Role of BDNF in the development of an OFC-amygdala circuit regulating sociability in mouse and human

Abstract

Social deficits are common in many psychiatric disorders. However, due to inadequate tools for manipulating circuit activity in humans and unspecific paradigms for modeling social behaviors in rodents, our understanding of the molecular and circuit mechanisms mediating social behaviors remains relatively limited. Using human functional neuroimaging and rodent fiber photometry, we identified a mOFC-BLA projection that modulates social approach behavior and influences susceptibility to social anxiety. In humans and knock-in mice with a loss of function BDNF SNP (Val66Met), the functionality of this circuit was altered, resulting in social behavioral changes in human and mice. We further showed that the development of this circuit is disrupted in BDNF Met carriers due to insufficient BDNF bioavailability, specifically during a peri-adolescent timeframe. These findings define one mechanism by which social anxiety may stem from altered maturation of orbitofronto-amygdala projections and identify a developmental window in which BDNF-based interventions may have therapeutic potential.

Introduction

Social behavioral deficits are key features shared among many major psychiatric disorders, including autism, schizophrenia, depression, and anxiety disorders [1]. While the prevalence of social deficits has made social behavior and its underlying circuits desirable targets for translational research, the intrinsic complexity of social deficits has made understanding of their underlying neural correlates difficult. Indeed, social deficits in humans can manifest in numerous ways such as anxiety towards specific or general social situations [2–4], social anhedonia [5, 6], compromised social memory and recognition, and excessive aggression [7]. In addition, although structural imaging analyses, facial perception-based functional analyses, and meta-analyses in humans have identified numerous cortical and subcortical regions associated with various types of social deficits [8–10], the heterogeneity of human populations combined with the relatively low spatial resolution of the human neuroimaging techniques have complicated the interpretation of those results.

Recent advances in optical tools such as optogenetic techniques in rodents have allowed for detailed circuit-level investigations of neural correlates underlying social behaviors [11, 12]. Pathways involving regions such as basolateral amygdala (BLA), ventral hippocampus [13–15], ventral tegmental area (VTA) [16], nucleus accumbens (NAc) [16, 17], and prefrontal cortex (PFC) [18, 19] have been found to modulate social interactions in rodents. However, it is unclear how these circuits are related to the different types of social deficits seen in humans. Given the intrinsic discrepancies between human and rodent social behaviors, mapping rodent behaviors to relevant human deficits has been challenging. While various rodent task designs have enhanced our ability to understand and translate rodent social behaviors [20, 21], cross-referencing results from parallel human and rodent models remain one strategy by which social behavioral changes in human are matched with the circuit level precision of the rodent models [22].

Using human and parallel rodent models linked by a common molecular level change, we identified a specific orbitofrontal (OFC)-amygdala projection that plays a critical role in perception of social cues and is disrupted in carriers of BDNF Val66Met allele (rs6265), a common SNP leading to decreased BDNF-regulated secretion [23] and associated with risk of anxiety disorders [22]. In human adults performing an emotional face perception task, amygdala and OFC activities were altered in opposite directions in BDNF Met carriers in response to threatening social cues, suggesting an inhibitory influence of OFC activity on amygdala. Similar reductions in activity were observed in BLA projecting medial OFC (mOFC) neurons in adult BDNF Val66Met knock-in mice (BDNF^Met/Met) compared to wild-type (WT) mice (BDNF^Val/Val) using fiber photometry during social novelty investigation. Behaviorally, the changes in neuronal activity were paralleled by increased self-reported social anxiety scores in human adult BDNF Met carriers and decreased social novelty approach in adult BDNF^Met/Met mice. Importantly, suppression of mOFC-BLA activity in adult WT mice using chemogenetic tools recapitulated the behavioral phenotype of BDNF^Met/Met mice, establishing a necessary role of amygdala-projecting OFC neurons in supporting social functions. Finally, diffusion tensor imaging (DTI), neuroanatomical tracing, and viral BDNF conditional overexpression and knockout strategies indicated that the maturation of the mOFC-BLA circuit can be influenced bi-directionally by the availability of BDNF during adolescence. While a reduction in BDNF bioavailability—such as in BDNF Met allele carriers—can compromise the integrity of mOFC-BLA projections and confer susceptibility to social deficits, BDNF-based interventions during a defined window in adolescence may offer long-term benefit to susceptible populations with regard to social anxiety behaviors.

Results

BDNF Val66Met SNP leads to social deficits in both humans and mouse

Cross-species analyses were performed to elucidate the impact of the BDNF Val66Met SNP on social behaviors in adulthood. We performed rodent behavioral analyses using BDNF^Met/Met mice and WT littermate controls (BDNF^Val/Val). We utilized a rodent home-cage social interaction task to quantify social interactions initiated by BDNF^Val/Val (n = 10) or BDNF^Met/Met mice (n = 10) with a novel WT conspecific. In this task, BDNF^Met/Met mice showed significantly reduced levels of social approach (t(18) = 3.304, p = 0.004) compared to BDNF^Val/Val mice when engaging novel interaction partners, but not familiar partners (Fig. 1a). Changes in social behavior were dissociated from novelty-induced investigations by exposing the animals to a novel object in a separate trial. There was no significant genotypic difference in the novel object investigation trials (Fig. 1b). Furthermore, because previous studies suggested a strong association between general anxiety and BDNF Val66Met polymorphism [23], the lack of behavioral differences in object investigation also suggested that there was no significant change in general anxiety to drive the social changes in our current experimental setting. Similarly, the novel WT non-littermate interaction partners approached experimental animals equally regardless of genotype (t(18) = 1.295, p = 0.211; 10 VV, 10 MM; Fig. S1), indicating that the social changes are intrinsic to the experimental animals, unaffected by the sociability of the WT partners. No significant differences were found in fear-related behaviors (freezing: t(18) = 0.9596, p = 0.350; 10 VV, 10 MM), sexual behaviors (mounting: t(18) = 0.5529, p = 0.5871; 10 VV, 10 MM), and social interaction frequencies (t(18) = 0.5529, p = 0.5871; 10 VV, 10 MM) (Fig. S2). The presence of a significant difference in interaction duration but lack of significant difference in interaction frequency indicates an overall shorter interaction duration per bout for BDNF^Met/Met mice compared to BDNF^Val/Val mice, which can indicate social anxiety-like behavior. However, no statistical significance was observed for duration per bout (t(18) = 1.387, p = 0.1824; 10 VV, 10 MM; Fig. S2).

Fig. 1.

Change in social behaviors in both BDNF^Met/Met mice and human BDNF Met carriers. a BDNF^Met/Met mice showed decreased social novelty approach behaviors compared to BDNF^Val/Val mice (t(18) = 3.304, p = 0.004; 10 VV, 10 MM). This genotypic difference was not seen during littermate interaction. b BDNF^Met/Met mice did not display changes in object investigation behaviors compared to BDNF^Val/Val mice. c Human BDNF Met carriers reported significantly increased levels of self-reported social anxiety compared to Val/Val participants (F(1,47) = 4.724, p = 0.034; 34 VV, 28 MM/VM). *p < 0.05 and **p < 0.01. Error bar = 1 SEM

The three-chamber social interaction task was used to further evaluate sociability and social novelty preference [24]. While no genotypic difference was observed in sociability trials (novel object vs. novel social partner), BDNF^Met/Met mice showed reduced social novelty preference when allowed to interact with a novel social partner and a familiar social partner (Fig. S3). While BDNF^Val/Val mice spent significantly more time with novel social partner compared to the familiar partner in social preference trial (corrected p = 0.037), interaction times did not differ significantly in BDNF^Met/Met mice (corrected p = 0.3215). The lack of significant change in sociability and the presence of genotypic difference in social novelty preference indicated that while BDNF Met allele does not significantly alter overall sociability per se, it can specifically affect social memory or social novelty approach. To distinguish a social memory deficit from reduced novelty preference, a littermate was used as the familiar partner to mitigate the impact of social memory on behaviors in the social novelty preference trial. Previous studies have shown that BDNF^Met/Met animals are able to recognize and respond to a non-littermate animal in resident-intruder task [23], indicating that BDNF^Met/Met animals are at least able to distinguish littermates from non-littermates. While using a littermate as familiar interaction partner does not eliminate the possibility of BDNF^Met/Met animals exhibiting a social recognition memory deficit, it largely removes the need for it from the task. No social novelty preference was seen in the BDNF^Met/Met mice during the social preference trial with littermate, suggesting a deficit in social novelty approach independent of social recognition memory (Fig. S3).

To constrain our interpretation of rodent social behavioral changes, we analyzed a self-report of social anxiety symptoms in a subset of adult participants (n = 62) from the Pediatric Imaging, Neurocognition, and Genetics (PING) Study [25]. The detailed inclusion criteria and data collection process can be found in Methods section or PING study website, and a summary of different human analyses in this paper can be found in Supplemental Table 1. There was a main effect of genotype with human BDNF Met carriers reporting significantly higher social-anxiety sub-score scores (F(1,47) = 4.724, p = 0.034) when controlling for gender, site, general anxiety score (composite score), and ethnic background (Fig. 1c). Contributions of covariates are shown in Supplemental Table 2 and other sub-scores such as panic disorder and traumatic stress disorder scores are shown in Supplemental Table 3. While the generalized anxiety score, previously reported in association with BDNF Met allele, had a major contribution to social anxiety score, the genotypic effect was independent of generalized anxiety. In addition, since the sample consisted of healthy individuals with many subjects reporting a social-anxiety score of zero, analysis was repeated after removing participants who reported zero for social anxiety to prevent the result from being distorted by a disproportionate amount of non-anxious individuals in each group. The removal of non-anxious individuals (final n = 47) did not alter the results (F (1,32) = 5.644, p = 0.0237) (Table S2). These results are consistent with the rodent findings, highlighting a social-specific change in BDNF Met carriers. More importantly, the human results serve to constrain the interpretation of this social change, suggesting that the lack of social novelty approach seen in BDNF^Met/Met is possibly a reflection of social anxiety-like behavior, not lack of interest in social novelty. In this context, previous studies have also shown no deficit in sucrose preference and object novelty preference test in the BDNF^Met/Met mice [26, 27], further suggesting that the behaviors are likely related to a deficit in social novelty approach instead of reduced interest.

In order to differentiate BDNF Met allele’s effects on social behavior from its previously reported effects on general anxiety [23], anxiogenic factors were minimized in the rodent home-cage interaction tasks by testing subjects in their home cages under infrared light (Fig. 1a), and anxiety was included as a covariate in the human questionnaire model (Table S2). Thus, we identified a robust genotypic effect in both mouse social novelty approach and human self-reported social anxiety scores that was independent of the underlying general anxiety phenotype.

Altered OFC and amygdala functions in response to aversive or novel social cues in human BDNF Met carriers and BDNF^Met/Met mice

To investigate the underlying neural basis for these genotypic differences in social anxiety in humans, we performed task-based functional neuroimaging in healthy adult subjects (BDNF Val/Val (VV) (n = 34) vs. BDNF Met carriers (n = 21)) in a go-nogo task in which emotional faces (fear, calm, happy) were used as social response elements. Briefly, in this task, a pair of emotional cues (i.e. fear and calm) are rapidly presented in pseudo-random sequence to the subject, and the subject must follow a preset rule to click or not click a button with each presentation based on what emotion was shown. Their performance as well as blood oxygen-level dependent signals can be analyzed to see how genotype can change subject’s response to different emotional cues. Detailed experimental setup can be found in the Methods section as well as previous publications [28–30]. We chose this behavioral paradigm because it provides an established measure of social emotional processing, and prior studies suggested that patients with social anxiety disorder (SAD) exhibit altered brain activation to negative facial cues relative to neutral facial cues [31–33]. Therefore, performance and brain activity in response to fear and calm cues were analyzed for genotypic differences. Behaviorally, task performance was measured by sensitivity index d′ accounting for both hits and false alarms towards a specific cue, with higher d′ indicating better performance. Human BDNF Met carriers performed significantly worse to fear cues compared to calm cues, and this deterioration was not seen in the BDNF VV group (Fig. S4). In addition, this genotypic difference was also not observed when performing to happy cues compared to calm cues (Fig. S4). Whole-brain analysis using voxel-wise threshold of 0.005 and cluster-size corrected α of 0.05 revealed a significant interactive effect of genotype and cue-type (fear-calm) in right amygdala (23 voxels) and right OFC (106 voxels) (Fig. 2a). Specifically, we saw heightened activation of the right amygdala (MNI coordinates: x = −30.8, y = 13.0, z = 15.4, 23 voxels; F(1,53) = 14.815, p < 0.001; 34 VV, 21 MM/VM) and diminished activation of right OFC (MNI coordinates: x = 19.7, y = −18.4, z = −18.5, 106 voxels; F(1,53) = 17.916, p < 0.001; 34 VV, 21 MM/VM) in response to fear cues relative to calm cues in the Met carriers that was not present in VV subjects. Furthermore, the relative activation of amygdala and OFC towards fear cues relative to calm cues showed an inverse correlation across both genotypes (r = −0.327, t(53) = −2.55, p = 0.014) (Fig. 2b), where individuals with higher relative OFC activity exhibited lower relative amygdala activity to fear cues. Similar patterns of activity in OFC and amygdala were not observed for happy-calm contrast (Fig. S5), indicating that the effect is specific to fear cues, not general arousal.

Fig. 2.

Human BDNF Met carriers responded to fear cues differently than BDNF Val/Val, and BDNF^Met/Met mice showed diminished mOFC-BLA activity during social novelty approach. a Compared to BDNF Val/Val, BDNF Met carriers showed significantly higher amygdala activation (MNI coordinates: x = −30.8, y = 13.0, z = 15.4, 23 voxels; p < 0.001; 34 VV, 21 MM/VM) and significantly reduced OFC activation (MNI coordinates: x = 19.7, y = −18.4, z = −18.5, 106 voxels; p < 0.001; 34 VV, 21 MM/VM) towards fear cues relative to calm cues. b Across both genotypes, relative activity of amygdala to fear inversely correlates with that of the OFC (r = −0.327, p = 0.014; n = 55). c Representative GCaMP injection site and fiber placement (left bar = 200 μm; right bar = 80 μm). d Representative calcium signals across 150 s during social interaction trial. Social zone entries are indicated in red. e Average mOFC calcium signal time locked to zone entry. There was a significantly increased dF/F in BLA-projecting mOFC neurons of BDNF^Val/Val mice compared to that of BDNF^Met/Met mice during the first second after the initiation of social novelty approach (F(1,11) = 17.59, p = 0.0015; 7 VV, 7 MM, 243 zone entries), but not during littermate or object interaction (Littermate: F(1,11) = 2.366, p = 0.152; 7 VV, 7 MM, 191 zone entries; Object: F(1,11) = 0.883, p = 0.779; 7 VV, 7 MM, 312 zone entries). Each data point represents average signals of the first second from a single entry (green line). There was a significantly increased dF/F in BLA-projecting mOFC neurons of BDNF^Val/Val mice compared to that of BDNF^Met/Met mice 0.5 s prior to the initiation of social approach (F(1,11) = 5.71, p = 0.0359; 7 VV, 7 MM, 243 zone entries). Each data point represents average signals of 0.5 s prior to a single entry (purple line). Within-subject errors were controlled in all analyses. *p < 0.05 and **p < 0.01. Error bar = 1 SEM. MO medial orbitofrontal cortex, VO ventral orbitofrontal cortex

To constrain our interpretation of these human functional neuroimaging findings, we examined the mOFC in adult mice utilizing a live imaging technique, fiber photometry, which allows for measurement of in vivo calcium signals from neuronal populations [16, 34]. Importantly, through fiber photometry recordings, we were able to establish the neural activity of OFC neurons projecting to the basolateral amygdala (BLA) with temporal precision, i.e., time locked to social novelty engagement. A viral vector carrying a Cre-dependent genetically encoded calcium indicator (AAV1/Flex/GCaMP6s) was injected unilaterally into left mOFC and a retrograde vector carrying Cre-recombinase was injected into ipsilateral BLA in order to selectively express the Ca²⁺ indicator in mOFC neurons projecting to BLA. A fiber-optic cannula was implanted above the injection site to deliver an excitation light of 470 nm and record activity (emitted fluorescence) from BLA-projecting mOFC neurons during behavioral tasks (Fig. 2c). While it is technically difficult to completely eliminate viral diffusion into surrounding tissues given the size of BLA, our anatomical tracking as well as previous studies have shown minimal projections from mOFC to the surrounding CeA or MeA [35] (Fig. S6). Therefore, measured signals are highly specific to mOFC-BLA project. The raw signal was normalized to baseline fluorescence (dF/F) using a rolling window and time locked to behavior as defined by entering a zone surrounding an object cue, a littermate, or a novel social partner (Fig. 2d). When time locked to interaction zone entrance, we saw an activity increase in the mOFC-BLA projection during social novelty approach in the BDNF^Val/Val mice that was absent in the BDNF^Met/Met mice (F(1,11) = 17.59, p = 0.0015; 7 VV, 7 MM, 243 zone entries) (Fig. 2e). In addition, this mOFC-amygdala recruitment was specific to social novelty encounters, as the increase and genotypic difference were not observed during novel object investigation or littermate investigation (Object: F(1,11) = 0.883, p = 0.779; 7 VV, 7 MM, 312 zone entries; Littermate: F(1,11) = 2.366, p = 0.152; 7 VV, 7 MM, 191 zone entries) (Fig. 2e). Interestingly, the increase in neural activity as well as the genotypic difference emerged prior to the initiation of interaction as the difference in neuronal activity was present 0.5 s before zone entry (F(1,11) = 5.71, p = 0.0359; 7 VV, 7 MM, 243 zone entries) (Fig. 2e), indicating that the increased activity in BLA-projecting mOFC cells may predict the initiation of a social interaction in BDNF^Val/Val mice but not BDNF^Met/Met mice. While this observation must be interpreted conservatively due to the subjective nature of the social interaction initiation timing (zone entry), it suggests the possible role of mOFC-amygdala pathway in initiation of social novelty interactions, in addition to cue valuation during a novelty interaction. Importantly, the functional results in mice mirrors the results of the human imaging study, confirming the involvement of mOFC, and specifically, its projections to the amygdala, in regulating the response to novel social cues.

BDNF Val66Met SNP alters the anatomical integrity of OFC–amygdala projection neurons

Prior studies have shown that the BDNF Met allele alters the developmental trajectories of neuronal circuits, resulting in differences in axon density and neuronal morphologies that persist into adulthood [23, 36, 37]. To investigate whether an anatomical change in adult human BDNF Met carriers can contribute to the functional change, DTI analysis and tract-tracing studies were conducted to investigate the integrity of orbitofronto-amygdala projections in humans and mice, respectively. In PING participants, using OFC and amygdala clusters from functional analysis as seeds, probabilistic tractography was performed to identify the white matter fibers between the two regions, and structural integrity was quantified by measuring fractional anisotropy (FA). The analysis revealed a main effect of genotype in the whiter matter integrity between the OFC and amygdala, with the VV participants showing significantly higher FA than the BDNF Met carriers (F(1,147) = 7.785, p = 0.006) (Fig. 3a). Based on our threshold, individual tracts largely remained between the OFC and amygdala clusters, within uncinate faciculus (UF) known to carry majority of the frontal-limbic projection axons (Fig. 3b). No significant tracts were identified outside of UF. To confirm the directionality of the white matter tract altered by the BDNF Met allele, an anterograde tracer, phaseolus vulgaris leucoagglutinin (PHA-L), was injected into the mOFC of adult BDNF^Val/Val and BDNF^Met/Met mice. Axon fiber terminals containing the tracer were quantified in ipsilateral BLA as a measure of projection density of mOFC neurons to BLA specifically (Fig. 3d, e). Consistent with the human results, the tract-tracing experiments revealed a significantly reduced axon fiber density projecting from the mOFC to BLA in BDNF^Met/Met mice, as compared to BDNF^Val/Val mice (t(10) = 8.408, p <0.0001) (Fig. 3c). In addition, there was minimal projection from mOFC to CeA or MeA, and there was no genotypic difference in those projections (CeA: (t(10) = 0.6694, p = 0.5184; MeA: (t(10) = 0.1557, p = 0.8794; Fig. 3f).

Fig. 3.

Changes in OFC–amygdala structural connectivity in both BDNF^Met/Met mice and human Met carriers. a White matter tract between OFC and amygdala had significantly reduced FA in BDNF Met carriers compared to BDNF Val/Val (F(1,147) = 7.785, p = 0.006; 91 VV, 72 MM/VM). b Representative path for white matter tracts used for analysis. c BDNF^Met/Met mice had reduced active fiber densities from mOFC to BLA compared to BDNF^Val/Val mice (t(10) = 8.408, p < 0.0001; 6 VV, 6 MM). d Representative PHA-L injection site in mOFC (site of injection bar = 200 μm; upper bar = 40 μm; lower bar = 20 μm) and e representative fiber density in BLA (top bar = 40 μm; bottom bar = 4 μm). f There was no significant difference in mOFC-CeA (t(10) = 0.6694, p = 0.5184; 6 VV, 6 MM) and mOFC-MeA (t(10) = 0.1557, p = 0.8794, 6 VV, 6 MM) projections between BDNF^Met/Met and BDNF^Val/Val mice. *p < 0.05 and **p < 0.01. Error bar = 1 SEM. MO medial orbitofrontal cortex, VO ventral orbitofrontal cortex, LO lateral orbitofrontal cortex, AI agranular insular cortex, PrL prelimbic cortex, M2 secondary motor cortex, MeA medial amygdala, CeA central amygdala

Inhibition of OFC–amygdala projection neurons in WT mice recapitulates social deficits seen in BDNF^Met/Met mice

To investigate whether BLA-projecting mOFC neurons are causally involved in regulating social behaviors, we used Designer Receptors Exclusively Activated by Designer Drugs (DREADD)-based chemogenetic tools [38] (AAV-hM4D(Gi)) to selectively inhibit bilateral projections from mOFC to BLA in adult WT (BDNF^Val/Val) mice (Fig. 4a). Our functional and anatomical results suggested a strong association between mOFC-BLA projection and social novelty approach. However, since the BDNF Met allele can alter multiple neural circuits, the specificity of this circuit in relation to social novelty approach has not been established. If the altered social behavior was due to the reduced prefrontal regulation of limbic regions, then inhibition of BLA-projecting mOFC neurons in BDNF^Val/Val mice should induce similar behavioral changes. Similar to fiber photometry, due to the sparse nature of the CeA or MeA projecting mOFC neurons [35] (Fig. S6), the inhibitory effect was largely confined to mOFC-BLA circuit. Indeed, one-way ANOVA showed a significant change in social novelty approach based on DREADD expression and CNO treatment (F(2,20) = 4.550, p = 0.0235). In post hoc analyses, we found that inhibition of BLA-projecting mOFC neurons using Clozapine-N-oxide (CNO) in BDNF^Val/Val mice expressing DREADD induced significant reduction in social novelty approach (t(14) = 2.913, p = 0.011) when compared to corresponding saline controls. Similar to the behavioral phenotype observed in BDNF^Met/Met mice, CNO did not induce behavioral changes during novel object investigation in the DREADD-expressing WT mice (Fig. 4b). In addition, administration of CNO in a non-DREADD-expressing (GFP) group did not alter social and object interaction behaviors, indicating that the changes seen were not due to nonspecific effects of CNO (Fig. 4a, b).

Fig. 4.

Selective inhibition of mOFC-BLA projection can induce social deficit. a DREADD and GFP expression time course and representative injection site of DREADD vector in bilateral mOFC (left bar = 200 μm; right bar = 80 μm). b There is a significant difference in social novelty approach based on treatments (F(2,20) = 4.550, p = 0.0235). Specifically, CNO injection in DREADD-expressing mice led to a significant reduction in social novelty approach behaviors compared to the control saline group (t(14) = 2.913, p = 0.011; 7 Saline, 9 CNO). Object investigation behaviors were not changed by CNO injection and CNO injection alone also did not change social or objection investigations (7 GFP). *p < 0.05 and **p < 0.01. Error bar = 1 SEM. MO medial orbitofrontal cortex, VO ventral orbitofrontal cortex, LO lateral orbitofrontal cortex

BDNF bioavailability within a developmental window is necessary for proper development of social functions

Given that the BDNF Met allele reduces BDNF bioavailability by affecting regulated BDNF secretion [39–41], reduced BDNF bioavailability within the mOFC-BLA projections may play a prominent role in leading to the social deficits in BDNF^Met/Met mice. To test the possibility of enhancing BDNF levels to reverse the social deficits, we conducted region-specific BDNF overexpression experiments (Fig. 5a). An AAV2 viral vector expressing BDNF was injected into bilateral mOFC of adult mice (postnatal day (P) 50) to selectively elevate BDNF levels within the mOFC. When tested at P70, BDNF^Met/Met mice that received exogenous BDNF at P50 continued to exhibit reduced social novelty approach (t(13) = 2.42, p = 0.031) (Fig. 5c). To investigate whether BDNF is instead necessary within a developmental timeframe for optimal maturation of the mOFC neurons, BDNF was overexpressed in mOFC of P25 BDNF^Met/Met mice, and behaviors were measured at P70. In contrast to the adult group, BDNF overexpression in bilateral mOFC during adolescence led to a rescue of the social deficits in BDNF^Met/Met mice (Fig. 5d). To determine whether these behavioral changes are associated with these anatomical changes in mOFC-BLA projections, mOFC-BLA fiber densities of a separate task-naïve BDNF^Met/Met and BDNF^Val/Val cohort with P25 GFP or BDNF-GFP injections were quantified at P70, and showed a significant enhancement of mOFC-BLA projections in BDNF^Met/Met animals receiving mOFC BDNF (t(16) = 5.792, corrected p < 0.001) (Fig. S7). Furthermore, to test whether the lack of BDNF-driven effect in the P50 injection group is due to the reduced duration of expression (only 20 days compared to 45 days in the P25 injection group), a separate cohort of P50-injected BDNF^Val/Val and BDNF^Met/Met mice were tested at P95 after 45 days of BDNF expression, and no behavioral rescue was observed (t(13) = 2.239, p = 0.0433; VV = 8, MM = 7; Fig. S8). Similarly, on the circuit level, no rescue in fiber density from mOFC-BLA was observed at P95 in BDNF^Met/Met mice after 45 days of expression (F(1,16) = 3.186, p = 0.5803; Fig. S9). Together, these data indicate that the developmental timing of BDNF availability, rather than its duration, is critical for mOFC-BLA circuit formation and social behavioral establishment.

Fig. 5.

Additional BDNF during adolescence is important for restoration and maintenance of social functions. a Representative injection site for mBDNF vector in bilateral mOFC (bar = 40 μm). b Representative injection site for Cre vector in bilateral mOFC (bar = 40 μm). c BDNF overexpression at P50 was not sufficient to restore social functions in BDNF^Met/Met mice at P70 (t(13) = 2.42, p = 0.031; 7 VV, 8 MM). d When BDNF was overexpressed at P25, social function was restored in the BDNF^Met/Met mice at P70, with no difference in social interaction relative to BDNF^Val/Val mice (7 VV, 7 MM). e Elimination of mOFC BDNF post-P50 did not lead to changes in social interaction and objection investigation (7 GFP, 8 Cre). *p < 0.05 and **p < 0.01. Error bar = 1 SEM. MO medial orbitofrontal cortex

To test whether BDNF during this postnatal development timeframe is sufficient in restore social functions in the long-term, BDNF was selectively deleted from bilateral mOFC of adult (P50) conditional BDNF^flox/flox knockout mice using an AAV2 viral vector carrying Cre-recombinase (Fig. 5b). This allowed us to maintain a normal level of BDNF in the mOFC during peri-adolescence, but selectively deplete it in adulthood. When tested at P70, there were no behavioral differences between the BDNF knockout and control groups (Fig. 5e), indicating that mOFC BDNF levels in adults do not significantly impact baseline social functions and that peri-adolescent BDNF availability can play a determining role in future adult social behaviors. Interestingly, no genotypic differences were found between human BDNF Val and Met allele carriers as well as BDNF^Val/Val and BDNF^Met/Met mice when social behaviors and circuit anatomies were investigated during adolescence (Fig. S10, Table S4), suggesting that while BDNF bioavailability during adolescence is crucial for modulating long-term social behaviors, its behavioral impact does not emerge until the transition from adolescence to adulthood.

Discussion

Using human functional neuroimaging and circuit-based techniques in rodent models, our study has expanded on prior findings and shed light on mechanisms underlying the neural basis of social behaviors, focusing on the OFC and BLA. While the PFC has been a target of social behavioral investigations in recent years, studies have largely focused on the ventromedial PFC (vmPFC) [17–19, 42–44], and less on the OFC. Traditionally, the OFC has been implicated in mediating cue evaluation [45, 46] and decision-making [47–49] in rodents, non-human primates, and humans. Specifically, orbito-striatal circuitry has been strongly implicated in action evaluation and action selection [50–53]. While not extensively investigated, the valuation function of OFC may also extend into social domains, as it has frequently been observed that OFC lesions can lead to impaired emotion recognition and social judgments [54–57]. Furthermore, although the precise functional role of the OFC is not clear, OFC activity has been associated with self-evaluations, social feedback responses, and social rewards [58–60]. In these studies, however, the downstream targets of OFC neurons in the context of social functions have not been elucidated. The amygdala, on the other hand, plays an established role in emotional processing, as well as fear/anxiety [61–65], a role that has recently been shown to extend into social domains. Indeed, a recent study in rodents using optogenetic techniques identified BLA as a key region for social interactions [13]. Moreover, clinical neuroimaging studies have observed elevated amygdala reactivity towards aversive social cues in SAD patients [31–33]. Despite detailed investigations into OFC and amygdala as separate regions, studies of specific orbitofronto-amygdala projections have not been equally extensive, especially with regard to social behaviors. Several studies have suggested that the OFC and its connections to the limbic regions may be important in cue evaluation, especially social cues and faces [66, 67], and clinical imaging studies have demonstrated concerted functional changes in OFC and amygdala in SAD patients [68, 69]. However, up to this point, no evidence was available to show the direct involvement of OFC–amygdala projections in actual social behaviors. Our behavioral and functional findings from human BDNF Val/Val and Met carriers and knock-in mice provide the first evidence of mOFC-BLA involvement in response to novel or aversive social cues as well as social anxiety (Figs. 1 and 2).

While the mOFC-BLA projection is likely a part of a larger and complex social network composed of previously implicated projections such as vHPC-BLA [13], VTA-NAc [16], PL-NAc [17], and microcircuits of PFC [18, 19], the mOFC-BLA may play a unique role within the network by modulating a specific subset of social behaviors that is distinct from the other pathways. While rodent social interaction paradigm is a reliable measure of general social behavioral changes, characterization of the exact social changes has been difficult due to the unspecific nature of the rodent interaction behaviors. Many factors such as perception of social rewards, perception of social threats, and social aggression can all potentially present as social interaction or social approach deficits. To decipher the basis of the social behavioral change, our study constrained the behavioral interpretation of the BDNF^Met/Met mice using modified rodent paradigms and behavioral results from human subjects carrying the same polymorphism. Results from object investigation and littermate interaction trials showed that the deficit is social specific and limited to social novelty approach only (Fig. 1, S3), indicating that rather than a general reduction in sociability, the difference in social behavior is triggered by challenging social situations. This is consistent with behavioral findings in human BDNF Met carriers who reported higher level of anxiety towards social situations. Taken together, these findings indicate that the direct mOFC projections to BLA likely modulate susceptibility to social anxiety through its top-down regulation of this limbic region. Finally, the altered activities of the mOFC projection neurons and their downstream targets in BLA seen in BDNF Met carriers may help explain the functional changes seen in OFC and amygdala of the SAD patients [68, 69]. Conversely, the similarity in neural signatures between SAD patients and BDNF Met carriers provides further support for involvement of the direct mOFC-BLA pathway in social anxiety modulation.

Mechanistically, given OFC’s function in social valuation, it is plausible that the OFC–amygdala projection, particularly between mOFC and BLA, is important in evaluation of and action selection in response to social cues. Specifically, incoming social cues initially processed by other brain regions are likely integrated with additional information by the prefrontal regulatory regions such as OFC. Based on this integrated or contextualized value, the OFC can then feedback on subcortical regions such as amygdala to control behavioral responses. Therefore, altered OFC–amygdala connectivity seen in BDNF Met carriers, and possibly SAD patients, may lead to compromised integration and feedback in response to an aversive or novel social cue, precipitating a social approach deficit or susceptibility to social anxiety.

Our current study indicated that the BDNF Val66Met SNP can alter mOFC-BLA projections anatomically and functionally to induce behavioral and neurological changes resembling that of patients with SADs. While the altered OFC and amygdala activities seen in patients with SADs and human BDNF Met allele carriers are likely the result of multiple factors including but not limited to genetic backgrounds, experiences, and environmental stress, the deficits seen in BDNF^Met/Met mice can be entirely attributed to the molecular changes induced by the BDNF Val66Met SNP. It is well established that BDNF plays an important role in emotional learning and memory [70–72], and the single amino acid substitution of Val66Met SNP can result in reduced activity-dependent BDNF secretion [39–41] and corresponding reductions in neuronal plasticity [73, 74]. The reduced BDNF availability, therefore, likely plays a major role in the development and function of various neuronal projections. Indeed, overexpressing BDNF in the mOFC neurons of BDNF^Met/Met mice during peri-adolescence effectively normalized their mOFC-BLA projections and social novelty approach behaviors in adulthood, while overexpressing BDNF later in life did not reverse the behavioral differences (Fig. 5), indicating that there is a limited developmental window during adolescence in which BDNF could exert its effect to normalize social behaviors. Interestingly, social behaviors and circuit anatomies were not changed in adolescent BDNF Met carriers or BDNF^Met/Met animals (Fig. S10). These findings indicate that while the genotypic divergence of phenotypes does not emerge until later, the availability of BDNF during adolescence is essential for determining the level of social function in adulthood. Furthermore, region-specific removal of mOFC BDNF in adulthood in conditional BDNF KO mice did not result in significant social behavioral change, indicating that having appropriate levels of BDNF during this peri-adolescent timeframe (approximately P25 to P50) is crucial to the development and long-term maintenance of proper OFC–amygdala projections for normal baseline social functions (Fig. 5). These results are in agreement with prior studies indicating that the precise temporal-spatial control of neuronal survival and differentiation by growth factors such as BDNF is critical for the proper formation of functional projections and synapses during development [75–77]. Similarly, peri-adolescent stress in rats has been shown to cause hypoactive mOFC and altered amygdala connectivity, resulting in changes in social behaviors in adulthood [78, 79]. While the stress-induced peri-adolescent changes in rats are largely driven by alteration monoamine system [80], change in BDNF availability can certainly exerts similar effects during this sensitive period. Our findings are also consistent with the imbalance model of emotional regulation in humans, in which there is an imbalance between the regulatory regions and emotional processing regions during adolescence [81]. As a result of this imbalance, the protracted regulatory regions such as PFC as well as their connections to subcortical regions must undergo significant maturation between adolescence and adulthood in order to properly assume regulatory functions in adulthood. Therefore, the adolescent availability of BDNF in prefrontal regions such as OFC, or lack thereof, could have significant influence on this maturation process and the eventual frontolimbic functions. In addition, it has been recently shown that global increase in BDNF expression during adolescence via fluoxetine administration was effective in rescuing anxiety-like phenotypes in adult BDNF^Met/Met mice [37], again highlighting the importance of BDNF during peri-adolescent development. From an interventional perspective, while BDNF-based treatment during adolescence may provide long-term benefit to BDNF Met allele carriers with regard to social behaviors, the decision to intervene may not be straight forward as neither circuit-level nor behavior-level deficit are present during adolescence. Therefore, BDNF Val66Met SNP may serve as a key screening tool for susceptibility detection in adolescents with other risk factors for SAD.

Together, our findings indicate that BDNF bioavailability during peri-adolescence could impact social anxiety-like phenotypes through its effects on the development of the mOFC-BLA pathway. By using a parallel human–mouse system linked by a common loss of function BDNF SNP, we were able to rely on human behaviors with a translational value to constrain our behavioral interpretations and rely on a mouse model system to demonstrate specificity of the genetic association. Moreover, the rodent studies involving fiber photometry, chemogenetic tools, and viral overexpression and knockout of BDNF allowed us to map relevant circuits with improved spatial resolution and establish a necessary role for BDNF in amygdala-projecting OFC neurons in supporting social approach behavior. Finally, the converging findings from both human and mouse model can improve the confidence of observed genotype–phenotype associations, as nonreplicable associations have been a major problem in the field. Developmentally, we showed that there is a timeframe within peri-adolescence in which the OFC–amygdala-driven social deficits can be permanently reversed in the BDNF^Met/Met mice. These findings suggests the intriguing possibility that non-invasive therapies and pharmacological agents shown to elevate BDNF, such as exercise [82–84], environmental enrichment [85], and antidepressants [86–89] may become invaluable tools in correcting BDNF Val66Met SNP-related behavioral alterations through timed screening and interventions during adolescence.

Methods

Animals

Male BDNF Val66Met mice (BDNF^Met/Met and BDNF^Val/Val) backcrossed to C57BL/6N background were used for most experiments, with each experiment conducted with an independent cohort. Male BDNF^flox/flox mice purchased from The Jackson Laboratory were used for the P50 conditional knockout experiment. All animals were group-housed (up to five animals per cage) in a vivarium with a regular 12-h light/dark cycle. For BDNF Val66Met colony, all cages used had at least one MM and two VV mice (the extra VV serves as littermate partner). All behavioral experiments were conducted during the light phase of the cycle at P95, P70, or P35. No female mice were used. Animal care was in accordance with Weill Cornell Medicine, Institutional Animal Care and Use Committee, National Institutes of Health Care and Use of Laboratory Animals.

Rodent three-chamber behavioral paradigm

The standard rodent three-chamber test was conducted as described previously [24]. Briefly, after two sessions of 5 min habituation, experimental animals were allowed to choose between a novel social partner (social cue; non-littermate WT) and a novel object (object cue) in the sociability trial, and choose between a familiar social partner (social cue from the previous trial) and a novel social partner (a new non-littermate WT) in the social preference trial. All trials were 5 min and occurred consecutively, with only minimal delays for cue manipulation. Between trials, experimental animals were confined in the center chamber while cue exchange occurred in the side chambers. Experimental animals were exposed to both cues simultaneously at the beginning of each trial. Circular zones surrounding the cups were designated as interaction zones, and zone entries were live tracked using Ethovision XT 11.5 (Noldus) with an infrared sensitive digital camera (Polestar II, Everfocus Electronics). To test for social behaviors independent of social memory, in a separate cohort of animals, a littermate WT was used as the social cue during the sociability phase and as the familiar interaction partner during the social preference phase. Analysis was done using R3.2.1 ANOVA package to detect gene-by-cue interaction on interaction times.

Rodent free social interaction behavioral paradigm

The rodent free social interaction task was conducted in a housing cage with bedding under infrared light. All animals were allowed to habituate to the testing room for at least 30 min prior to the test. At the beginning of a trial, the experimental mouse was allowed to habituate to a new housing cage for 15 min before a novel object, a familiar interaction partner (WT littermate), or a novel interaction partner (WT non-littermate) were added to the cage for 5 min each, consecutively without delays in between. The order of cue exposure was designed to minimize stress and anxiogenic factors during the test. A new housing cage (L 30 cm ×W 18 cm × H 12.5 cm) with fresh bedding was used for each experimental animal, but the cage is not changed or cleaned between trials to reduce disturbance to the experimental animal. The amount of time the experimental animal’s nose spent in a zone surrounding the object (~1 cm wide) as well as object itself (~0.5 cm wide edge) was used as a measure of object investigation, and the amount of time an experimental animal’s nose spent in proximity (<1.5 cm) to a partner’s tail-base was used as a measure of active social approach. The areas of detection for both object and social trials are roughly 2.25π cm [2]. The animals were simultaneously live tracked using Ethovision XT 11.5 (Noldus) with an infrared sensitive digital camera (Polestar II, Everfocus Electronics). Two-animal tracking module was used, and experimental animal was differentiated from interaction partner through tail markings. Analysis was done using R3.2.1 t-Test package to detect the effect of genotype on interaction times.

Fiber photometry of modified social interaction behavioral paradigm

At P50, 200 nl of Cre-dependent calcium indicator vector (AAV1-Syn-Flex-GCaMP6s-WPRE-SV40; Penn Vector Core) was injected into left mOFC (AP +2.4, ML −0.5, DV −2.3) of BDNF^Val/Val mice and BDNF^MetMet mice over 4 min. Two hundred nanoliters of retrograde-Cre vector (rAAV2-Retro-CAG-Cre; UNC GTC Vector Core) was injected into ipsilateral BLA (AP −1.6, ML −2.75, DV −3.5) over 4 min. The needle was kept at the injection site for an additional 8 min before being withdrawn to minimize leakage. A 400-μm diameter 0.48 numerical aperture (NA) mono fiber-optic cannula (Doric Lenses) was implanted 200 μm dorsal to the calcium indicator injection site in mOFC (AP+2.4, ML −0.5, DV −2.1). The cannula was fixed to the skull using two layers of adhesive dental cement (C&B Metabond, Parkell). Animals were tested 20 days post-surgery at P70, and animals with no GCaMP expression in the mOFC region were removed from analysis.

The social interaction test with fiber photometry recording was conducted under infrared light in an open field arena (64 cm × 36 cm). All animals were habituated to the patch cable attachment (Doric) for 30 min one day prior to the test and were allowed to habituate to the experimental room for at least 45 min on the test day. At the beginning of a trial, the patch cable was attached to the chronically implanted optical fiber and the experimental animal was allowed to explore the arena for 10 min, followed by an object interaction phase and a novel social interaction phase of 5 min each. Both the object and the interaction partner were confined under a cup to prevent interference with the cable. A zone surrounding the cup (~1 cm wide on all sides) was designated as the interaction zone consistent with previously published reports [90], and zone entries were tracked using Ethovision XT 11.5 (Noldus). Zone entries of less than 1 s in duration (in which the test mouse briefly entered the interaction zone and immediately retreated) were excluded from the analyses reported in Fig. 2. During interaction trials, a 470 nm LED excitation light (Thorlabs) was delivered at 521 Hz, and emitted fluorescence signals were isolated using a dichroic beam splitter (Semrock, FF495-Di03) and a bandpass filter (Semrock, FF01–535/50), captured by a photodetector (Newport, Model2151), and recorded by a real-time processor (Tucker Davis Technology RX8). A TTL pulse was sent from Ethovision to the fiber photometry rig at the beginning of each trial to time-lock behaviors and calcium signals. The detailed fiber photometry rigs design has been described elsewhere [16, 34]. The arena, cup, and fiber were cleaned thoroughly with ethanol between each animal.

Raw fluorescent signals were imported and analyzed using MATLAB R2016a (MathWorks). A dynamic baseline fluorescent value was obtained by identifying the median value a 30 s rolling window to remove trends overtime. The normalized signals (dF/F) were computed by subtracting the baseline value from each data point then dividing by the baseline value. The normalized signals from 2 s before and 3 s after the zone entry were averaged across all interactions of all animals to plot activity changes time locked to specific behaviors by genotype (Fig. 2e). Quantitatively, average dF/F of 1 s after and 0.5 s before the zone entry were calculated for each individual zone entry to investigate genotypic difference and gene-by-target interaction using a mixed-effects model controlling for within-subject errors using lme function in R. The 1-s window post zone entry was used for quantitative analysis because it represents a window in which the animal is guaranteed to be present in the interaction zone. Since zone entries less than 1 s in duration were removed from the analysis, neuronal signals from the first second of the remaining zone entries were not contaminated by other activities.

Stereotaxic virus injection

All experimental and surgical procedures were approved by the Weill Cornell Medicine’s IACUC. Animals received ketamine/xylazine cocktail i.p. (85 mg/kg ketamine and 12 mg/kg xylazine) prior to the surgery. Animals were leveled, and coordinates were identified based on bregma and lambda skull landmarks in a stereotaxic apparatus (Kopf Instrument). All injections were done using 10 μm NanoFil Hamilton syringe and 33-g blunt needle (WPI). Analgesics were administered for 3 days post-surgery.

Anterograde tract-tracing

At P70 or P35, 50 nl of PHA-L (2.5% solution in 0.05 M pH 7.4 TBS; Vector Laboratories) was injected into left mOFC (AP +2.4, ML −0.5, DV −2.3) of BDNF^Val/Val and BDNF^Met/Met mice over 4min (Fig. 3d). Ten days post-surgery, animals were anesthetized with Euthasol (0.1 ml/10 g body weight) and perfused through the heart with 30 ml 0.9% saline followed by 120 ml 4% paraformaldehyde in a 0.1 M phosphate buffer. Brains were removed and post-fixed in 4% paraformaldehyde in a 0.1 M phosphate buffer at 4°C overnight and transferred to a sucrose solution (30% sucrose in 0.1 M phosphate buffer at 4°C for 48 h). Forty micrometer coronal sections were cut using a freezing microtome. Free-floating serial sections (every three sections) were washed (3 × 10 min) in TBS, followed by a 30-min blocking in solution containing 4% normal donkey serum and 1% BSA in TBS with 0.2% Triton X-100. After blocking, rabbit anti–PHA-L primary antibody (1:1,000; Vector Laboratories) diluted in the blocking solution was applied to the sections for a 24 h incubation at 4°C. Alexa Fluor-labeled donkey anti-rabbit 555 (1:500; Invitrogen) was then applied to the sections for a 2h incubation. Finally, the sections were mounted, cover-slipped, and sealed with nail polish. Injection sites were confirmed using Allen Mouse Brain Map (Allen Co.) under a Nikon 80i Fluorescent Microscope. MicroFire 11 and Fire-Frame software (Optronics) were used for digital photography, and StereoInvestigator software (MicroBrightfield) was used for stereological cell and fiber density estimation. Briefly, contours of sub-nuclei MeA, CeA, and BLA were drawn (with its total volume estimated using the Cavalieri method), and random sampling was applied to the contours. Section tracing was done under a ×4 lens, and perimetrics probe analysis was performed under a ×40 lens. Counting frame was set to 25 × 25 μm, and the radius of the Merz coherent test system was set to 5 μm. Total length of all sampling sites was calculated automatically, and fiber density was obtained as the quotient of the length-summation and the area-summation of all sections. No differences in injections were observed across genotypes. Analysis was done using R3.2.1 t-Test package to detect the effect of genotype on mOFC-BLA fiber densities.

Designer Receptor Exclusively Activated by Designer Drug (DREADD)

At P50, 200 nl of Cre-dependent DREADD vector (rAAV8-hsyn-DIO-hM4D(Gi)-mCherry; UNC GTC Vector Core) or a control GFP vector was injected into bilateral mOFC (AP +2.4, ML ±0.5, DV −2.3) of BDNF^Val/Val mice over 4 min. 200 nl of retrograde-Cre vector (rAAV2-Retro-CAG-Cre; UNC GTC Vector Core) was injected into bilateral BLA (AP −1.6, ML ± 2.75, DV −3.5) over 4 min. The needle was kept at the injection site for an additional 8 min before being withdrawn. Animals were tested 20 days post-surgery at P70, with the DREADD group receiving either CNO (2 mg/kg) or saline i.p. and the GFP group receiving CNO (2 mg/kg) i.p. 45 min before the test. Animals were habituated to i.p. injection by daily saline i.p. for 3 days prior to the test day. Animals showing no bilateral DREADD expressions in the mOFC region were removed from analysis. Analysis was done using R3.2.1 t-Test package to detect the effect of CNO treatment on interaction times.

BDNF overexpression and deletion

At P25 or P50, 200 nl of BDNF vector (rAAV2-CMV-GFP-2A-mBDNF; Vector Biolabs) was injected into bilateral mOFC (AP +2.4, ML ±0.5, DV −2.3) of BDNF^Val/Val mice and BDNF^Met/Met mice over 4min. Needle was kept at the injection site for an additional 8 min before withdrawn. Animals were tested 45 or 20 days post-surgery at P70. At P50, 200 nl of Cre-recombinase vector (rAAV2-hsyn-mCherry-Cre; UNC GTC Vector Core) was injected into bilateral mOFC (AP +2.4, ML ±0.5, DV −2.3) of BDNF^flox/flox mice over 4 min. Needle was kept at the injection site for an additional 8 min before being withdrawn. Animals were tested 20 days post-surgery at P70. Animals with no bilateral expressions (GFP-BDNF or mCherry-Cre) in the mOFC region were removed from analysis. Analysis was done using R3.2.1 t-Test package to detect the effect of overexpression of knockout treatment on interaction times.

Emo-go-nogo and functional magnetic resonance imaging participants

Sixty-six healthy adult and 35 healthy adolescent consented participants were scanned using functional magnetic resonance imaging (fMRI) while performing Emo-go-nogo task. Study protocol was approved by IRB of Weill Cornell Medicine. Seven of the 66 adult participants were excluded due to poor task performance (Go accuracy <0.7), and 4 of the 66 adult participants were excluded due to excessive motion in the scanner (defined as >10% of time points exceeding >1.56 mm translational motion, half a voxel, or >1 degree rotational motion overall), resulting in 55 usable subjects (mean age = 23.27; range = [18–32]). In the 55 usable subjects, there were a total of 21 Met (5 MM, 16 MV; 12 females) and 34 Val (21 females). Three of the 35 adolescent participants were excluded due to poor task performance, and 4 of the 35 adolescent participants were excluded due to motion in the scanner, resulting in 28 usable subjects (mean age = 15.45; range = [12–18]). In the 28 usable subjects, there were a total of 8 Met (1 MM, 7 MV; 6 females) and 20 Val (20 females). Data from this sample have been published previously on different subsets of the data [28–30]. A Taqman 5′ exonuclease assays (ABI) and a 7900HT apparatus was used to genotype saliva-derived DNA at the BDNF Val66Met (rs6265) SNP. Adults and parents provided informed written consent and minors provided assent. The study was approved by the institutional review board of Weill Cornell Medicine.

Anxiety questionnaire and diffusion MRI participants

Data used were obtained from the Pediatric Imaging, Neurocogniton, and Genetics (PING) study database (pingstudy.ucsd.edu), with an initial sample size of 205 healthy adult participants (18–21 years) and 413 healthy adolescent participants (12–17 years). One of the 205 adult participants was excluded due to excessive motion in the scanner (defined as more than one-half of the gradient slices exceeding motion threshold >1.25 mm), and 41 participants were excluded due to insufficient information of sex, ancestry, and/or genotype, resulting in 163 usable subjects (mean age = 19.52; range = [18–22]). In the 163 usable subjects, there were a total of 72 Met (12 MM, 60 MV; 48 females) and 91 Val (44 females). Sixty-three of the 163 subjects completed anxiety questionnaires. Subject with social anxiety score beyond 95% CI were removed from analysis, resulting in a usable sample size of 62, consisting of 28 Met (6 MM, 22 MV; 15 females) and 34 Val (17 females). Three of the 413 adolescent participants were excluded due to excessive motion in the scanner, and 71 participants were excluded due to insufficient information of sex, ancestry, and/or genotype, resulting in 339 usable subjects (mean age = 14.82; range = [12–18]). In the 339 usable subjects, there were a total of 114 Met (11 MM, 103 MV; 53 females) and 225 Val (104 females). Of the 339 usable subjects, 141 completed anxiety questionnaires. Subject with score beyond 95% CI were removed from analysis, resulting in a usable sample size of 139, consisting of 52 Met (8 MM, 44 MV; 22 females) and 87 Val (37 females).

BDNF Val66Met genotype was acquired by the PING Genomic Core by genome-wide genotyping on saliva-derived DNA using Illumina Human 660-Quad Beadchip. Supervised clustering approach of ADMIXTURE software was used to determine ancestry and admixture proportion of each participant corresponding six continental populations: African, Central Asian, East Asian, European, Native American, and Oceanic.

Behavioral anxiety measures

Data used were obtained from the PING study database (pingstudy.ucsd.edu), with 63 subjects (see above) completing the self-report for age-adjusted Screen for Child Anxiety-Related Emotional Disorders (SCARED). Social anxiety self-report score was used for analysis and composite score was used as a covariate. Social anxiety questions include prompts such as “I don’t like to be with people I don’t know well” or “I feel nervous with people I don’t know well.” The detailed data collection and adjustment methods can be found on PING data resource website (http://pingstudy.ucsd.edu/) and previous publication [91]. Analysis was done using R3.2.1 ANOVA package to detect main effect of genotype (VV vs. MM/VM) on social anxiety while controlling for gender, site, general anxiety score, and ancestry.

Behavioral paradigm for functional fMRI (Emo-go-nogo)

Functional MRI participants completed six runs of a go-nogo task in scanner using happy, fear, and calm facial expressions as target (go) and non-target (nogo) social cues [92]. In each run, two social cues of different expressions were presented as either target cues, to which they were instructed to press a button, or non-target cues, to which they were instructed to withhold button press. Instructions were given prior to each run. Expressions were pseudorandomized across the run to control for presentation order. All combinations of expressions were used as target–non-target pairs, resulting in a 3 (expression: happy, fear, calm) × 2 (response: go, nogo) factorial design with a unique target–non-target pair for each of the six runs. The detailed task design was previously described [28]. Accuracy on the task was determined using the sensitivity index d′, which accounts for both hits and false alarms [93]. Overall d′ for each participant was calculated by subtracting normalized false alarm rates on nogo trails from normalized accuracy on go trials. Similarly, d′ for each emotion was calculated by subtracting normalized false alarm rate from normalized accuracy for each emotion. Analysis was done using R3.2.1 ANOVA package to detect gene-by-cue interaction on d′ while controlling for gender.

Functional MRI

Two different scanners at Weill Cornell Medical College Citigroup Biomedical Imaging Center were used to acquire anatomical and functional images. A scanner was included as a factor of two levels to control for scanner effect and only overlapping brain regions were used in the analysis. General Electric Signa 3.0T fMRI Scanner: High-resolution T1-weighted scan was acquired in 124 1.5 mm sagittal slices (256 × 256 mm in-plane resolution; 240 mm field of view). Functional images were acquired as 34 4.0 mm coronal slices (TR = 2500 ms, echo time = 30 ms, field of view = 200 mm, flip angle = 90, 3.125 × 3.125 mm resolution) covering the whole brain except for the posterior portion of the occipital lobe. Siemens Magneto Trio 3.0T fMRI Scanner: High-resolution T1-weighted scan was acquired in 160 1.2 mm sagittal slices (256 × 256 mm in-plane resolution; 256 mm field of view). Functional images were acquired as 38 4.0 mm axial slices (TR = 2500ms, echo time = 30 ms, field of view = 200 mm, flip angle = 90, 3.1 × 3.1 mm resolution) covering the whole brain.

Analysis of Functional NeuroImages (AFNI) software version 16.0.00 was used to process functional imaging data [94]. Preprocessing steps included slice-time correction using sinc interpolation, volume registration with six6-parameter rigid-body transformation accounting for head motions, and normalization to the Montreal Neurological Institute (MNI) 152 1 mm T1 template using 12-parameter affine transformation and nonlinear transformation (AFNI 3dQWarp function). Transformed images were then resampled to 3 mm voxels and smoothed using a 6 mm full-width/half-maximum Gaussian kernel. Signal intensity of each voxel time series was normalized to percent signal change. A general linear model (GLM) was used to estimate the voxel-wise activation to different emotional cues for each individual. To examine participants’ reaction to each emotional social cue, we modeled each emotional cue controlling for both response types and performance, with 6 emotional combinations (e.g. fear-go happy-nogo, happy-go calm nogo), 2 response types (go and nogo), and 2 performance outcomes (correct or incorrect), for a total of 24 regressors (e.g. fearful-go-correct, happy-nogo-incorrect). In addition, six motion estimation parameters were included. Baseline trends were estimated to capture shifts in signal change. Responsive activations were modeled with a three-parameter gamma hemodynamic-response function (HRF). Time points with motion greater than half a voxel (1.56 mm), as well as the preceding and following time points, were censored.

Regression coefficients from individual GLMs were submitted to a whole brain group-level linear mixed-effect model analysis using AFNI 3dLME function with type III sums of squares. In the general model, random deviations from the group mean were used for participant intercepts, with BDNF genotypes (VV vs. MM/VM), gender, and scanner as between-subject variables; and cue types (fear or calm), response types (go or no-go), and performance (correct or incorrect) as within-subject variables. In addition, the data were fitted to a parallel model using happy and calm as cue-types to distinguish the effect of fear cue from the effect of general arousal.

For investigation of gene-by-cue interaction, a voxel-wise p-value of 0.005 was used. Cluster-size was thresholded at an α of 0.05 after correction for multiple comparisons. Multiple comparison correction was performed using the 3dClustSim program in 2016 AFNI version 16.0.00 in which the false-positive rate was adjusted [95]. In’ the 3dClustsim program, the new mixed model autocorrelation function option (-acf option) was used to further reduce false-positive rate based on recent recommendations [96]. Post hoc analyses on extracted beta weights were done using R3.1.2 ANOVA package to detect gene-by-cue interaction on beta weights (of amygdala and OFC) while controlling for gender and scanner.

Diffusion MRI

Diffusion MRI data were obtained from the PING study database. High-resolution T1-weighted scan and diffusion-weighted scans were acquired with standardized protocols across sites with most parameters constant across all locations. Site was included as a factor in analysis models. Siemens Magneto Trio 3.0T fMRI Scanner: High-resolution T1-weighted scan was acquired in 160 1.2 mm sagittal slices (256 × 192 mm in-plane resolution; 256 mm field of view). Two series of 30 directional diffusion-weighted images were acquired in 68 slices [TR = 19 s; echo time = 91 ms; anterior to posterior phase encoding]. General Electric Discovery 750 3.0 T fMRI Scanner: High-resolution T1-weighted scan was acquired in 166 1.2 mm sagittal slices (256 × 192 mm in-plane resolution; 240 mm field of view). Two series of 30 directional diffusion-weighted images were collected in 51 slices (TR = 8 s; echo time = 80.7 ms; right to left phase encoding). General Electric Sigma HDx 3.0T fMRI Scanner: High-resolution T1-weighted scan was acquired in 166 1.2 mm sagittal slices (256 × 192 mm in-plane resolution; 240 mm field of view). Two series of 30 directional diffusion-weighted images were collected in 51 slices (TR = 13.6 s; echo time = 83 ms; right to left phase encoding). Phillips Achieva 3.0T fMRI Scanner: High-resolution T1-weighted scan was acquired in 170 1.2 mm sagittal slices (256 × 256 mm in-plane resolution; 240 mm field of view). Two series of 32 directional diffusion-weighted images were collected in 60 slices (TR = 9 s; echo time = 91.14 ms; anterior to posterior phase encoding).

Imaging files were processed using customized pipeline using both AFNI [94] and FMRIB Software Library software packages [97]. Both Eddy current and susceptibility-induced distortions were corrected. Head movements during the scan were calculated and realigned across gradient direction slices. Gradient vectors directions were then rotated based on motion correction. Any gradient slices with head displacement greater than half voxel (1.25 mm) were excluded from analyses. White matter structural connectivity was investigated based on functional seeds obtained from functional MRI scans. Probabilistic tractography was used to obtain tract masks on a participant-by-participant basis. Diffusion data were modeled using a crossing fiber method described previously [98]. A two-tensor model was used to perform probabilistic tractography and FA analysis. FA maps were transformed into MNI space using standard FA map, and probabilistic tractography was performed using orbitofrontal and amygdala clusters obtained from the functional MRI scan as seed and waypoint masks. Individual probabilistic maps are then filtered at a threshold (5000) to isolate clean white matter tract masks. Resulting masks are individually inspected to ensure its conformity to known white matter tracts [99]. FA values are extracted from FA map using the tract masks and average across all voxels to obtain a single FA value for each participant. Post hoc analyses on extracted FA were done using R3.2.1 ANOVA package to detect main effect of genotype (VV vs. MM/VM) on FA while controlling for gender, site, and ancestries.

Statistics

All statistical analyses were conducted using R3.2.1 Student’s T-test package using two-tailed tests, R3.2.1 ANOVA package (type III sum of square) using two-tailed tests, and R3.2.1 nlme (linear mixed-effect model) package. Detailed statistical analyses and modeling for human imaging studies are outlined in each respective section. F, t, p values, and n are given in results and figure legends when applicable. GraphPad Prism 6.0 was used to produce graphs for visual purposes only. No blinding process was used during behavioral experiments as mice were ran in staggered order based on genotype or treatment, and behavioral scoring was automated to eliminate human bias. Experimenter was blinded for immunohistochemistry experiments to prevent human bias. Rodent sample size was determined by power analysis (power = 0.8) based on previously published literature [100].

Resources availability

The following antibodies are used for staining: (1) Anti-Phaseolus Vulgaris Agglutinin (E+L), Unconjugated, Made in Rabbit (Vector Laboratories); Catalog Number AS-2300; (2) Donkey anti-Rabbit IgG (H+L) Highly Cross-Adsorbed Secondary Antibody, Alexa Fluor 555 (Invitrogen); Catalog Number A-31572;

Data availability

All the relevant data that support the findings of this study are available from the corresponding author upon reasonable request. Data from PING data set is available publicly online (pingstudy.ucsd.edu).

Code availability

The custom written code files for fiber photometry data processing and human imaging analyses are available upon request.