Altered social behavior in mice carrying a cortical Foxp2 deletion

Abstract

Genetic disruptions of the forkhead box transcription factor FOXP2 in humans cause an autosomal-dominant speech and language disorder. While FOXP2 expression pattern are highly conserved, its role in specific brain areas for mammalian social behaviors remains largely unknown. Here we studied mice carrying a homozygous cortical Foxp2 deletion. The postnatal development and gross morphological architecture of mutant mice was indistinguishable from wildtype (WT) littermates. Unbiased behavioral profiling of adult mice revealed abnormalities in approach behavior towards conspecifics as well as in the reciprocal responses of WT interaction partners. Furthermore mutant mice showed alterations in acoustical parameters of ultrasonic vocalizations, which also differed in function of the social context. Cell type-specific gene expression profiling of cortical pyramidal neurons revealed aberrant regulation of genes involved in social behavior. In particular Foxp2 mutants showed the downregulation of Mint2 (Apba2), a gene involved in approach behavior in mice and autism spectrum disorder in humans. Taken together these data demonstrate that cortical Foxp2 is required for normal social behaviors in mice.

Introduction

Genetic disruptions of the forkhead box transcription factor FOXP2 in humans cause developmental verbal dyspraxia (OMIM#602081), which is characterized by impaired precision and consistency of movements underlying speech (1). In addition, individuals carrying FOXP2 mutations also show receptive and expressive language deficits (2–5).

Functional genomics studies (6–8) suggest that FOXP2 regulates a variety of physiological processes such as neurite outgrowth and synaptic plasticity. Specific transcriptional targets have linked FOXP2 to genes involved in other forms of speech and language impairment (9), such as the sushi repeat-containing protein X-linked 2 (SRPX2), involved in epilepsies originating in cortical speech areas (9). Furthermore the FOXP2 target contactin associated protein-like 2 (CNTNAP2), a neurexin-family member, is associated with typical specific language impairment (OMIM#612514) and has also been implicated in autism spectrum disorder (ASD) OMIM#209850 (7,10–12,70). Apart from other transcriptional targets such as MEF2C (13), FOXP2 also directly interacts with transcription factors linked to ASD such as TBR1 (14–16) and FOXP1 (17). Although FOXP2 has not been directly associated with ASD, these findings suggest that FOXP2-dependent molecular networks are involved in social cognition and behavior.

The FOXP2 sequence is highly conserved and the protein is expressed in homologous brain regions in cortex, striatum, thalamus and cerebellum (18–27). Expression starts during embryogenesis and is maintained in the adult brain, suggesting roles both during neurodevelopment and in adult neuronal function. The contribution of Foxp2 in different brain areas to social cognition and behavior however remains largely unknown.

In the developing cortex, Foxp2 has been implicated in the transition of radial glial cells to neurogenic intermediate progenitor cells, radial migration and neurite maturation (28–30). Foxp2 overexpression in early cortical neurons has been associated with decreased formation of excitatory synapses, potentially via direct repression of Srpx2 (31).

In the mature cortex Foxp2 expression is found prominently in layer VI and few layer V excitatory projection neurons. Foxp2⁺ neurons in layer VI colocalize with Tbr1, a transcription factor which is crucial for layer VI development, and Tbr1 haploinsufficiency has been shown to impair social interaction and USV (32,18,27,33). De novo TBR1 mutations associated with sporadic autism were shown to disrupt protein functions and interaction with FOXP2 (14,16) which in turn may impair the coordinated regulation of common downstream targets, possibly involving ASD candidate genes (14). As layer VI contains a morphologically heterogeneous and functionally less well characterized population of cortico-cortical and cortico-thalamic projection neurons (34–37), the identities of Foxp2⁺ layer VI neurons and the extend of overlapping expression with Tbr1 remains to be determined.

Throughout postnatal development and into adulthood a few layer V subcortically projecting neurons from specific cortical areas show co-expression of both Foxp2 and the transcription factor Ctip2 (24,27). In particular a sparse population of neurons in layer V of the primary motor cortex (M1) in mice were found to project directly to the nucleus ambiguous which is important in motor control of the larynx (laryngeal motor cortex—LMC). Such a connection previously had been thought to be present only in vocal learning species, and has been implicated in frequency modulation of mouse vocalizations. Retrograde trans-synaptic tracing of laryngeal muscles with pseudorabies virus showed that some of these cells in M1 express Foxp2, although at a lower level, which may either be specific to these cells or can also be attributed to viral toxicity (38).

Several studies of USV in adult Foxp2 mouse mutants employing different experimental set ups and call analysis paradigms suggest specific changes ranging from an absence of gross abnormalities in song repertoire and only refined alterations in syllable acoustic structures to a major reduction of syllable production, disappearance of long syllables and abnormal rhythmic structure of a courtship songs (38–41). Adult mice carrying a humanized Foxp2 knock-in allele do not show differences in spectral call structure with the exception of calls that differ in frequency jumps (42).

While these studies made clear that Foxp2 affects a variety of neural processes during development and adulthood, the contribution of Foxp2 in different brain areas to social cognition and behavior remains largely unknown. Here, we explored a potential contribution of Foxp2 in postmitotic excitatory neurons to cortical development and social behavior in mice. We found that homozygous Foxp2 deletion does not impair the gross morphology of the cortex. In behavioral studies, we find that these mutant mice show subtle changes in USV and social interactions accompanied by corresponding changes in ribosome-bound transcript profiles.

Results

Conditional homozygous Foxp2 deletion in neurons does not affect gross cortical morphology

In order to generate mice carrying a homozygous deletion in the cortex (subsequently named cKOs), we intercrossed Foxp2^lox/lox and Nex-Cre^+/− animals. Nex-Cre induces recombination in postmitotic newborn glutamatergic projection neurons generated from dorsal telencephanlic (Cre negative) progenitors starting from around embryonic day 11 (E11) (43,44). Cre expression leads to excision of Foxp2 exons 12–14, encoding the DNA-binding domain (45). Immunostainings (Fig. 1A), RT-PCR (exon-12-13) (Fig. 1B) and western blot (Fig. 1C) on adult cKOs showed efficient deletion in the cortex while Foxp2 expression in striatum, thalamus and cerebellum was maintained. Analyses of Nissl-stained brain sections did not reveal any gross histological abnormalities, characteristic cortical layers could be identified (Fig. 1D). Measurements of cortical thickness across different areas showed that cortical size was normal in cKOs (Fig. 1E), suggesting that progenitor fate transition into neurons was unaffected which is consistent with the restriction of Cre⁺ expression to postmitotic newborn neurons during development.

Figure 1

Foxp2 deletion in the cortex in Nex-Cre; Foxp2^lox/lox mouse (cKO) line. (A) Representative immunohistochemical images, scale bars 2 mm. (B) Reverse transcription PCR and (C) Western blot on tissue punches from striatum and cortex. (D) Somatosensory cortex Nissl staining. (E) Cortical thickness measurements on Nissl stained sections (genotype F (1,80)=0.42, P > 0.05, n = 3–4). Mean + SEM.

We then investigated a potential impact of homozygous cortical Foxp2 deletion on the number of layer VI Tbr1⁺ neurons. An unbiased automatic estimation of the overlap between Tbr1⁺ and Foxp2⁺ expression in WT animal showed that 69.8 ± 3.2% of Foxp2⁺ neurons in layer VI co-express Tbr1 and 55.0 ± 2.2% of Tbr1⁺ cells co-express Foxp2⁺ (Fig. 2A). There was no cortical area examined that showed 100% co-expression, in contrast to what has been reported previously in the early postnatal somatosensory cortex (27).

Figure 2

Tbr1 expression and localization analysis in the cortex of WT and Foxp2 cKO mice. (A) Tbr1 and Foxp2 colocalization in cortical layer VI of WT animals. (B) Tbr1 immunostaining (picture) and quantification (bar graph), scale bars 2 mm (genotype F (1,50)=1.63, P > 0.05, n = 3–4). Data presented as mean + SEM. (C) Representative western blot of lower cortical layer punches.

We then assessed the number of Tbr1⁺ layer VI cells and found no difference between cKO and WT littermates (Fig. 2B and D). Likewise western blot analysis of lower cortical layer tissue punches did not show significant differences between genotypes (n = 7, t (12) = 0.2594, P = 0.7997) (Fig. 2C). These data indicate that Foxp2 may not be required to regulate the expression of Tbr1 or the number of Tbr1⁺ neurons.

We next explored the impact of homozygous cortical Foxp2 deletion on layer V subcerebral projection neurons in the motor cortex including the LMC. Layer V is characterized by the specifically high expression of the transcription factor Ctip2 (46,47). Therefore, we first counted Ctip2⁺ layer V cells in coronal section levels corresponding to the LMC as well as in sagittal sections. We did not detect differences in the number of Ctip2⁺ layer V neurons (Fig. 3). Moreover, there was also no difference in the total cell number or cell density as counted using pan-nuclear Hoechst staining (not shown). Taken together our morphological studies did not detect gross alterations of the cortical histoarchitecture following homozygous Foxp2 deletion.

Figure 3

Ctip2 immunostaining and quantification in the cortex of WT and Foxp2 cKO, scale bars 1 mm (coronal: genotype F (1,16) = 1.20, P > 0.05, n = 5; sagittal: genotype F (1,12) = 0, P > 0.05, n = 3). Data presented as mean + SEM.

Altered ultrasonic vocalizations in mice carrying a cortical Foxp2 deletion

Ultrasonic vocalizations are a component of social interactions in adult mice and can provide a sensitive and reliable proxy for genetic effects on social behavior (48,49). The use and structure of USV varies with context and affective states (40,50–52). We therefore studied USV of adult cKO and WT male mice in two social contexts, during courtship interactions with WT females and during same sex interactions with WT males.

The majority of USV are stereotypic calls of single frequency while the minority are ‘complex’ harmonic calls containing frequency jumps. Since the ethologically salient features of USV are largely unknown, a number of classification schemes have been developed. We have analyzed structural and temporal parameters of our USV recordings (Fig. 4) including the adaptation of three previously described methods. The analyses by Enard et al. (25) in mice carrying a humanized knock-in allele of Foxp2 relied on call duration and presence of pitch jumps. Holy and Guo (53) used an unbiased approach to categorize call elements of male USV into syllables with non-random transition probabilities, reminiscent of some characteristics of songbird vocalizations. Arriaga et al. (54) classified calls with jumps ≥10 kHz and considered the direction of the frequency change: down (D), up (U), DD, DU, UD, DDU, UDUD. The major findings resulting from the analysis of each method are summarized in (Figs 5 and 6) and described here:

Figure 4

Ultrasound vocalization analysis. (A) Representative sonogram showing ultrasonic calls from male–female interaction. Red lines delineate call start and end times. (B) Ultrasonic call example; measurement of the pitch (frequency at maximum power) in each time bin in the spectrogram. Pitch determination allows to calculate specific call features such as mean and range. (C) Power Spectral Density displays the used vocalization power across the frequency spectrum. The black line represents the mean for eight recordings and the red bars represent SEM. (D) Sonogram segmented into phrases using the determined gap cut-off value.

Figure 5

USV abnormalities of Foxp2 cKO male mice detected during courtship interactions. Summary of differences in USV modulation according to various call classification schemes: (A) fraction of calls as in Enard et al. short jumps t (16)=2.666; (B) fraction of calls as in Arriaga et al. D: t (16)=2.499, DU: t (16)=2.111; (C) pitch range, Holy and Guo calls classification, calls with jumps: t (16)=2.316; (D) pitch range differences in calls with jumps classified by the height of jump, ≥5 kHz t (16)=2.450, ≥10 kHz t (16)=2.960, ≥15 kHz t (16)=2.858. Mean ± SEM, n = 8–10.

Figure 6

USV abnormalities of Foxp2 cKO males, detected during same-sex interactions. (A) Distribution of vocalization power (PSD) for calls with and without frequency jumps. Arrow points to the apparent drop of power in 50 kHz calls of Foxp2 cKO males. (B) Results of a temporal analysis of song elements within a phrase, ranging the minimum gap between phrases between 0.3–1.5 s: significant differences were detected in the following temporal correlations: (1) inter-call pause to following call duration (t (12)=2.482. ^*P = 0.0286, minimum gap = 0.3099 s (t (12)=2.266, ^*P = 0.0428, minimum gap = 0.3599 s), (2) inter-call pause to following inter-call pause (t (12)=2.250, ^*P = 0.044, minimum gap = 0.8599 s, t (12)=2.712, ^*P = 0.0189, minimum gap = 0.9099 s, t (12)=2.549, ^*P = 0.0255, minimum gap = 1.3599 s, t (12)=2.379, ^*p = 0.0348, minimum gap =1.4099 s. Lines represent the boundaries of random chance correlation as determined by n = 10⁴ permutations with |$\alpha = 0.05$|⁠. Red asterisks mark the values differing between the genotypes while being correlated in nonrandom manner. Mean ± SEM, n = 6–8.

Figure 7

Social behavior alterations of Foxp2 cortical knockout hosts in male–male interaction test. (A) Chronograms of significantly different events, depict the presence and duration of events for a single interaction dyad (host follows visitor, events number t (17) = 2.342, P = 0.0316, duration t (17) = 2.885, P = 0.0103; oral (host)-genital (visitor) contacts, events number t (17) = 2.585, P = 0.0193, duration t (17) = 2.762, P = 0.0133; close contacts, events number t (17) = 0.4659, P > 0.05, duration t (17) = 2.131, P = 0.0480). (B) Manual blind score for the duration of social interaction (following and close contact) (t (17) = 2.729, ^*P = 0.0143), and paw control in a close contact (t (17) = 3.225, ^**P = 0.005). (C) Differences in sequential behavioral transition probabilities: the arrows represent the string of events, the thickness of the arrow is proportional to the probability that the event occurs; grey arrows—events common for all the dyads of tested males, blue arrows—events which are specific for WT–WT dyads, red arrows—events which are specific for cKO-WT dyads; H: host males, V: visitor males. Mean ± SEM, n = 9–10.

Figure 8

Social behavior alterations of WT visitors in male–male interaction test. (A) Chronograms of events: oral (WT visitor) to genital (cKO or WT host) contacts, events number t (17) = 1.714, P > 0.05, duration t (17) = 2.177, P = 0.0439. (B) Differences in sequential behavioral transition probabilities; H: host males, V: visitor males. Mean ± SEM, n = 9–10.

Figure 9

Validation of Ntsr1 bacTRAP. (A) Confocal immunofluorescence shows colocalization of Ntsr1-eGFPL10a transgene with Foxp2 (arrow). (B) qRT-PCR validation of enrichment in IP fraction for Ntsr1 as compared to total input. (C) No difference of Ntsr1 mRNA levels in IP fractions between genotypes (n = 6; t-test, P = 0.908), Mean + SEM.

Call number and call fractions

The total number of calls did not differ significantly between genotypes irrespective of the employed analyses method. In male–female dyadic recordings call numbers ranged between 207–1530 calls (n = 10 WT, mean = 1067, SEM = 127; 8 cKO, mean = 1033, SEM = 167, P = 0.8715) while male–male dyadic recordings ranged between 3–650 calls (n = 10 WT, mean = 161, SEM = 73; n = 9 cKO, mean = 72, SEM = 23, P = 0.2679). To ensure proper sampling in downstream analyses of call features, only recordings with at least 10 calls were considered. All male–female recordings passed this criterion, while four WT and one cKO male–male recordings were excluded from downstream analysis. The Enard et al. analysis revealed a decrease (P < 0.018) in the number of calls with short duration in cKO male–female dyads, when the presence of pitch jumps were considered as ≥50% to its starting frequency (Fig. 5A). The Arriaga et al. analyses detected an increased number of complex D class calls (P = 0.0236) and a trend (P = 0.0504) for a decreased number of DU calls in cKO male–female dyads (Fig. 5B). For male–male interactions, a trend for an increased number of U calls (n = 9–10, t (17) = 2.077, P = 0.0543) was observed. The Holy and Guo analysis did not show significant differences between genotypes.

Mean frequency and pitch range

The mean of the dominant frequency (pitch) across all calls was not different between genotypes. Based on the genotype differences in the fraction of specific complex calls above, we explored the pitch range in complex calls as well as for calls with no frequency jumps in more detail. Defining calls with jumps according to Holy and Guo yielded an ∼14% decrease in the range of pitch in cKO male–female dyads (P < 0.05; Fig. 5C). This may suggest that complex calls in cKOs are less modulated. Using an unbiased method to minimize the size of a frequency jump yielded a similar decrease (P < 0.05) in pitch range for calls with a minimum of 5, 10 or 15 kHz jumps (Fig. 5D). In male–male dyads no genotype effects for pitch range were detected, suggesting that social context influences this USV parameter.

Power spectral density

The power spectral density (PSD) estimate represents the individual usage of the USV frequency spectrum across all calls (Fig. 4C). This analysis provides an overview understanding of how mice are using frequency bands in the spectrum. For example, if mice had increased usage of higher frequencies versus lower frequencies or vice versa, this would be identifiable as a shift in the PSD curve. To infer whether or not the overall frequency usage was different between genotypes, we used a modified Kolmogorov–Smirnov statistic to test for differences in the shape of the PSD. (see Methods). Although the PSD curves were not significantly different at the global level, we did notice a trend for a drop in usage of low frequency 50 kHz audio in recordings from the cKO male–male interactions (Fig. 6A). Mice emit low frequency calls in aversive behavioral situations such as acute restraint stress, male–male aggression or social defeat reflecting a negative affective state (52,55). A caveat to this finding is the lower sampling of calls in male–male interactions as compared to male–female dyads (48,50,52).

Temporal analysis

USV utterance occurs in bouts of vocalization behavior followed by long pauses (53). To explore the relationship between temporal features within vocalization recordings, each recording was first chunked into bouts in an unbiased manner by ranging the minimum inter-bout gap time between 0.3 and 1.5 s (Fig. 4D). Determining bouts allows subsequent estimation of the true inter-call pause interval by only considering pause times between adjacent calls of a bout and not long periods of silence between bouts. Three linear correlation coefficients (Pearson’s R) were computed across a recording: (a) the correlation between the length of inter-call pause interval to duration of the call immediately following, (b) the correlation of duration of a call to the duration of the subsequent call and (c) the correlation of inter-call pause intervals to the following inter-call pause interval. To determine whether any correlation was greater than expected by random chance, the bout adjacency data were permuted 10⁴ to generate an empirical distribution of Pearson’s R values (black horizontal lines, Fig. 6B). In general, call duration was correlated positively to subsequent call duration, while inter-call intervals were significantly correlated to following call duration or following inter-call interval. Between genotypes, we detected that cKO showed a slight increase to the correlation of inter-call pause interval to subsequent duration time, and a slight reduction to the correlation of inter-call pause interval to subsequent inter-call pause interval (Fig. 6B). Taken together, the results demonstrate differences in USV between cKO and WT animals, including changes to both spectral repertoire and temporal dynamics.

Foxp2 cortical knockout mice display altered social interactions

Previous studies in mice carrying Foxp2 germline mutations revealed abnormal postnatal developmental trajectories such as reduced weight gain and maturation of righting reflexes, which may confound behavioral analyses (56–59). We studied the postnatal development of cKO mice by using a battery of postnatal milestone tests and monitored weight gains. These studies revealed no differences between cKOs and WT littermates (Supplementary Materials, Fig. S1 and Table S1), suggesting that cortical Foxp2 deficiency does not contribute to the impaired postnatal development observed in Foxp2 germline mutants. Furthermore, while Foxp2 knockdown in rat pups has been shown to delay their maternal retrieval (60), our data suggest that cKOs received sufficient care.

We went on to study social behaviors in adult cKOs and WT littermates by employing an unbiased automated screening (ICY Mice Profiler). This test scores 29 parameters during dyadic male–male interactions in a social interaction test (61,62). The tested mouse (either cKO or WT) is designated here as ‘host’ while the ‘visitor’ is an unfamiliar WT. They are brought in contact in an arena unfamiliar for both mice to test for novel social interaction rather than aggression as is traditionally done when adding one mouse into the home cage of another (frequently designated as resident—intruder test). In addition, we have shown that this test situation also favors the production of USV (48,63). Among the scored parameters, we found that the following behaviors were significantly (t-test, n = 9–10, P < 0.05) reduced in cKO-WT as compared to WT-WT interactions: the number and duration of following events initiated by the host, the number of chasing events (following behavior when speed of the host > visitor), the number and duration of oral–genital contacts initiated by the host and the duration of staying in close distance (<1 cm) (Fig. 7). We also observed a significant increase in the duration of oral–genital contacts initiated by WT visitors interacting with cKO as compared to WT hosts (Fig. 8A). Furthermore the number of events when mice were closely (<3 cm) juxtaposed in similar positions (head to head, tail to tail) tended (P = 0.058) to be decreased in cKO-WT interactions.

In addition, manual blind scoring of video recordings for unambiguous direct physical interactions (paw control: the approaching mouse touches the others back with forelimbs) showed that cKOs contacted the visitor significantly less often than WTs (Fig. 7B). In the elevated plus maze test, we found no genotype differences, indicating that the observed decreased approach phenotypes and lower dominance in cKOs are not related to a generally increased anxiety (Supplementary Material, Table S2).

The dynamic interaction repertoire was visualized in transitional behavioral graphs to summarize the action sequences and transition probabilities from one event to another (62; Figs 7C and 8B). This data representation showed that WT hosts are more likely to follow the visitor whereas cKO hosts are more likely to escape when visitors approach them (Fig. 7C). Furthermore, the behavior of visitors appears to be different depending on the genotype of the host. For instance visitor positioning behind a cKO host was more likely to be followed by a stop event of the visitor, which is thought to constitute a behavioral decision point (62; Fig. 8B). One possible interpretation is that the WT visitor differently integrates in its decision the actions of WT and cKO interaction partners, as we have shown previously in other conditions (64). Taken together these results suggest that cortical Foxp2 deficiency is associated with alterations of social behavior and with different reciprocal actions in conspecifics.

Autism candidate gene Mint2 is downregulated in the cortex of cKOs

We assessed whether alterations in social behavior are correlated with transcriptional abnormalities in the cortex of adult Foxp2 mutant mice. To this end, we employed bacterial artificial chromosome transgenic mice expressing eGFP-tagged ribosomal subunit L10a specifically in Ntsr1⁺ layer VI (Ntsr1-L10eGFP). The Ntsr1 promoter has been shown to drive transgene expression in nearly all Foxp2⁺ corticothalamic layer VI neurons (37,65–67) as we confirmed in immunostainings (Fig. 9A). The approach permits cell specific detection of actively translated mRNAs by translating ribosome affinity purification (bacTRAP) in adult mice (68). The Ntsr1-L10eGFP mice were intercrossed with Foxp2 heterozygous knockout animals (Foxp2^S321X/+ (58)), which models the haploinsufficieny in humans and allows a simple one-generation cross (58). Ribosome bound mRNAs (IP) were isolated, RNA-seq libraries were generated by the Smart-seq2 protocol (69) and sequenced. Validation experiments by qPCR showed that layer VI specific Ntsr1 transcripts were enriched in the IP fraction as compared to input (Fig. 9B). Ntsr1 expression levels were not different between genotypes, indicating that Foxp2 deficiency does not alter Ntsr1 expression (Fig. 9C), thus the amount of eGFP-tagged polysomes should be similar in WT and Foxp2^S321X/+ mice.

Among the 14 215 detected genes (>10 reads in at least six samples) a total of 34 were differentially expressed between Foxp2^S321X/+ and WT littermates at FDR 0.05, of which 11 genes showed also an enrichment in the IP fraction (Table 1). Foxp2 showed ∼2-fold lower expression in Foxp2^S321X/+ animals, consistent with our previous results (58). Among the differentially expressed IP transcripts, KIAA0319 is associated with reading disability (dyslexia) (71) and altered activation in language-related brain areas (72). Particularly noteworthy in the context of our study was the downregulation of Mint2 (Apba2/X11L) in Foxp2 mutants. MINT2 is implicated in ASD and schizophrenia (73). Mint2 is a neuronal adaptor that binds neuron-specific cell surface protein neurexin 1 (NRXN1), an ASD candidate gene (74,75). Mint2 KO mice show characteristic deficits in motivated approach behaviours without abnormalities in other domains such as anxiety, sensorimotor gating or locomotor activity (76). We validated Mint2 expression in Foxp2 cKO males. Using quantitative PCR and western blotting from cortical tissues, we could confirm decreased mRNA and protein levels in cKOs (Fig. 10). Overall, these results suggest that Mint2 levels are decreased in the cortex of Foxp2 cKO mice, and this alteration is possibly contributing to the social abnormalities observed in these animals.

Discussion

We show that homozygous Foxp2 deletion in newborn neurons during corticogenesis does not lead to gross abnormalities of cortical histoarchitecture. This is consistent with studies on other Foxp2 mutants and suggests that the functional behavioral abnormalities are linked to more subtle cellular phenotypes such as neurite growth during development or synaptic plasticity in the adult brain (25,31,58,77).

Number and layer specific expression patterns of layer VI Tbr1⁺ neurons are not affected in cKOs, thus Tbr1 may be regulated either independently of Foxp2 or this regulation happens already at the stage of cortical progenitor cells. The observation that in the adult WT cortex Tbr1 and Foxp2 are partly co-expressed in a fraction of neurons illustrates the cellular heterogeneity of the Foxp2⁺ neuronal population in layer VI. Further experiments are needed to explore, whether the Foxp2⁺/Tbr1⁺ neurons constitute a specific microcircuity for social information processing. This would also help to better understand the functional relevance of autism-associated TBR1 mutations that prevent its interactions with FOXP2 (14).

Layer V neurons in different cortical areas have been implicated in social behaviors (78). In the LMC, subpopulations of layer V neurons that are activated during courtship song production, establish monosynaptic projections to brainstem laryngeal motor neurons (54). This population at least partly expresses Foxp2 and shows a slightly ectopic position in the motor cortex of Foxp2 heterozygous mutant mice (38). In homozygous conditional KO mice, this sparse population of cells and their potential contribution to the observed USV abnormalities remains to be studied.

In addition to changes of Foxp2 cKO behaviors, we observed that WT conspecifics spent longer time in oro-genital contact with mutant males than with their WT littermates. Moreover, two other behaviors showed a trend for augmentation: time the WT visitor was spending behind cKO host and back-to-back posture. The positioning behind the host, followed by the high probability of stop behavior and oro-genital sniffing, may indicate an enhanced interest towards the socially abnormal partner. Back-to-back is a posture which could reflect a risk-prone behavior and is potentially associated with an enhanced collective vigilance pattern (62). This altered perception by an interaction partner suggests that cortical Foxp2 may be involved in biologically relevant social functions, which may incorporate distinct forms of communication, such as USV.

In both types of interactions, male–male and male–female, USV may serve affiliative functions to gather social information for instance about hierarchical status. In addition male courtship USV may be predictors of mating opportunities associated with reward expectations and may also serve to reduce female aggression (79,80).

In cKO male–female interactions the fraction and pitch range tuning particularly of structurally complex vocalizations are altered. In cKO male–male interactions vocalization deficits may be detectable at the level of phrasal timing organization. We interpret these observations with caution for several reasons. USV changes in cKOs appear to be subtle and variables may not be independent of each other indicating the need for larger sample sizes to increase statistical power. While genotype differences mainly affect few specific complex vocalizations, the perception of such differences and their ethological significance for fitness are mostly unknown.

Table 1

Differentially expressed genes Foxp2^+/− versus WT in L10eGFP⁺ fraction

	Transcript enrichment in IP vs Input	Transcript enrichment in Foxp2+/−vs WT	FDR
Acsl3	1.0141	0.7861	0.0127
Slc1a4	1.0412	0.7070	0.0212
Mint2/Apba2	1.0224	0.6829	0.0218
Rab3b	1.2096	1.1188	0.0251
Cntn1	1.2397	0.8586	0.0265
Foxp2	1.5261	0.6782	0.0278
KIAA0319	1.2464	0.7078	0.0369
Cyp7b1	1.2954	0.5390	0.0369
Stt3a	1.3546	0.8004	0.0369
Sema3e	2.0968	0.7315	0.0406
Acp2	1.1531	0.7533	0.0417

Benjamini Hochberg FDR used for multiple testing correction, n = 3/group.

The difference in the fraction of call with downward jumps (D) and downward–upward calls (DU or hdu) in male–female interaction is of interest, since similar call structures have been previously described as being stimulus-specific (male call towards female versus calls toward males stimuli) (50). Furthermore, it has been suggested that enrichment in DU calls is a measure of emotional ‘enthusiasm’ (40,50). A fraction of both call types (DU and a subfraction of D calls-hh) are specifically altered in Foxp2 heterozygous mutants in response to female urine as compared to water (40). One possible interpretation is that cortical Foxp2 regulates a fraction of female-oriented complex calls which play an important role in mating success. Intriguingly recent experiments suggest that males produce more complex and louder vocalizations in response to fresh female urine as compared to live females. However, in playback experiments females prefer these complex vocalizations. Therefore, complex calls have been suggested to serve as a remote attraction signal, when females are detected by olfactory cues. Once females are attracted into close vicinity, males may switch to simpler, stereotyped calls (51). Thus, playback experiments of male cKO USV to females could begin addressing the issue whether cortical Foxp2 is indeed required for mate attraction. The host–visitor paradigm used here was previously shown to involve the establishment of a dominant–subordinate relationship that can be, in some strains, associated with agonistic behaviors (48). A recent integrated analysis of social interaction behavior and USV showed that vocalizations are emitted during specific behavioral sequences especially during close contact or approach behaviors (52). The results from our Mouse Profiler experiments and manual scoring suggest alterations of these behavioral sequences and hence may influence our USV results. Recording USV from males exposed to female urine only may further allow reducing these confounders. In this context, more detailed studies of the 50 kHz peak in male–male interactions may be warranted. Furthermore, while it is thought that in host–visitor paradigms, the host animal is the vocalizing subject (79) a recent study demonstrates that also females emit USV during courtship interactions (81). Such vocalizations of WT visitors likely may have been captured in our recordings and may potentially deflate genotype effects.

Figure 10

Mint2 downregulation in the cortex of Foxp2 cKOs. Left panel: qRT-PCR results on the full-depth cortices, n = 4–5, ^*P < 0.05. Right panel: western blot, performed on tissue punches from lower cortical layers, n = 7, ^*P < 0.05. Two-tailed t-test, Mean + SEM.

RNA profiling results in the cortex of heterozygous Foxp2 mutants suggest dysregulation of genes involved in social cognition and behavior. Downregulation of one particularly relevant candidate gene Mint2 was confirmed in the cortex of Foxp2 cKO mice. Mint2 is a neuronal adaptor that binds to proteins essential for synaptic vesicle exocytosis and to intracellular domains of Nrxn1, a synaptic adhesion molecule strongly implicated in ASD and other neurodevelopmental disorders (74,75,82). Several genomic variants of MINT2/APBA2 have been associated with ASD (73). Gene co-expression network analysis of ASD cortical tissues yielded MINT2/APBA2 as hub gene in the network module with the highest correlation to disease status (83). In addition the gene may show signs of recent adaptive evolution in humans (84,85). Behavioral studies of Mint2 KO mice revealed characteristic deficits in motivated approach behaviors in the absence of generally increased anxiety (76). Mint2 KO mice showed subordinate behaviors under competitive conditions with WT littermates. Moreover, approach and following behaviors during social interactions was significantly decreased in these mutants. These phenotypes are remarkably similar to those observed in this study and support our conclusion, that cortical Foxp2 in mice is required for the normal expression of specific social behavior.

Materials and Methods

Animals

Mice were on C57BL/6 background. Experiments on adult mice were performed on males, age 3–5 months. Animals were maintained under a 12 h light/dark cycle. Food and water were supplied ad libitum. Experiments were performed in accordance with French (Ministère de l’Agriculture et de la Forêt, 87–848) and European Economic Community (EEC, 86–6091) guidelines for the care of laboratory animals.

Histology

Perfusion

Mice were rapidly anesthetized with pentobarbital and perfused transcardially with saline solution, followed by 4% (w/v) paraformaldehyde PFA in 0.1 M PB, pH 7.5 (PB-PFA 4%). Brains were removed and postfixed overnight, then cryoprotected in 10% sucrose-PB solution and snap-frozen in Isopentane. Brains were cut in cryostat (CryoStar™ NX70, Thermo Scientific, Waltham, Massachusetts).

Nissl staining was performed with 0.025% Thionine dye on three consecutive 30 μm sagittal sections.

Immunofluorescence

sections (10 μm) were rinsed in PBS and subjected to antigen retrieval in 10 mM citrate buffer with 0.05% Tween20, pH 6, for 10 min ∼100°C. After cooling down sections were permeabilized in PBS containing 0.25% Triton (PBS-T) for 5 min, then blocked in PBS-T with 0.2% gelatin from porcine skin for 1 h. Slices were incubated overnight at 4°C with primary antibodies: goat anti-Foxp2 N16 (Santa Cruz, 1:500); rabbit anti-Tbr1C (a gift from Robert F. Hevner, Washington University, Seattle, USA) 1:2500); rat anti-Ctip2 (Abcam, 1:500) in blocking solution. The next day, sections were washed two times in PBS-T, blocked for 5 min and incubated for 2 h at RT with species-specific fluorophore (Cy5, Cy3 or 488)-coupled secondary antibodies (Jackson Immunoresearch, West Grove, Pennsylvania). Sections were washed three times in PBS-T, stained with Hoechst in PBS and mounted in Vectashield (Vector Laboratories, Burlingame, California).

Image analysis

Pictures were acquired with NanoZoomer scanner (Hamamatsu, Japan), at 20× resolution. After conversion into Tiff, 8-bit format with ImajeJ plugin NDPI tools, pictures were analyzed using ICY software (86).

Brain regions were identified using mouse brain atlas (Paxinos and Franklin, 2007). For cortical thickness analysis three sagittal sections levels corresponded to the following lateral coordinates: 3.36 mm, 1.56 mm, 1.08 mm. In cell counting analysis three sagittal section levels corresponded to 3.6 mm, 1.32 mm, 1.2 mm. Coronal sections levels corresponded to the following bregma coordinates: +0.26 mm, +0.14 mm, +0.62 mm.

Cortical thickness was measured using linear ROI, each ROI positioning was anchored to particular morphological features of the area. Cells were counted with a help of ICY Spot Detector plugin (87), applying the following parameters: for Tbr1-scale 4 (13 pixels), sensitivity 160 and for Ctip2^high-scale 4 (13 pixels), sensitivity 90.

Tbr1–Foxp2 colocalization was detected with the ICY Colocalizer protocol, using the following parameters: Colocalizer Max distance 8, Spot Detector Foxp2-scale 4, sensitivity 250, Spot Detector Tbr1-scale 4, sensitivity 160.

Gene and protein expression

Full cortical tissue was obtained by midline coronal segmentation on rostral and caudal parts of fresh brain tissue, followed by dissection of cortical hemispheres from diencephalon, basal ganglia structures and hippocampus. Brain punches preparations were previously described (88). Punches were collected at the level of rostral cortex (frontal association–primary somatosensory cortex), caudal cortex (auditory cortex) and striatum. Lower cortical layers punches were collected using 0.75 mm punch of Brain punch set (Stoelting Co, Chicago, USA) from caudal cortex areas including caudal motor and primary somatosensory parts, secondary somatosensory, parietal association, auditory and visual cortices (bregma −1.06 to −2.7 mm).

RT-PCR and qPCR

Total RNA was extracted using TRIzol reagent (Ambion, Foster City, California, Life technologies, Waltham, Massachusetts), cDNA was generated using random hexamers (Invitrogen, Waltham, Massachusetts, Life technologies) and Superscript III Reverse Transcriptase (Invitrogen, Carlsbad, CA, Life technologies) following manufacturer protocols. qPCR was performed on the Agilent Mx3005P real-time PCR machine by using SYBR green master mix (Applied Biosystems, ABI, Foster City, CA). Hprt1 was used as an endogenous control for normalization (89). Relative gene expression was determined using the comparative CT method as described (90). Specific intron flanking primers were designed using Primer3 (91) and the sequences (5′–3′) were the following: Foxp2 forward TAGACCTCCCTTCACTTATGCAA, reverse TCCACACTGCTCCTTTAACATTT, Mint2 forward TGGCATCATTTCCAAGCTTTG, reverse CTTAGCAGCCTTCTGCATCC. Equal efficiency of target and endogenous reference primers were verified using standard curve where the difference between the absolute value of slopes between target amplification and endogenous reference amplification was <0.1. Specificity of DNA products was verified with dissociation curves, agarose gel electrophoresis and sequencing.

Immunoblotting

Western blots were performed as previously described (88). Protein lysates (50 μg) were separated by SDS–polyacrylamide gel electrophoresis before electrophoretic transfer onto nitrocellulose membranes (Hybond Pure, Amersham, Orsay, France). Membranes were incubated for 1 h at RT in a blocking solution of TBS-T (0.01%) with 1 or 5% fat-free powdered milk. Membranes were than incubated overnight at 4°C with primary antibodies (anti-Foxp2 and anti-Tbr1 were the same as used for immunostaining, rabbit anti-Mint2 (Sigma, 1:1000), mouse anti-Actin (Millipore, 1:10000). Secondary antibodies were IgG IRdye800CW-coupled or IgG IRdye700DX-coupled (Rockland Immunochemicals, Gilbertsville, PA). Fluorescence was analyzed at 680 and 800 nm using the Odyssey infrared imager (Li-Cor, Lincoln, NE).

Quantitative blots for Tbr1 and Mint2 were performed on 40 μg total protein from lower cortical layer punches. Tissues were treated in RIPA buffer (50 mM Tris–HCl, pH 8.0, 0.1% (w/v) sodium dodecyl sulfate, 0.5% (w/v) sodium deoxycholate, 1% (v/v) Nonidet P-40, and 150 mM NaCl) with protease inhibitors (Complete protease inhibitors cocktail tablets, Roche Diagnostics), on PVDF membranes (Immobilon-P, Millipore). Protein intensity was quantified using ICY ROI Statistics.

Behavior

Neonatal behavior development was evaluated as previously described (92,93).

Elevated plus maze

The test was performed according to Komada et al. (94). A mouse was placed in the center area of the maze with its head directed toward a closed arm. Mice were allowed to move freely about the maze for 10 min. The distance traveled, the number of entries into each arm, the time spent in each arm and the percent of entries into the open arms were measured using an automated video tracking system (Viewpoint, Lyon, France). The number of head-dips and rearings were scored manually.

Male–male interaction was assessed as described (62) in a host-visitor paradigm, where the tested mouse was a host (or isolated host-mouse) and visitor (or social visitor) was a WT conspecific. Semi-automated social behavior scoring was performed for 4 min using ICY plugin Mice Profiler. Manual blind scoring of video recordings for paw control and index of aggressiveness was performed according to Coura et al. (61).

Male–female interaction was assessed as described (79). Briefly, male tested mice were single housed for 5 days before an experiment. On the third and fourth days of separation animals were habituated to the test environment for 15 min; on the test day animals were habituated to the environment for 5 min before presentation of a WT female. Interaction was performed in the male home cage and recorded for 5 min. Females were superovulated by an injection of pregnant mares serum gonadotropin (5 U per mouse), and 48 h later, of Chorulon (5 U per mouse), 12 h later females were presented to the males. Hormones we purchased from Laboratoire INTERVET, France.

Ultrasonic vocalizations

USV were recorded during social interaction experiments (62) as we described previously (52). A condenser ultrasound microphone was placed above the experimental chamber high enough for receiving angle of the microphone to cover the whole area of the test cage. USV were recorded under sampling frequency of 250 kHz;16-bit format. All recording hardware and software were from Avisoft Bioacoustics, Germany. Raw digital audio data was converted to voltage units given peak-to-peak Voltage of the UltraSoundGate hardware at 10 dB gain (± 0.25 mV). Spectrograms of recordings were prepared using a FFT size of 512 samples, with 50% overlap, multiplied by a Hamming window, which provides a temporal resolution of 1.024 ms and a frequency resolution of 488.3 Hz. Two filter frames were used for subsequent analysis: bandpass filter 25–120 kHz (according to Holy and Guo (53)) and 40–120 kHz (an optimal noise thresholding for the current dataset). Locations of ultrasonic calls were detected using the method of Holy and Guo (53), using the custom whistimes() function in MATLAB with the following parameters: minimum spectral purity: 0.2, maximum spectral discontinuity: 1.0, minimum duration: 5 ms, minimum intercall pause: 30 ms.

Spectral characteristics of the pitch were determined by finding the fast Fourier transform (FFT) frequency with peak power in each time window over a call, for all calls. In order to segment the spectrogram into phrasal units, a range of inter-phrase gap sizes was tested. This range uses the peak of the distribution of pause times as an estimate of the true mean, and ranges cut-off values out to five standard deviations from this value (∼0.1–1.5 s). The actual values of the cutoffs differed slightly in each experiment (male–female, male–male) since the cut-offs are determined from the empirical distribution of pauses. To understand group differences in usage of the frequency spectrum, PSD was estimated from the FFT using the periodogram method. Periodograms were computed as (|X(k)|²)/(N·F), where the scaling 1/(N·Fs) (N = FFT size, Fs = sampling rate) ensures conservation of power by Parseval’s theorem (Σx(t)²Δt = 1/(N·Fs) (Σ_k|X(k)|²), where Δt = 1/Fs, and k is frequency indexed from 0:N−1). Two-sided periodograms (i.e., including negative frequencies) were converted to one-sided by doubling the values for frequencies with negative equivalents, resulting in a real-valued spectrum for N/2 + 1 frequencies in standard units of V² Hz⁻¹. Periodograms were time averaged within each call and over all calls for each animal.

Cell type-specific RNA purification by translating ribosome affinity purification (BacTRAP)

Mice (3 months old) were decapitated and fresh brains immediately processed according to Heiman et al. (95). Cortical hemispheres from one animal were dissected from diencephalon and basal ganglia structures, hippocampus and a tip of a rostral cortex, and processed in 2 mL of tissue-lysis buffer. A total of 30 μl of the supernatant (S20) was collected as an input sample immediately before immunopurification (IP). Affinity matrix was prepared with Dynabeads Protein G (Life Technologies, Novex) (50 μl per sample) and anti-GFP rabbit serum (Life Technologies, Molecular Probes) (10 μl per sample), and beads were washed in a low-salt buffer. IP lasted for 4–6 h, on ice. Buffers composition differed from Heiman et al. in HEPES-KOH (10 mm) and MgCl₂ (5 mm). RNA quantity was accessed with Quant-iT RNA Assay kit (Life Technologies, Invitrogen). RNA yield for one animal varied between 1.2–5.7 ng, RNA quality was assessed using Bioanalyzer (RIN ≥ 7.8).

RT-qPCR analysis was performed as above. Primers for Ntsr1 were used from PrimerBank (ID: 9055296a1)

Library preparation for RNA sequencing was performed using the Smart-seq2 protocol (69).

Libraries were pooled after incorporation of dual Nextera indices in equimolar amounts. Sequencing was done on an Illumina HiSeq1500, using the Rapid Mode. Samples were demultiplexed using deML (96). The reads were mapped with ngm (version 0.4.10) (97) against the mouse genome (version mm10) [supplemented by eGFP sequence]. PCR duplicates were removed using custom perl scripts. HTseq was used to count reads overlapping genes unambigiously. For this, the ensembl annotation (version GRCm38.74) was adapted by including GFP sequence annotation.

Statistics

Histology and behaviour

Analysis was performed using either unpaired two-tailed T-test when two groups were compared or 2-way ANOVA for > two groups.

USV analysis

For comparisons between the groups means of single variables (fractional calls comparisons, mean and range of the pitch, phrase pause length, correlations between pauses and duration times, a chi-square statistic for Markov transition probabilities) simple t-tests were employed.

As an omnibus test to determine whether there were changes in usage of the frequency spectrum, as represented by PSD(x) (Amplitude² Hz⁻¹ for x, a frequency, in Hz), a bootstrapping procedure was performed using a Kolmogorov–Smirnov like statistic; comparison of averaged power spectra (represented as estimated power spectral densities, PSD in units of Amplitude2 Hz-1) were performed using a bootstrapping procedure, with the test statistic in question being the Kolmogorov–Smirnov D statistic:

$$ D=\max \left( abs\left({\frac{PSD(x)}{\sum PSD(x)}}_{WT}-{\frac{\mathrm{P} SD(x)}{\sum PSD(x)}}_{cKO}\right)\right) $$

which represents the maximum difference between two spectra. In each iteration, bootstrapped samples were computed by sampling with replacement from the vectors representing the PSD for either WT or cKO, and an average computed for each followed by computing Di for the ith iteration. This generates a distribution of maximum differences (Dtest) in the spectra between genotypes for which a 95% confidence interval can be computed. To generate a null distribution for D, the procedure was repeated but shuffling genotype labels (Dnull). The procedure was repeated 104 times which is the maximum magnitude of difference between two normalized cumulative distributions. The test and null distributions were Dnull and Dtest were compared by three possible decision criteria: (a) if there is no overlap between the confidence intervals of Dnull and Dtest, reject the null hypothesis that the spectra come from the same underlying population of spectra (most conservative); (b) allow some overlap between the intervals and compute a P-value representing the probability of overlap (number of Dtest samples found within Dnull divided by the total number of bootstrapped samples), probability of overlap as the number of bootstrap samples that occur in the region of overlap divided by the total, if P overlap <0.05, reject the null hypothesis; (c) compute the probability of observing a value greater than or equal to the mean of Dtest or greater in the Dnull distribution (number of Dnull samples greater than or equal to mean (Dtest) divided by the number of samples). This procedure is analogous to computing D directly from their means instead of bootstrapping the test samples.

BacTRAP sequening analysis

Analysis was performed in R (DESeq2, Bioconductor, version 1.4.0). The two major pairwise comparisons, Input against IP and WT against Foxp2 heterozygous knockout, yielded approximately the same number of significant calls. Descriptive analysis included principal component analysis and sample-wise clustering based on normalized counts. The samples could be clearly and correctly separated by RNA source. Likelihood ratio test was used to determine the significance of the covariate genotype in a reduced model compared to a full model including the covariate RNA source and genotype in explaining average gene expression differences. The P-value was adjusted for multiple testing (Benjamini-Hochberg FDR correction).

Data repository

The RNA seq raw data are available under GEO accession number GSE109596.

Acknowledgements

JDD is a NARSAD Independent Investigator of the Brain and Behavior Research Foundation. We thank Dr Ines Hellmann for supporting mRNA-seq experiments, Dr Robert Hevner for providing Tbr1 antibody and Dr Cataldo Schietroma for helpful comments.

Conflict of Interest statement. None declared.

Funding

Ecole des Neurosciences de Paris-ENP (VM, MG), the Inserm/CNRS ATIP-AVENIR programme (MG), the Institute de France/Fondation NRJ (MG), Fondation pour la Recherche Médicale en France-FRM (MG, DEA20090615981), the Agence nationale de la recherché (MG, ANR-13-ISV4-0004), the ‘Investissements d’Avenir’ program managed by ANR (ANR-11-IDEX-0004-02) and the NIH (JDD, 1R01MH107515 and 5R01HG008687).