, Wilmington, MA) weighing between 350–425 g at the time of experiment served as subjects. Three of the rats used in experiment 1 had previously received exposure to saccharine in different test chambers. Rats were housed individually on a 12:12 light/dark schedule and had ad libitum access to standard rat chow and water until the beginning of behavioral testing procedures. All procedures were conducted in accordance with Johns Hopkins University and NIH guidelines for animal research.
All training procedures took place in the rats’ home cages, which were standard plastic tubs (24″ × 12″ × 12″), topped with a steel wire lid where chow and water were normally presented. Prior to testing, all food and normal cage water was removed. When fluids were presented in each session, a standard glass bottle with a rubber stopper and steel sipper tub was placed on the cage top so that the tube protruded into the box. Rats were placed on a schedule of restricted access to water one week prior to training during which they received 15 min of access to water per day at approximately 4PM. Rats had unrestricted access to chow during the pre-training phase.
The solutions used during testing were either reverse-osmosis-filtered (RO) water or a 10% (w/v) sucrose solution in RO water. Fluids were presented in glass bottles affixed with a steel sipper tube. The sipper tube for the test presentations was similar to the normal cage water bottle, but differed in that it used a ball bearing in the tube, while the normal cage bottle used a narrow-gauge opening at the tip.
To reduce neophobia to the sucrose solution, two days prior to testing, rats were presented with the sucrose on their home cages for 2 min during the normal water presentation time in the pre-training phase. Bottles were weighed before and after the sucrose exposure to ensure that each rat experienced the solution. Rats that failed to consume any sucrose had the bottle placed back on the home cage for an additional 2 min. At the end of pre-training, rats were allowed to drink normal cage water for another 13 min to meet their water requirements for the day. During pre-training and also during the test phase of experiment 1, rats were presented with solutions in the colony room.
In experiment 2, odorants were presented along with solutions. Odors employed in experiment 2 were biologically neutral and easily discriminated. The odors, hexenol b gamma extra and iso amyl butyrate (International Flavors and Fragrances, New York) were diluted to 1.25% (v/v) in propylene glycol. In this experiment, rats were moved to one of two different rooms for testing. Both rooms were approximately the same size, and were dimly illuminated with ambient red light. To minimize the risk of the different odors commingling, a consistent odor was used in each room during all testing.
The experimental design follows from the temporal features of IEG expression of Arc and H1a. Specifically, salient neural activity that drives plasticity will result in H1a mRNA being expressed in the cell nucleus 25–30 min following the event, while Arc mRNA can be detected 5–8 min after such an event. Because Arc and H1a have been shown to be co-expressed after exposure to salient events (Vazdarjanova et al., 2002), the temporal differences in nuclear expression of these IEGs can be used to explore how the same cell responds to two different events. To do this, the experiments used two 5-min behavioral epochs separated by 20 min, followed by sacrifice. As a result, events 25–30 minutes prior to sacrifice (epoch 1) will drive IEG expression that will be seen as nuclear H1a, while events 5 minutes prior to sacrifice (epoch 2) will be seen as nuclear Arc expression. Thus, each cell will either show H1a only (activity only in the first epoch), Arc only (activity only in the second epoch), both H1a and Arc (activity in both epochs) or no IEG activity. Single IEG expression (either Arc or H1a) reflects selectivity for one of the two tastants, while double labeling demonstrates that the cell treated the two tastants similarly.
For behavioral testing, in experiment 1, rats were presented with either sucrose or water on their home cages in one of two time-limited epochs (Table 1). In the first epoch, rats received a presentation of a fluid for 1 min on the home cage. Twenty-four min later, rats received another 1-min presentation of a fluid. Five min after the last presentation, rats were sacrificed. In this experiment, each rat was assigned to one of three different general conditions. In the first condition, rats received the same outcome in both epochs (SAME), either 10% sucrose (n = 7) or 5 Froot Loops (one per minute) (n = 1). In the DIFFERENT condition, rats received 10% sucrose in one epoch and plain water in the other epoch. The order of sucrose presentation was counterbalanced; 2 rats received 10% sucrose in the first epoch and water in the second epoch, and 3 rats received the solutions in the opposite order. In the final WATER control group, rats (n = 4) drank plain water in both epochs.
Experiment 2 was designed to assess the extent to which taste-associated cues were able to drive IEG expression in GC in the absence of the tastants themselves, and further, whether such activity was similar to that seen during primary gustation. The summary of these designs are seen in Table 1. All rats were trained to drink during two different epochs and in different contexts. Rats were run in two sessions per day. Each session was 5 min long, separated by a 20 min inter-session interval. During each session, the bottle of the fluid being presented was placed on the cage one min following the start of the epoch. At the stopper end of the spout, an odor saturated (0.5 ml) disc of filter paper acted as a mechanism to deliver the odor cue. Scented odor cues were used for one epoch per day that employed an odor-outcome pairing, but not for the control epoch in which water was presented alone. The odorant was at a concentration such that the maximum intensity was achieved when the rat was drinking from the spout, though some diffusion of odorants was likely present at a significantly lower concentration elsewhere in the testing bin. Rats drank the fluid in three 1-min bouts, each bout separated by 30s during which the bottles were removed from the cage. Because of our concern over diffusion of odors, different rooms were used as the contexts for these sessions to minimize contamination between odors; all testing with one odor was done in one context and testing with a different odor in the other context. Many provisions, however, were taken to make the contexts as similar as possible otherwise. Both rooms were approximately 10′ × 10′ and were dimly illuminated with red lamps. Lamps were located under the tables on which the rats were placed to ensure diffuse light and to reduce the impact of visual contextual stimuli. Half the rats started in one context and the other half in the other context.
The rats underwent eight days of training. During this phase, rats drank from standard bottles with a steel sipper tube placed on the home cage. In each of the contexts were exposed to one of two odors (either hexenol b gamma extra (context A) or iso amyl butyrate (context B), placed on the 1″ diameter piece of filter paper. For each animal, one odor was always paired with a bottle containing 10% sucrose solution (O1), and the other odor was always paired with normal tap water (O2). However, the identity of the specific odor/flavor pairings was counterbalanced among the rats.
Each day, rats received an odor/outcome pairing in one context (e.g., O1) and unscented plain water in the second context. This was done because it was expected that residual levels of odorants specific to the context would be ambiently present in each room even in the absence of explicit presentations. We addressed this by allowing rats to drink unscented water in both contexts to minimize effects of any such lingering odors on contextual encoding and thus enhance associability of the explicit odor cue with its paired outcome. Thus, across all eight days of training, for each day rats received one context with odor (either O1 or O2, depending on the day) and the other context with odorless water, for a total of four O1 presentations, four O2 presentations, and eight unscented water-only presentations (four per context; Table 1). This training protocol also acclimated rats to the timing of the two-epoch protocol used on the test day.
On the day of test, rats were run on the same schedule as during training, but were assigned to one of two conditions. In the first condition, rats (n = 5) that had been trained to learn the associations between two odors, only one of which was paired with a flavored sucrose solution, were tested with exposure to each of those odors in different epochs. Prior to sacrifice, rats were given exposure to the sucrose-paired odor (O1) presented with water instead of sucrose in one epoch, while the water-associated odor (O2) was still paired with water presentation in the other epoch. The order of O1 and O2 presentation was counterbalanced by subject during the tests.
It is known that GC and BLA share reciprocal connections, and as such, it was of interest to know whether BLA lesions modulated associative taste coding in GC. Existing evidence suggests that rats with BLA lesions fail to show behavioral evidence for specific associative representations (Hatfield et al., 1996; Balleine et al., 2003; Pickens et al., 2003). However, there is little evidence to suggest that BLA lesions affect taste perception at all (Pickens et al., 2003). As such, we predicted that lesions of BLA would also allow us to determine how BLA lesions affect GC coding of real (primary) taste representations. Thus a second group of animals in the O1 vs O2 condition (n = 4) were given BLA lesions and run with the same training and test as control subjects.
In the second condition, rats (n = 5) received the sucrose-paired O1 odor presented with water in one epoch, and actual 10% sucrose alone (i.e., no odor) in the other epoch during the test. The sucrose solution was given in the opposite context as the O1/sucrose context to minimize the effect of any associative contextual cues. This condition, therefore, assessed the degree of co-expression between “real” sucrose and the associatively-activated sucrose representation. In this condition, order of O1 and sucrose presentation was counterbalanced by subject.
The effects of neurotoxic lesions of the BLA were assessed in experiment 2 in the O1 vs O2 condition to determine whether the occurrence of odor activated taste representations in GC depended on the integrity of BLA. Those rats (n = 4) were anaesthetized with isoflurane gas mixed with oxygen, and mounted in a standard stereotaxic apparatus (David Kopf, Tijunga, CA). A midline incision was made and the scalp and fascia were retracted. The skull was leveled by bringing bregma and lambda to within 0.1 mm on the DV axis. Holes were drilled in the skull at −2.8 mm AP and ±5.1 mm ML relative to bregma. Injections of the neurotoxin NMDA (20 mg/ml in 0.1 M PBS) were made using a narrow-tapered glass pipette attached to a picospritzer (General Valve, Fairfield, NJ). Coordinates were taken at the skull surface at bregma, then the pipette was moved to −2.8 mm AP and ±5.1 mm ML relative to bregma. At both bilateral injection sites, the pipette was then slowly lowered to −8.7 mm DV from the skull surface at bregma and allowed to rest for 5 min. Neurotoxin (0.15 μl) was injected at rate of 0.025 μl/min over the course of 6 min. The pipette was then allowed to remain in place for another 5 min to allow the solution to diffuse and be absorbed into the surrounding neurons. The pipette was then raised to −8.3 mm DV and allowed to rest another 3 min before injecting 0.07 μl at a rate of 0.025 μl/min. Following another 5 min resting period, the pipette was slowly raised out of the brain. The scalp was then sutured shut with silk sutures and the wound was treated with topical antibiotics. Rats were allowed to recover for 30 min under warming lamps, after which they were returned to the colony room.
Exactly 1 min after the end of the 2nd testing epoch, rats were placed in a plastic tub with a tight-fitting lid saturated with isoflurane vapors for 1.25 min, and then rapidly decapitated with a guillotine. Brains were removed, notched on the rat’s left side, and blocked rostrally just caudal to the rostral pole of the prefrontal cortex and caudally through the middle of the cerebellum. The brains were then lowered into super-cooled (approximately −80°C) isopentane cooled externally by a dry ice and ethanol slurry and allowed to freeze for 5 minutes in the isopentane before being removed and wrapped in foil. Frozen brains were stored in a −80°C freezer until sectioning on a cryostat.
On the cryostat, the brain was warmed to −20°C then fixed to the platform with Tissue Tek OCT. Sections taken were either 16 μm (experiment 1) or 20 μm (experiment 2) which were immediately mounted onto slides. Mounted slides were then sealed in boxes and subsequently stored in a −80°C freezer until used for processing.
For rats that had received lesions of BLA, a separate series of sections was allowed to dry at 37°C. Sections were first fixed for 2 hr in 0.1M PB-buffered 10% formaldehyde. For staining, slides were defatted in a 1:1 mixture of ethanol and chloroform for 2 hr, then were gradually rehydrated in decreasing concentrations of alcohol. Sections were stained with a 0.2% thionin stain, then gradually dehydrated. Finally, tissues were bathed in xylenes and coverslipped using DPX. Lesion validity was assessed by visual inspection with a microscope. Morphological changes such as tissue shrinkage, holes and gliosis were taken as evidence of tissue damage.
Processed sections underwent processing for double label fluorescent in situ hybridization (FISH) for Arc and H1a mRNA. In situ procedures were adapted from Guzowski & Worley (2001) and Petrovich et al.(2005). Arc and H1a plasmids were linearized by separately incubating in a solution of BSA, 10× enzyme buffer and EcoR1 enzyme at 37°C for 2 hours. Cutting was confirmed by running on a gel. Linearized cDNA was then purified in phenol/chloroform and precipitated in 100% alcohol. The solution was stored at −20C overnight, after which the solution was centrifuged and the aqueous layer retained. The solution was rinsed in 70% alcohol and centrifuged three more times. The alcohol was then removed and the pellet allowed to air dry, after which TE buffer was added and the cDNA was resuspended. Finally, this cDNA was bound into a conjugate with digoxigenin-UTP (Arc) or fluroescein-UTP (H1a) and incubated at 37C for 1.5 hr. At the end of this process, the probe was purified by spinning through spin columns (Roche) for 5 min. The resulting conjugate probe was typically used that day, though in some situations the probe was kept at −20C and was used within a week.
For the in situ reactions, sample tissues were allowed to warm to −20C in a cryostat for 1 hr before processing. This tissues were then fixed in fresh buffered 4% paraformaldehyde, washed in 2× SSC, then bathed in 0.5% acetic anhydride in TEA buffer. Tissues were then stabilized in a 50:50 mix of methanol and acetate, and washed again in 2× SSC. Tissues were incubated in the hybridization buffer at room temperature for 30 minutes in a humid chamber. Hybridization buffer was gently removed, then slides were incubated in the same hybridization buffer plus both Arc and H1a riboprobes (approximately 10 μg/ml for each probe). Coverslips were placed over the tissue, sealed in place with DPX and then slides were incubated at 60C overnight (approx. 20 hr). Following this incubation, slides were washed in 2× SSC at 60C, then treated with 3 μl/ml RNAse at 37C to eliminate any non-hybridized RNAs in the tissue. Slides were washed in descending concentrations of SSC (from 4× to 0.1×) plus 0.2% w/v sodium thiosulfate. Following a rinse in Tris buffered saline plus 2 μl/ml Tween (TBS-T), slides were moved to a humid chamber, then blocked with TSA TNB solution (Perkin-Elmer) at RT. Slides were then incubated with HRP-antibody conjugate anti-digoxigenin-POD (1:1000) in blocking solution at 4C overnight. The following day, slides were washed in TBS-T, then reacted with the fluorophor cyanine-3 (CY3) (Perkin-Elmer) for 30 min at RT in the humid chamber. Slides were washed in TBS-T, and then rinsed in 2% H2O2 to quench residual peroxidase activity. Tissues were then washed in TBS-T and removed to the humid chamber. Slides were then incubated with the antibody anti-fluorescein-POD in TSA TNB buffer (1:1000) (Perkin-Elmer) at 4C overnight. The following day, slides were rinsed in TBS-T then reacted with the fluorescent marker fluorescein (Perkin-Elmer). Tissues were dried then coverslipped with Vectashield coverslipping media containing the nuclear Nissl-type dye 4′,6-diamidino-2-phenylindole (DAPI) to counterstain cell nuclei. Slides were kept at 4C until imaging at confocal microscope.
Tissues were imaged using a Zeiss LSM 510 Meta confocal microscope using three lasers (488, 520 & 570 nm), each corresponding to the peak emission spectrum for DAPI (Nissl stain for cell nuclei), Fluorescein (H1a mRNA) and CY3 (Arc mRNA), respectively. Data were acquired using a z-stack (0.5 μm thickness per section in stack), the height of which was determined by the penetration of detectable probe for each sample. For this experiment, the objective lens was set at 60× magnification for one block of animals (n = 9; experiment 1), 40× for another block (n = 14; experiment 2), and 20× for a third block of subjects (n = 8; experiment 1). Early sessions employed the higher magnification to ensure accurate expression rates and double labeling in each cell. However, lower magnification in later replications allowed for a greater sampling of each region of interest while decreasing chances of sample bias. Typically 8 samples were taken from each animal, taking bilateral samples from four anatomical planes anterior to bregma at +0.95, +1.20, +1.45 and +1.70 mm anterior to bregma. However, where extensive tissue damage occurred due to in situ processing, data were not collected and the data were omitted.
Data were analyzed using LSM 510 Image Browser software (Zeiss), and cell counts were done manually. In order to eliminate bias, images were first viewed with only the DAPI-labeled (Nissl) cells present, (e.g., turning off the filter for the CY3 and Fluorescein channels). To do this, a piece of transparency was taped to the computer monitor and valid cells were circled in black. Cells were counted for analysis if the nucleus was present on at least 5 sections of the z-stack. Once cells were determined to be valid, the CY3 filter was turned back on to reveal only the CY3-positive markers. A second transparency was placed over the DAPI transparency and CY3-positive markers were circled in red. That sheet was removed, and the CY3 filter was turned off, while the Fluoroscein filter was turned on. A third transparency was overlaid on the DAPI transparency, and Fluorescein-positive cells were circled in green. Analysis was done by overlaying all three transparencies. Cells which were positive for both DAPI and CY3 were considered Arc-positive, cells with both DAPI and Fluorescein were considered H1a-positive, and cells with DAPI, CY3 and Fluorescein were considered double-labeled for both mRNA.
All behavior and neural data were compiled for each subject and evaluated by repeated measures ANOVA that treated subjects as a random effect. In some situations, significant violations of parametric assumptions demanded analysis with appropriate non-parametric analysis. Tests were computed using Statistica software (StatSoft, Tulsa, OK).
Rats consumed substantial amounts of both normal water and the sucrose solution. Across all subjects, rats consumed an average of 9.23 mL of sucrose per epoch and 7.97 mL of water per epoch. There were no significant differences in consumption for either fluid, behavioral condition or epoch (data not shown). The rat in the SAME condition consumed all five Froot-Loop pieces during each epoch. Because we expected that consumption of identical sucrose solutions or identical cereal pellets in both epochs would induce similar rates of GC expression, data from the single Froot-Loop-consuming rat were combined with those from the sucrose-consuming rats in the SAME condition.
We then analyzed the rates of IEG expression for Arc and H1a in GC. GC was defined by anatomical boundaries determined by several factors. Several studies have shown important functional divisions in the insula in relation to the cytoarchitectonic structure of the region. For example, whereas the insular region is typically comprised of an allocortical agranular cortical region (AI), a neocortical granular cortical region (GI) and a transitional dysgranular cortical region (DI) between them, the majority of taste responsive neurons have been shown to be centered on the DI region specifically (Kosar et al., 1986; Yamamoto et al., 1989). Further, results from both lesion (Balleine and Dickinson, 2000) and recent 2-photon imaging studies (Accolla et al., 2007) have demonstrated a greater importance for the more anterior portions of insula ranging from approximately +0.7 to +1.7 mm anterior to bregma for encoding taste-related information. Finally, anatomical tracing studies have shown that these same regions are reciprocally connected with limbic structures such as BLA and OFC, while more posterior sections of the insula have connections with extended amygdala structures such as central nucleus of the amygdala and bed nucleus of the stria terminalis (Allen et al., 1991). In a series of animals (n = 4), we confirmed these boundaries using histochemical analysis of retrograde (Fluorogold) and combined anterograde/retrograde (cholera toxin b) tracers injected into either GC or BLA (Supplemental Fig. S1).
In support of these findings, we assessed total levels of IEG expression in a pilot subject after sampling tastants. For this subject, samples were taken in the DI layer of the insula from +2.0 to −0.3 mm from bregma. For this sample, IEG expression in response to sucrose exposure was greatest in the regions between +1.0 and +2.0 mm anterior to bregma, and notably diminished in the more posterior regions from +0.3 to −0.3 mm from bregma (Supplementary Fig S2). Given this information, all subsequent analyses were restricted to regions extending from +0.95 to +1.7 mm anterior to bregma, and we referred to this circumscribed region as GC.
Samples in this experiment were taken in two separate blocks. In the initial block (n = 9; 6 in the SAME condition, 2 the DIFFERENT condition and 1 in WATER condition), samples were taken at 60× magnification resulting in a lower number of cells per sample. In this block, we collected an average of 346 ± 66 DAPI-labeled cells per animal. In the second block (n = 8; 2 in SAME, 3 in DIFFERENT, 3 in WATER), images were collected at 20× magnification, resulting in a yield of 2447 ± 117 cells per animal. No animal’s tissue was imaged at both 20× and 60×. Importantly, all IEG expression analysis was done on rates of expression, which was defined as the percentage of IEG-positive nuclei (expressing Arc, H1a or both) out of the total number of DAPI-labeled cells for each sample. Thus, because analysis was done on rates rather than total numbers of observations, data were comparable across different samples and different magnification. Using these rates there were no statistical differences between IEG expression ratios in the 60× and 20× magnification groups and as such, data from these groups were combined for statistical comparisons. Representative samples of analyzed in situ tissue appear in Fig. 1. The overall expression rates for each rat in each condition were analyzed with a repeated-measures group X test ANOVA that treated rats as a random effect.
To ensure that any effects were as a result of treatment and not to incidental differences in expression rate for Arc or H1a, we first compared the total rate of Arc to the total rate of H1a, regardless of group or stimulus order. We found no differences in the average expression rate between Arc (20.9%) and H1a (20.3%), t(16) = 0.33, p = 0.75. Thus, any differences reported here are not due to any inherent bias in expression for Arc compared to H1a. For this reason, data will be reported by treatment condition and not necessarily by order of presentation.
We first tested whether sucrose was able to drive greater IEG expression in GC than tasteless water. For these comparisons, we contrasted the total percent of cells that expressed an IEG (H1a) in epoch 1 with the total percent expressing Arc in epoch 2. When cells expressed both IEGs, these cells were added to both the epoch 1 and epoch 2 total count.
For both the SAME and WATER conditions, rats received identical solutions in both epochs. However, for the DIFFERENT condition, the order of the sucrose presentation was counterbalanced. For those subjects, we expected that the rate of sucrose-related IEG expression should not depend on order (or IEG). Low numbers of subjects in the DIFFERENT condition (two with sucrose first, three with water first) precluded statistical comparisons; however, numerically, solution-related IEG rates were similar (for sucrose, 27% when presented first, and 22% when presented second; for water, 12% when presented first and 13% when presented second). Thus to aid in analysis and display, we combined the IEG rates associated with sucrose and water into their respective “epochs”, regardless of the order in which those solutions were presented.
In general we found that sucrose was more effective at driving IEG expression in GC than water alone (Fig. 2). A 2-way ANOVA using group (SAME, DIFFERENT, WATER) and epoch revealed a significant main effect of group, F(2, 14) = 4.92, p < 0.05. This effect was carried by a significant linear trend where SAME showed the highest overall IEG rate, WATER the lowest and DIFFERENT at intermediate rates, F(1, 14) = 9.05, p < 0.01; the quadratic trend was not significant (F<1). The main effect of epoch confirmed the greater role that sucrose played in increased IEG expression, as there was reliably greater IEG expression in epoch 1 (plus combined sucrose epoch in MIXED) than in epoch 2 (plus combined water epoch in MIXED), F(1, 14) = 8.74, p < 0.05. In both the SAME and WATER conditions, there were similar levels of IEG expression in epochs 1 and 2. Therefore, the epoch effect was primarily due to increased IEG expression in the sucrose epoch in the MIXED condition. In support of this assertion, there was a significant interaction between epoch and group, F(2, 14) = 6.03, p < 0.02. An a priori contrast showed that there was a significant increase in IEG expression rates in the sucrose epoch compared to the water epoch in the DIFFERENT condition, while there were no such differences between epochs in the SAME or WATER conditions, F(1, 14) = 12.05, p < 0.005. Thus, sucrose reliably drove greater rates of IEG expression than water both across conditions, and also within the same subjects when given both the sucrose and water stimuli.
In this analysis, it was possible that greater variability in the DIFFERENT group may have contributed to an overestimation of the variance and thus altered our interpretation. We tested, but found no evidence for outliers for any of the groups presented here, using a z-score of greater than 1.96 standard deviations from the mean. Indeed, it was found that the two most extreme points in the sucrose epoch for the DIFFERENT condition actually decreased the effect size (Cohen’s f = 0.88 with those points compared to 1.08 without those points), suggesting that our effect was robust even without the more extreme points.
If GC is important for encoding taste properties of stimuli, then it was expected that repeated exposure to identical tastants should recruit highly overlapping neural ensembles, while different tastants should activate separate ensembles. To assess this, we compared the rate of double labeling as a percentage of those cells expressing both H1a and Arc to all valid cells. A one-way ANOVA showed a significant main effect, F(2, 14) = 6.54, p < 0.01, Fig. 3. This effect was due to increased rates in the SAME condition compared to the DIFFERENT (Tukey: p < 0.05) and WATER conditions (Tukey: p < 0.05), though DIFFERENT and WATER were not different from each other (p = 0.92). However, it was uncertain whether low rates of double labeling in the WATER and DIFFERENT conditions were simply due to low rates of single labeling. For example, tasting water in both epochs may lead to a high degree of double labeling as a percentage of total cells activated, but the double labeling rate would be minimal because of a lower baseline of IEG activity. To test this across treatments, the rate of double labeling was normalized as the percentage of cells active in the first epoch that were subsequently reactivated in the second epoch. This Epoch 1 Reactivation Rate (E1RR) was calculated as:
E1RR that approaches 100% indicate a greater rate of overlap between ensembles, while lower numbers (near 0) show distinct ensembles. As with the overall rate, a significant effect of group, F(2, 14) = 4.88, p < 0.03 showed that E1RR was reliably greater in the SAME condition (E1RR = 53%) than in either the DIFFERENT (E1RR = 33%) or WATER (E1RR = 36%) conditions, F(1, 14) = 9.74, p <0.01. Further, despite receiving the same water solution in both epochs of the WATER condition, rates of E1RR in that condition were not greater than in the DIFFERENT condition, p > 0.5. Thus, identical flavored tastants, but not tasteless water or the production of similar consummatory responses, are able to increase reactivation rates of GC neural ensembles across multiple exposures.
In the rats that received neurotoxic lesions of BLA, tissue damage was assessed by absence of nuclear staining, gliosis and cell shrinkage (Fig. 4). In all subjects, damage was bilateral and extended rostrally to −1.8 mm and caudally −3.8 mm from bregma. Damage was focused on the basolateral and lateral nuclei of the amygdala, leaving the central nucleus largely intact (Fig. 5). Anatomical studies (Allen et al., 1991; Saddoris, 2008) indicate that this region is densely and reciprocally connected with the levels of GC that were analyzed in Experiment 1. Thus these bilateral lesions effectively remove BLA input into GC. In some cases, damage often extended both ventrally into posterior piriform cortex and laterally into perirhinal cortex. However, this damage was only seen at a few levels and failed to encompass more than partial damage to those regions at any given level. Follow-up analyses failed to show any correlation between size of damage to these regions and IEG encoding rates in GC (data not shown).
All rats in Experiment 2 quickly overcame neophobia to the novel sucrose solutions during pre-training. In the O1 vs O2 condition, no rats in either group required multiple presentations to consume the requisite amounts of sucrose in the 2-min time limit. During subsequent training (Supplemental Fig. 3), there was a significant difference in the rates of consumption for the different fluids, F(2, 24) = 11.58, p < 0.02. Rats consumed significantly less of the water when presented along with O2 than either sucrose paired with O1 or water alone, F(1, 12) = 20.23, p < 0.001, though rats consumed similar amounts of sucrose paired with O1 and water alone, F(1, 12) = 0.54, p = 0.47. However, there were no effects of training day (days 1–8), lesion (sham vs BLA) or interactions with those factors, all p values > 0.2. Finally, during the test session, overall consumption varied across condition, F(2, 11) = 16.82, p < 0.001; not surprisingly, rats in the O1 vs Sucrose condition consumed more overall than rats in the O1 vs O2 condition, F(1, 11) = 33.39, p < 0.001. Notably, rats in each condition drank similar amounts of fluid in the two test epochs. Neither the main effect of epoch, F(1, 11) = 0.85, p = 0.38, nor the interaction of epoch and condition, F(2, 11) = 0.48, p = 0.63, was significant. Thus, observed differences in IEG expression rate within each condition (below) were not attributable to differences in consumption during test.
In the O1 vs O2 condition, we collected an average of 525±86 cells in the sham-lesioned rats, and 3689±207 cells in the BLA-lesioned rats. In the O1 vs. Sucrose condition, we collected an average of 674±156 per subject. Differences in cell yields were due to differences in magnification for image capture and analysis; images for the sham-lesioned animals (both in the O1 vs. O2 and O1 vs. Sucrose conditions) were taken at 40× magnification, while tissue from BLA-lesioned rats was imaged at 20×. In addition to similar rates found across different magnifications in Experiment 1, other analyses in our lab has shown that the larger sample reduces variability across samples but does not affect the observed rates of IEG expression.
As in Experiment 1, we first analyzed the total rate of IEG expression for Arc and H1a to ensure that effects were due to treatment and not to incidental differences in the expression rates of Arc and H1a. A 2-way ANOVA compared rates of IEG expression for Arc and H1a between the three different groups (O1 vs. O2 (sham), O1 vs. O2 (BLA-lesion), and O1 vs. Sucrose), and there were no effects of IEG type (Arc: 14.28%, H1a: 14.36%; F(1, 11) = 0.005, p = 0.95) nor an interaction between group and IEG type, F(2, 11) = 0.55, p = 0.59. Indeed, encoding rates between Arc and H1a within each group were strikingly similar (Tukey: O1 vs. O2 (sham), p = 1.0; O1 vs. O2 (BLA-lesion), p = 0.97; O1 vs Sucrose, p = 0.97). Thus, any effects seen in each group in the experiments were due to behavioral treatment, and not to differential activity between IEGs.
Because different stimuli were used during test, data from O1 vs. O2 and O1 vs. Sucrose were considered separately. In the O1 vs. O2 conditions, we found a reliable increase in the rate of IEG expression for the sucrose-paired odor (O1) compared to the water-paired odor (O2) despite the fact that identical water solutions were consumed in both test epochs, F(1, 7) = 10.66, p < 0.02 (Fig. 6A). Indeed, this effect was robust as there was greater IEG expression in the O1 epoch compared to the O2 epoch in all five sham-lesioned rats and all four BLA-lesioned rats. However, BLA lesions appeared to moderately decrease the overall rate of IEG expression in GC compared to sham-lesioned rats, F(1, 7) = 5.90, p = 0.045, but there was no interaction between lesion status and odor identity, F(1, 7) = 1.16, p = 0.32. Indeed, sham and BLA-lesioned rats showed similar differences between the O1 and O2 epoch, t(7) = 1.08, p = 0.32. Thus, despite tasting the same water solution in both epochs, the presence of the sucrose-paired odor O1 potentiated IEG expression.
In a separate analysis, IEG rates in the O1 vs. O2 sham group were compared in the GC and BLA of the same animals. Within subjects, GC neurons were more likely to show potentiated expression for O1 over O2, while BLA neurons showed similar rates of expression for both O1 and O2 (Supplementary Fig. 4). Thus, there was no main effect of region (GC vs BLA) or stimulus (O1 vs O2), but there was a significant interaction, F(1, 3) = 12.84, p < 0.05. These findings indicate that BLA and GC play complementary roles in the encoding of tastants in associative representations.
A different pattern of results emerged in the O1 vs Sucrose condition. Here, despite tasting two discriminable solutions in the two epochs (water and sucrose), IEG expression in the O1 epoch was not reliably different from that in the sucrose alone epoch, t(4) = 1.24, p = 0.28, Fig. 6B. Indeed, these rates were numerically similar to those found in Experiment 1 in which rats sampled sucrose alone (25% in Experiment 1 and 20% in O1 vs. Sucrose). Thus, a sucrose-associated odor was able to drive IEG expression at rates statistically similar to that when rats taste sucrose alone.
Next we compared rates of double labeling for all conditions in Experiment 2. As noted above, double labeling is a mechanism by which to explore the extent to which ensembles of neurons overlap in activity across epochs. Data from the O1 vs Sucrose illustrates the utility of this measure. In that condition, it was possible that similar IEG expression rates driven by exposure to O1 and sucrose was carried by two different populations of neurons. For example, associative representations may reflect activity from limbic afferents, while primary gustatory responses may be driven by non-overlapping, bottom-up sensory processes. In that case, one would expect similar rates of single labeling for both O1 and Sucrose, but low rates of double labeling between those outcomes. In contrast, if associative representations of sucrose reactivate ensembles engaged in primary gustatory processing, then there should be a high degree of double labeling between O1 and Sucrose ensemble activity. Thus all data were analyzed for rates of double labeling. Recall that in Experiment 1, we found that rates of double labeling were similar in the DIFFERENT condition regardless of order of presentation, suggesting that different neural ensembles are recruited for the different tastants, rather than one tastant activating a subset of the neurons that encoded the other tastant. Thus, for this analysis in Experiment 2, we looked at overall double labeling for all three conditions (O1 vs O2: Sham, O1 vs O2: BLA and O1 vs Sucrose), regardless of whether the first epoch was O1, O2 or sucrose. We found a significant increase in the rate of double labeling in the O1 vs. Sucrose condition compared to the O1 vs. O2 conditions, F(2, 11) = 205.92, p < 0.0001, Fig. 7. Tukey unequal-n post-hoc tests revealed that such double labeling was significantly greater in O1 vs. Sucrose than in both the O1 vs. O2 sham (p < 0.001) and O1 vs. O2 BLA (p < 0.001) groups. Further, within the O1 vs. O2 subjects, double labeling in the sham group was significantly greater than in the BLA-lesioned group (p < 0.05).
As before, we suspected that lower basal rates of IEG expression might account for this difference in total double labeling between O1 vs. O2 groups. Using the E1RR index for double labeling instead of the overall rate, we again found a significant difference across the three groups, F(2, 11) = 19.35, p < 0.0005. This difference was due to almost exclusively to an increase in the E1RR for the O1 vs. Sucrose animals (E1RR = 59%) compared to either of the O1 vs. O2 groups; post-hoc comparisons show a significant increase for O1 vs. Sucrose over the O1 vs. O2 sham (Tukey: p < 0.001; E1RR = 32%) and O1 vs. O2 BLA-lesion (Tukey: p < 0.005; E1RR = 33%) groups. However, with this metric lesions of BLA did not alter the rate of double labeling compared to sham controls in the O1 vs. O2 condition, Tukey: p = 0.98. Thus, differences in double labeling with overall rates between the two O1 vs. O2 groups were due to a general decrease in overall IEG expression rather than a lesion-induced alteration in content of the GC associations. This argues that BLA lesions may modulate the magnitude of representational encoding in GC while leaving intact the content of those representations.
Previous work has shown that GC plays a role in both the perception and memory for tastants. Here we used a 2-IEG in situ hybridization technique to characterize coding in ensembles of GC neurons. First, a tastant (sucrose) drove greater expression of plasticity-related IEGs in GC than tasteless water. Second, when sucrose was consumed in two different test epochs, overlapping ensembles of neurons were recruited in encoding those tastants, whereas dissimilar tastants or water alone failed to re-engage those ensembles to the same extent. Third, odor cues associated with sucrose provoked GC IEG expression similar to that induced by sucrose itself. Despite consuming only water in both epochs, rats in the O1 vs. O2 condition in experiment 2 showed a greater percentage of neurons recruited when presented with a previously sucrose-paired odor than with a previously water-paired odor. Moreover, in rats in the O1 vs. Sucrose condition in experiment 2, presentation of a sucrose-paired odor and sucrose itself recruited similar ensembles of neurons. Thus, features that were coded for in a primary gustatory stimulus were also captured in associatively-activated representations of that tastant. Finally, although many studies have implicated BLA in the encoding of outcome-expectant representations, BLA lesions failed to alter the content of these associative GC representations.
These findings provide evidence that IEG markers of plasticity track important aspects of both primary and associative taste encoding. This is consistent with the results of other studies, which demonstrate the importance of protein synthesis in GC for the maintenance of taste-associated learning. For example, infusions of NMDA or AMPA receptor antagonists in GC prevent the formation and maintenance of CTA (Yasoshima et al., 2000). Mirroring these observations, recent work has shown a critical role for NMDA-dependent Arc in neural plasticity by regulating the number of AMPA receptors at locally activated synapses (Park et al., 2008; Waung et al., 2008). Thus, the present experiments provide evidence to support the expression of these IEGs during learning about taste and taste-associated cues.
Compared to some previous experiments in which this two-gene technique was employed, we found a much lower rate of coexpression across epochs given identical tastants. For example, Vazdarjanova et al. (2002) found that repeated exposure to an identical context induced E1RR that was greater than 90% in parietal cortex and hippocampus, while repeated exposure to identical sucrose solutions or sucrose and a sucrose-related cue in the present experiment induced a much lower 50% reactivation rate. These data, however, are in line with previous electrophysiological studies which have shown that taste-responsive neurons in GC have broadly tuned response characteristics (Katz et al., 2001, 2002; Stapleton et al., 2006). Indeed, the majority of neurons recorded in GC do not appear to be selective for chemosensory taste properties. As few as 14% of cells using averaging techniques (Yasoshima and Yamamoto, 1998) and as many as 41% using more sensitive temporal analyses (Katz et al., 2001; Soares et al., 2007) selectively encoded taste information. Thus, such broad tuning as well as the integration of somatosensory and taste information inherent in the region may limit the sensitivity of IEG ensemble analysis in GC. A recent study that examined CTA learning in GC and BLA with Arc reported similar overall rates of overall encoding (7–10%) and double labeling (~4%) in these regions when presented with taste-illness pairings (Barot et al., 2008). Thus, our data are consistent with mapping the ensembles selective for taste information even within the limits of such a system.
GC appears to be important not only for encoding tastants, but also the associative representations between cues and the motivational value of specific tastants. Previous studies have shown that GC lesions interfere with performance on CTA learning for multiple reinforcers (Braun et al., 1981) or a selective-satiation devaluation task where rats must use such associative representations to guide outcome-specific operant responding (Balleine and Dickinson, 2000). In these cases, GC lesions do not interfere with the ability for animals to acquire accurate learning about either cues or operant responses, but rather impair the ability to use a learned representation of outcomes to guide responding. Indeed, evidence suggests that both GC and BLA may be involved with a form of mediated learning known as taste-potentiated odor aversion (TPOA), in which illness is paired with a cue-evoked representation of food. While normal rats show a decrease in the amount of fluid consumed in the presence of the odor CS alone (Palmerino et al., 1980), lesions or inactivation of GC or BLA cause impairments in TPOA (Lasiter et al., 1985; Hatfield et al., 1992; Ferry et al., 1995; Hatfield and Gallagher, 1995) while training in this procedure in normal rats produces increased IEG activity in GC and BLA (Dardou et al., 2006, 2007). Thus, the present data support this role for GC and BLA for playing a role not only in taste perception, but also associative representations of tastes and taste memory.
Although other studies have shown increased activity in GC in the presence of taste-predictive cues, those studies have used between-subjects designs to explore differences between different treatment conditions. By using a within-subjects design in the present study, we found greater activity for the sucrose-paired cue than the water-paired cue. However, it is not clear that this potentiated IEG expression was necessarily due to taste-related processing. For example, O2 was always paired with water during training and also during test, while O1 was paired with sucrose in training but surprisingly paired with water in test. Thus, the increase may instead be capturing the novelty of the new situation compared to the well-predicted nature of the O2 epoch.
Data from the O1 vs. Sucrose condition argue against this interpretation. Here we saw that sucrose-paired O1 was able to drive IEG expression in GC ensembles as much as sucrose itself, a solution that rats had received many times over the course of training and pre-exposure. Indeed, the rate of coexpression across these epochs was nearly the same as when rats had sucrose in both epochs as in Experiment 1. This suggests that at the level of GC, tastants and associative representations of those tastants are encoded by the same ensembles of neurons. One objection to these findings is that sucrose is more hedonically motivating than water, even to thirsty rats. At the concentrations used, we always found a preference for the sucrose solution compared to water when the rats had a 2-bottle forced choice. Thus, GC representations may simply be encoding the rewarding value of the expected outcome. Previous lesion data suggest that GC is important for assigning a motivational value to specific tastant representations (Balleine and Dickinson, 2000). However, future studies will be required to distinguish the hedonic and sensory-specific aspects of these GC associative representations.
Surprisingly, lesions of BLA failed to abolish this potentiated IEG expression in GC. Many studies have shown that BLA plays an important role in assigning value to cue-outcome associations. In many learning paradigms, electrophysiological recordings show that BLA neurons develop cue-selective firing even before animals are making accurate choices (Schoenbaum et al., 1999; Saddoris et al., 2005), while BLA lesions impair the ability of OFC neurons to develop firing patterns that reflect learned cue-outcome associations (Schoenbaum et al., 2003). This associative encoding in BLA appears to be specific to the sensory properties of the reinforcer, as lesions of BLA prevent animals from avoiding a lever paired with a devalued outcome (Balleine et al., 2003) or potentiating a specific operant response that shares an outcome with a presented Pavlovian cue (Blundell et al., 2001; Corbit and Balleine, 2005).
Those studies suggest a critical role for BLA in developing sensory-specific associative representations that can be used in guiding behavior in the absence of primary reinforcers. Nonetheless, in the present studies, GC ensembles in BLA-lesioned rats showed a similar increase in IEG expression for O1 compared to O2 as in sham-lesioned animals, suggesting that lesioned animals were able to form odor-taste associations in the absence of BLA input. Indeed, some studies have shown that rats are able to form specific cross-modal associations between neutral stimuli without an intact BLA (Blundell et al., 2003; Dwyer and Killcross, 2006), a feature that is similar to the stimulus-stimulus associative functions in adjacent perirhinal cortex (Nicholson and Freeman, 2000). BLA, then, appears to be engaged when assigning the motivation significance to associations formed in the cortex. However, the overall decreased rate of IEG expression in BLA lesioned rats might suggest that BLA plays a modulatory role in GC, diminishing but not eliminating the content of associative GC ensembles. Thus if associative outcome information in GC is indeed intact following BLA lesions, then BLA may not be essential for the encoding of outcome-specific representations in the brain.
Previous findings showing such dependence may consequently reflect an interdependence of BLA with GC input (Blundell et al., 2001; Balleine et al., 2003; Corbit and Balleine, 2005), and that specifically, GC-generated representations may need to be communicated to BLA for assignment of motivational significance. We will explore this possibility in future experiments.
Figure S1. A. Fluid consumption over training for the O1 vs. O2 and O1 vs. Sucrose rats. Pairing in the abscissa represents the nth time a rat received that solution. In general, rats drank more solution in the O1/sucrose epochs than in the O2/water epochs, F(1, 15) = 14.82, p < 0.005. There was also a general effect of group, as a post-hoc contrast revealed a general increase in fluid consumption for O1 vs. Sucrose compared to either BLA or SHAM lesioned animals. This may have been an effect of surgical treatment, as the O1 vs. Sucrose rats were naive at the time of testing. B. Consumption rates were similar across epochs in all testing groups. In the O1 vs. O2 rats, rats drank similar amounts of watern when paired with either O1 or O2. Likewise, O1 vs. Sucrose rats drank as much water when paired with O1 in one epoch as unscented sucrose in the other epoch.
Figure S2. A. Fluorogold (FG) injection site in BLA (deposit marked in red). B–E, Photomicrographs of insular cortex in the immunohistochemically processed tissue after FG injection into BLA. Subregions of insular cortex are labeled. Note that labeling the granular (GI) and dysgranular (DI) cortex is present in the anterior sections (B–D), but is not present in more posterior sections (E). Number in bottom right is mm anterior to bregma. F. Sample tracer injection into GC (deposit outlined in red). G–H. Retrograde FG-positive neurons are present in the basolateral (BLA) but not lateral nucleus (LA) of the BLA. Labeling was predominantly found in anterior sections of BLA. I. One injection of cholera toxin b was placed in GC at +0.00 anterior to bregma (not shown). Dense staining of fibers and cell bodies was found in central nucleus (CN) and not in BLA. These data suggest an anterior-posterior division in insular cortical connections with the amygdala.
Figure S3. Average IEG expression rate for a single subject. Rates are total combined percent of cells in the SAME condition expressing at least one IEG at each anatomical level (positive numbers are mm anterior to bregma, negative are posterior to bregma). Error bars are standard deviation for all samples scored at that level. In general there was significantly greater overall expression in the anterior sections of GC (from +1 to +2 AP) than in the posterior sections (−0.3 to +0.3 AP). Interestingly, this difference corresponded to a different pattern of connectivity in GC and amygdala, with more anterior sections (+1 to +2) preferentially projecting to BLA and posterior sections (+0.0) projecting preferentially to CN. We used these parameters – connectivity with BLA and IEG expression rates – as a metric for selecting our region of interest in GC. Anterior and posterior GC may thus contribute differently to different aspects of taste, reward and motivation.
Figure S4. IEG expression rates from animals in the O1 vs. O2 experiment in which one odor (O1) was paired with sucrose while another odor (O2) was paired with water. During the time of test, rats received water in both epochs, one paired with O1 and the other with O2. A. Data shown are for individual animals that had both GC and BLA tissues analyzed for IEG expression. Thus, GC data here are similar to those found in Fig 3 but with one less subject. A 2×2 repeated-measures ANOVA showed no main effect of either brain region (GC or BLA) or odor (O1 vs O2). However there was a significant interaction between region and odor, F(1, 3) = 12.84, p < 0.05. As expected, this interaction was due to a significant difference between O1 and O2 in GC (Tukey HSD, p < 0.05), but no difference between O1 and O2 in BLA (p = 0.99). B. Rates of double labeling varied across region as well. GC neurons showed a significantly higher rate of double labeling compared to BLA within subjects, t(3) = 5.15, p < 0.02. Thus, GC and BLA encode different associative aspects of expected outcomes. Consistent with these results, GC may use distributed and overlapping ensembles of activity to encode learned taste-related information, while BLA may use more discrete ensembles to encode information about separate motivationally salient cue-outcome associations. * p < 0.05 within-subjects t-tests.
This work was supported by grants from the NIMH grant MH 060179 (MG) and NSF (MPS). We thank Michael McDannald, Hongjoo Joanne Lee, Gorica Petrovich and Weidong Hu for excellent technical assistance with in situ hybridization and immunocytochemistry, and Geoffrey Schoenbaum, Amy Shelton, Paul Worley, Steven Hsiao and Charles Pickens for helpful discussion and comments on earlier drafts of this manuscript.