Anterior Insula–Associated Social Novelty Recognition: Pivotal Roles of a Local Retinoic Acid Cascade and Oxytocin Signaling

Data were analyzed using the statistical methods stated in each figure legend, and all statistical analyses were performed using SPSS. Sample sizes were chosen on the basis of previous studies. When parametric tests such as t tests and analyses of variance (ANOVAs) were performed, the assumptions of normality (normal distribution) and homogeneity (similar variance among groups) were tested using Shapiro-Wilk’s test and Levene’s test, respectively. Data were mostly analyzed by one-way ANOVAs, two-way ANOVAs, and two-way repeated-measures ANOVAs followed by post hoc test. One-way ANOVAs followed by Tukey’s multiple comparisons were performed for testing statistical differences between means of three or more independent groups. Two-way ANOVAs followed by Bonferroni post hoc tests were used when examining the influence of two different factors on a dependent variable. However, if any repeated factor was present, two-way repeated-measures ANOVAs followed by Bonferroni post hoc tests were used. For simple pairwise comparisons, Student’s t tests were used.

We first investigated whether the AI is a critical brain area for social cognition in mice. We introduced bilateral excitotoxic lesions in either the AI or its adjacent brain region, the orbitofrontal cortex (OFC), in C57BL6/J wild-type male mice (see Figure S1 in the online supplement). In the three-chamber social interaction test, normal mice prefer the enclosure with a social subject (mouse) (trial 1: sociability) and show a selective preference to a novel mouse compared with a familiar mouse (trial 2: social novelty recognition) (Figure 1A). We observed a deficit only in trial 2 in AI-lesioned mice, whereas OFC-lesioned mice behaved normally in both trials (Figure 1A). Thus, AI-lesioned mice, but not OFC-lesioned mice, exhibited a significant deficit in social novelty recognition but not in sociability in this assay. To address whether the deficit was specific to the social context, we employed a five-trial social novelty recognition paradigm guided by social or nonsocial cues. Both OFC-lesioned and AI-lesioned mice displayed normal behaviors for nonsocial cue (odor): gradual reduction of exploration time to a banana scent during four consecutive trials, but substantially increased exploration when the scent was changed to a strawberry scent on the fifth trial (Figure 1B). In the context of social cue, the OFC-lesioned mice also displayed a normal response (Figure 1C). In contrast, AI-lesioned mice showed significant deficits in response to social cue: absence of habituation during the first four trials and no increase in exploration when exposed to a new conspecific cue in the fifth trial (Figure 1C). To avoid the possibility that these AI-specific deficits are associated with olfactory function or locomotor activity, we examined the hidden food test and open field test and showed that these are not confounding factors (see Figure S2A,B in the online supplement). In addition, AI-lesioned female mice behaved similarly to male mice, displaying a deficit in social novelty recognition (see Figure S2C in the online supplement). Together, these results indicate that the AI plays a role in social novelty recognition in mice, which clearly differentiates it from the OFC.

To study the molecular, cellular, and circuitry mechanisms that underlie this AI-mediated behavior, we looked for a lead molecule to build a useful mouse model at both the basic and translational science levels. We hypothesized that the target would be enriched in the layer 5 pyramidal neurons of the AI, whose gene is significantly highlighted in genome-wide association studies (GWASs) for brain disorders. We combined the information from our own microarray data of fluorescence-activated cell sorting (FACS)-isolated AI layer 5 pyramidal neurons (see the supplementary methods section in the online supplement for details) and an anatomic gene expression atlas (34), which were cross-referenced with the GWAS risk genes so that we highlighted 47 candidate molecules (see Figure S3A in the online supplement). Following successful precedents in which mouse models that genetically modify enzymes critical to the biosynthesis of key mediators (e.g., serine racemase for d-serine and nitric oxide synthase for nitric oxide) have been highly useful in exploring neurobiological mechanisms (35, 36), we further narrowed the candidates down to two molecules that fall in this category, Cyp26B1 (cytochrome P450 26B1, a retinoic acid [RA]–degrading enzyme) (37, 38) and Ckb (creatine kinase brain type) (see Figure S3A in the online supplement). Cyp26B1 has been highlighted by an analysis using a comprehensive transcriptome study of postmortem brains from patients with autism spectrum disorder, schizophrenia, and bipolar disorder (see Figure S3B in the online supplement) (39). Additionally, RA signaling is one of the top four signaling cascades in which risk genes for brain disorders are enriched (Figure 2A). Furthermore, we observed that a chronic infusion of RA locally into the AI resulted in a phenocopy of the AI lesion, exhibiting a significant deficit in social novelty recognition (trial 2) but not in sociability (trial 1) in the three-chamber social interaction test (Figure 2B; see also Figure S4A in the online supplement). Finally, we experimentally confirmed that Cyp26B1 is almost exclusively expressed in the pyramidal neurons at the deep cortical layers (layers 5 and 6) of the AI (Figure 2C–E). We therefore decided to study the unique role of the AI in social novelty recognition by using RA signaling, in particular Cyp26B1, as a molecular lead.

To directly examine the role of Cyp26B1 in the AI on social behavior, we bilaterally knocked down the expression of Cyp26B1 in the AI by adeno-associated virus (AAV)–mediated RNAi (AAV-U6-shCyp26B1-CMV-mCherrry [AAV-Cyp26B1 KD]) (see Figure S4B in the online supplement). Efficient knockdown (KD) of Cyp26B1 and a reciprocal augmentation of RA signaling were confirmed in vitro and in vivo at mRNA and protein levels (see Figure S5A–D in the online supplement). Consistent with the effect of local RA infusion into the AI (Figure 2B), Cyp26B1 KD in the AI (AI-Cyp26B1 KD) strongly induced a deficit in social novelty recognition without affecting sociability in the three-chamber social interaction test, whereas control mice (AI-Con) injected with AAV-U6-shControl-CMV-mCherrry (AAV-Con) displayed typical social behaviors (Figure 2F). Restoring the expression of Cyp26B1 by introducing RNAi-resistant Cyp26B1 (Cyp26B1^R) with AAV-hSyn-Cyp26B1^R-HA (AAV-Cyp26B1^R) in the AI rescued the deficit in social novelty recognition elicited by Cyp26B1 loss-of-function (Figure 2F). The AI-Cyp26B1 KD mice also displayed recognition deficits in the context of social but not nonsocial cues (see Figure S5E in the online supplement). These data indicate the direct involvement of Cyp26B1 in the AI in social novelty recognition. To confirm and extend the results obtained from the RNAi approach, we used Cyp26B1 conditional knockout mice (40): we selectively depleted Cyp26B1 in αCaMKII-positive cells in the AI by injecting AAV-αCaMKII::GFP-Cre in the AI of Cyp26B1-floxed (Cyp26B1^flox/flox) mice (see Figure S4C in the online supplement). The genetic depletion of Cyp26B1 in αCaMKII-positive cells (pyramidal neurons) of the AI led to deficits in social novelty recognition (Figure 2G), which is consistent with the phenotype in AI-Cyp26B1 KD mice.

Because RA signaling is primarily involved in transcriptional regulation, we hypothesized that Cyp26B1 KD/KO affected these behavioral phenotypes via altered gene transcription. Thus, we performed an unbiased molecular profiling analysis of FACS-enriched AI cortical layer 5 pyramidal neurons. There was a small subset of differentially expressed genes between the control and KD groups (false discovery rate <0.1), while the expression profiles of control and rescue groups were similar, with good reproducibility among biological triplicates (Figure 3A). The majority of the genes we identified (15 of 20) included a retinoic acid response element (41) and/or were known to bind with retinoic acid receptors (RARs) (42) (see Table S1 in the online supplement), suggesting that these genes are likely to be regulated by RA signaling. Results from quantitative real-time PCR analyses were consistent with the molecular profiling data (see Figure S5F in the online supplement). Bioinformatic analyses strongly suggested that chronically upregulated RA signaling might cause abnormal dendritic spine regulation in the AI layer 5 pyramidal neurons (Figure 3B). Consistent with this prediction, we experimentally confirmed that AI-Cyp26B1 KD caused a significant reduction in spine density in the proximal apical dendrites, compared to the control group, while the expression of Cyp26B1^R effectively normalized this deficit (Figure 3C). These results suggest that the decreased expression of Cyp26B1 was the sole cause for the spine deficit.

We also validated the spine changes at the electrophysiological level. The AI layer 5 pyramidal neurons in acute slices from AI-Cyp26B1 KD mice exhibited a significantly lower frequency of spontaneous excitatory postsynaptic currents (sEPSCs) compared to those from control mice, whereas both groups of neurons showed no difference in amplitude (Figure 3D). The decrease in sEPSC frequency in the AI layer 5 pyramidal neurons implies that there would be reduced activity in these neurons and their downstream targets. In contrast, acute bath application of RA to brain slices from wild-type mice did not affect the sEPSCs of the AI layer 5 pyramidal neurons (Figure 3E). The results of electrophysiological experiments taken immediately after the acute RA application may reflect acute changes in intracellular signal transduction but are unrelated to chronic influences elicited by RA, such as gene transcription-mediated spine changes. Accordingly, the contrastive results in electrophysiology between acute and chronic applications of RA imply that electrophysiological changes in AI-Cyp26B1 KD mice are likely to reflect a functional outcome associated with the spine changes. Taken together, these findings show that proper control of RA signaling is critically required for the maintenance of dendritic architecture and activity of the AI layer 5 pyramidal neurons. Thus, we optogenetically activated the AI pyramidal neurons during the three-chamber social interaction test in mice infected with Cyp26B1 KD AAV and channel rhodopsin (ChR2)-EYFP AAV (see Figure S6 in the online supplement). Optogenetic excitation of the AI pyramidal neurons did not affect sociability (trial 1) in the AI-Cyp26B1 KD mice (Figure 3F). In contrast, optogenetic excitation in the AI restored proper social novelty recognition through a decrease in time spent with a familiar mouse (trial 2) (Figure 3F). These data suggest that modulation of spontaneous neuronal activity of the AI layer 5 pyramidal neurons could affect social novelty recognition.

Next, we looked for circuit mechanisms by which the activity of AI pyramidal neurons is properly maintained in social contexts. We examined plausible afferent projections to the AI by using a retrograde tracing method in which we injected cholera toxin subunit B-conjugated with Alexa 488 (CTB-Alexa 488) into the AI (Figure 4A). As expected, we observed a strong signal from the contralateral AI and basolateral amygdala, well-known brain regions that tightly connect with the AI (Figure 4B) (13, 43). Then we assessed possible projections to the AI from the hypothalamic paraventricular nucleus (PVN) and supraoptic nucleus (SON), where OT neurons originate (44). A recent publication reported that OT could directly influence social affective behavior mediated by the posterior insula (PI) (33). Nevertheless, it remains elusive how OT affects the function of the AI. It is well known that the AI and PI (see Figure S1A in the online supplement) are quite distinct anatomically and functionally in many aspects (45, 46). Several reports have indicated that the projection of OT neurons to the AI is quite sparse or negligible (24, 30). Our experimental data were indeed in accordance with these reports, indicating that there are no detectable projections to the AI from the hypothalamic PVN and SON (Figure 4C).

If OT does not influence the AI neurons via direct projection of OT neurons, another scenario is via diffusion of OT into the AI. To test this possibility, we locally infused OT into the AI of AI-Cyp26B1 KD mice and examined whether exogenous OT could ameliorate the AI-associated social novelty recognition deficits (see Figure S7A in the online supplement). Local infusion of OT into the AI of AI-Cyp26B1 KD mice did not alter their deficit in social novelty recognition (Figure 4D). We also locally infused desGly-NH₂-d(CH₂)₅ [D-Tyr², Thr⁴] OVT, a selective OTR antagonist (OTA), into the AI of control mice and did not observe any changes in social behavior (Figure 4D; see also Figure S7A in the online supplement). Nevertheless, intracerebroventricular (i.c.v.) administration of OT normalized the social novelty recognition deficit through a decrease in time spent with a familiar mouse in AI-Cyp26B1 KD mice in which RA signaling is selectively perturbed only in the AI (Figure 4D; see also Figure S7B in the online supplement). The i.c.v. administration of OTA also influenced social novelty recognition in control mice (Figure 4D; see also Figure S7B in the online supplement). These effects were selective to social novelty recognition, except i.c.v. administration of OTA, which also altered sociability (Figure 4E). Together, these results suggest that although OT specifically participates in the regulation of AI-mediated social novelty recognition, this is unlikely to be via the influence of OT directly on the AI neurons. Instead, OT may influence AI-mediated social novelty recognition indirectly, by remotely affecting the neurons that project to the AI from different brain regions.

In the retrograde tracing experiments to look for neuronal projections to the AI, we examined several brain regions in addition to the PVN and SON. Among them, we observed a distinct signal in the dorsal raphe nucleus (DRN) (Figure 4C), which is known to provide major serotonergic projections to the forebrain. Roles of serotonin receptors in the AI in higher brain functions have been suggested (47, 48). Thus, we examined how modulation of serotonergic inputs might affect AI-mediated social novelty recognition. Based on their high expression levels in the AI (http://mouse.brain-map.org), we focused on two serotonin receptors (5-HT_1A and 5-HT_2C). First we tested the effects of their agonists and antagonists by systemic injection. Only CP 809101 (a 5-HT_2C receptor agonist) normalized the social novelty recognition deficit in AI-Cyp26B1 KD mice, whereas SB 242084 (a 5-HT_2C receptor antagonist) disrupted social novelty recognition in control mice (see Figure S8A in the online supplement). To further validate whether the impact of the agonist and antagonist on behavior is directly mediated via the AI, we conducted local infusion experiments. Stimulating the AI serotonergic system via 5-HT_2C normalized the social novelty recognition deficit in AI-Cyp26B1 KD mice (Figure 5A), while inhibition of 5-HT_2C in the AI of control mice impaired social novelty recognition (Figure 5A). Treatment with the 5-HT_2C receptor agonist increased both the amplitude and frequency of sEPSCs in AI-Cyp26B1 KD pyramidal neurons without affecting presynaptic features (Figure 5B; see also Figure S8B in the online supplement). These data suggest that the 5-HT_2C receptor agonist ameliorated deficits in the AI layer 5 pyramidal neurons of AI-Cyp26B1 KD mice.

We have shown that AI-mediated social novelty recognition is properly maintained by serotonergic regulation via 5-HT_2C receptors expressed on AI pyramidal neurons. Meanwhile, the same social behavior is also regulated by OT, which likely plays a role in the AI by remotely affecting the neurons that project to the AI from different brain regions. Given that serotonergic innervation in the AI is from the DRN, one possible hypothesis is that OT directly influences the neurons in the DRN, which in turn regulates the activity of the AI pyramidal neurons. Indeed, OT neurons from the PVN reportedly project to the DRN (25), where OT receptors are reportedly highly expressed (49). To test our hypothesis, first we infused OTA specifically into the DRN of control mice and tested its effects on social novelty recognition. We observed that this treatment selectively impaired social novelty recognition (Figure 5C; see also Figure S9 in the online supplement). To test whether interfering with OT signaling in the DRN is specifically associated with the AI-mediated social novelty recognition, we infused OT specifically into the DRN of AI-Cyp26B1 KD mice. OT infusion into the DRN significantly restored social novelty recognition (Figure 5C; see also Figure S9 in the online supplement).