Male zebra finches (Taeniopygia guttata) were bred in our colony at the Freie Universität Berlin. All procedures were performed according to the guidelines of the governmental law (TierSchG), under permits granted by the local Berlin authorities governing research involving animals. All birds were sacrificed with an Isoflurane overdose.

Co-Immunoprecipitation (Co-IP) was done using Dynabeads® Protein G (Invitrogen, Cat.No.100.04D) following the manufacturer’s protocol with a few changes as follows. We first incubated the antibody-Dynabeads mixture for 15 min at room temperature (for antibody concentrations refer to Table 1). We subsequently incubated the protein extracts plus antibody-coated Dynabeads mixture for 30 min at 4°C with rotation. We eluted with 20 μl of elution buffer (50 mM Glycin pH 2.8) and before a second round of immunoprecipitation, we neutralized the elution buffer by adding 4.8 μl 1 M TrisHCl pH 7.5 and then added washing buffer to bring the volume to a total of 200 μl. After elution we added 15 μl of 2× Laemmli and prepared lysates for denaturing conditions.

Transfected HeK293 cells were treated with M-PER (Thermo Scientific, Prod#78505) lysis medium for 15 min on ice. Extracts were centrifuged for 10 min at 1500 g and supernatant was stored at −80°C until used. Cell extracts and co-IP’ed proteins were separated by SDS PAGE (8%–10%), transferred to a polyvinylidene fluoride membrane (Roche, Indianapolis, IN, USA), and blocked with Roti-Immunoblock for 2 h or overnight at 4°C. The membranes were then incubated with the desired antibody (Table 1) overnight at 4°C. Membranes were subsequently washed 3× PBS/0.1% Tween 20 followed by incubation with an HRP-conjugated antibody raised in the appropriate animal (1:2000 dilution; Amersham Biosciences) for another 30 min. Binding was detected on X-ray films using an ECL detection system for HRP (Perkin-Elmer, Boston, MA, USA). Films were developed in a Curix 60 developing machine (Agfa, Cologne, Germany). After the first detection membranes were washed in PBS/0.1% Tween for 5 min, then washed in 0.5 NaOH for 10 min, washed again in PBS/0.1% Tween for 5 min and blocked again and detected with a second/third primary antibody as described before.

After sacrificing the bird, the forebrain was quickly dissected, the hemispheres separated, embedded in TissueTec and frozen on dry ice. Hemispheres were stored at −80°C until further processing. Area X microbiopsies were punched as described previously (Olias et al., 2014; Adam et al., 2016). Protein was extracted from pooled microbiopsies of the same animal after we had confirmed correct targeting of the desired brain region (Olias et al., 2014; Adam et al., 2016).

For SV40 we seeded ~30000 HeLa cells in 200 μl DMEM medium (GIBCO) containing Penicillin-Streptomycin (Lonza-DE17–602E, 10 UI/ml) per well of a 96-well white flat bottom plate (Nunclon, Cat.No.136101, Denmark). Transfection was performed using Lipofectamine™ 2000 (Invitrogen) following manufacturer’s protocol. Briefly, after seeding plates were incubated for 24 h at 37°C at 5% CO₂. For transfection, the medium was exchanged with 100 μl of antibiotic free medium before adding the 50 μl transfection mix in each well. Transfection mix consisted of two parts: (a) 25 μl of OptiMEM (GIBCO) containing 30 ng of pGL4.13 (Luciferase gene driven by the SV40 promoter which is known to be regulated by FoxP subfamily members) and 30 ng of pGL4.75 (Renilla gene driven by the CMV promoter that is not affected by FoxP subfamily members, used for normalization of expression changes) and 250 ng total vector over-expressing the different FoxPs for each well; and (b) 1 μl of Lipofectamine™ 2000 (Invitrogen) in 25 μl OptiMEM that was premixed for 5 min at RT. After combining (a) and (b) the mix was incubated for 20 min at RT, then it was added to the cells. After 4–6 h of incubation we changed the medium to 75 μl of antibiotic containing medium and incubated for further 48 h at 37°C in a CO₂ incubator.

For VLDLR and CNTNAP2 we used HeK293 cells, and luciferase assays were done as described previously (Adam et al., 2016). Depending on the promoter, we performed between 4 and 7 luciferase assays, each after an independent transfection. Within each assay we used triplicates, e.g., three wells containing the same transfection reagents and quantity of cells. The mean of the triplicate was used for statistical analysis. Each plate was measured once in the ELISA reader.

We measured luminescence using the Dual Glo Luciferase Kit (Promega) following manufacturer’s protocol in an Elisa plate reader (Tecan, GENios; Switzerland). Mean background from untransfected wells was subtracted from all other wells. We present Luciferase results as mean Relative Light Units (Luciferase RLU/Renilla RLU) calculated from the normalized values of 4–7 independent assays.

The pGL4-VLDLR promoter was cloned as described (Adam et al., 2016). The CNTNAP2 promoter was cloned using DNA obtained from a blood sample of an adult bird as the template and amplified with the forward primer 5′-TTGCCTCATTGATTGCAGAA-3′ and reverse primer 5′-CCTGCTTTTCTCCACTTTGG-3′ using High Fidelity Taq (Fermentas K0191). The resulting PCR product was examined on an agarose gel, cleaned from nucleotides with the Nucleo Spin Gel and PCR Clean-up (Macherey-Nagel, Germany, Ref 740609.250), and cloned into the pCR4Blunt-TOPO vector of the Zero blunt PCR cloning kit (Invitrogen) according to the manufacturer’s protocol. Inserts from three independent CNTNAP2 clones were fully sequenced to confirm the sequence of the promoter region of CNTNAP2. Zebra finch CNTNAP2 sequences were deposited to GenBank, accession number NCBI KX943238. We then used forward primer 5′-GATGCTAGCTTGCCTCATTGATTGCAGAA-3′ and reverse primer 5′-GATAGGCCTCCTGCTTTTCTCCACTTTGG-3′ and subcloned the PCR product into the NheI and StuI sites of the pGL4.13 vector to generate the pGL4-CNTNAP2 vector used in Luciferase assays.

We identified putative CpG islands in the CNTNAP2 promoter with the CpG plot tool². We calculated the GC percentage in the GC rich region of the promoter region of CNTNAP2 using the CpG island calculator³.

Overexpression vectors of FoxP1/2/4 tagged with FLAG or V5 were previously generated (Haesler et al., 2004; Mendoza et al., 2015). For the generation of Myc tagged FoxP2 we used the FoxP2-V5 tagged sequence as a template and used the forward primer 5′-CGCGGATCCGCCACCATGATGCAGGAATCTGCGACAG-3′ and reverse primer 5′-GCGGAATTCCTACAGATCCTCTTCTGAGATGAGTTTTTGTTCTTCCAGATCTTCAGATAAAGGCTC-3′ and cloned it into pcDNA3, 1 + vector (Invitrogen).

Proteins for electrophoretic mobility shift assays (EMSAs) were purified using the Protino Ni-NTA Agarose (Macherey-Nagel, Germany, 745400.25) according to manufacturer’s protocol. Protein was quantified using BCA1 (Sigma). EMSA assays were carried out as published previously, using the Oligo described for VLDLR (Adam et al., 2016). The oligo we used for CNTNAP2 was 5′-TATTATTATTTATTTTTGTACTCTACATTCCTTGTTATTTGATACT-3′ (in bold presumed FoxP binding sites containing the ATTT core sequence).

Statistical tests were performed using the data analysis software R (R Core Team, 2013). After testing for normality, differences between Luciferase experiments were calculated with an analysis of variance (ANOVA) and a Tukey’s HSD post hoc test for pairwise comparison (R Core Team, 2013). Graphs were prepared with GraphPad Prism 4.0 (GraphPad Software, San Diego, CA, USA). Data are expressed as mean of means ± SEM.

To determine whether zebra finch FoxP1/2/4 can homo- and hetero-dimerize in cell lines, as described for the mouse FoxPs (Li et al., 2004a) and human FOXPs (Sin et al., 2015), we transiently overexpressed full length constructs of zebra finch FoxPs tagged either with FLAG or V5 in HeK293 cells and performed co-immunoprecipitation assays. In each Co-IP experiment we used four protein lysates; empty vector lysate (control 1), lysate with one of the two proteins to be tested for co-immunoprecipitation (control 2), another lysate with the other protein to be tested (control 3), and the lysate containing both proteins (experimental). Lysates were co-IP’ed with FLAG antibody, and membranes were first detected with V5 antibodies and sub-sequentially detected with FLAG antibodies. FLAG tagged proteins were immunoprecipitated as expected from lysates containing FLAG-tagged proteins whereas this was not the case for cells expressing an empty vector, or from cell lysates containing V5 tagged proteins (Figures 1A–F). In all experiments we detected a protein co-immunoprecipitated in lysates co-transfected with two differently tagged FoxPs (Figures 1A–F), indicating protein-protein interactions of the co-expressed FoxPs. Lastly, in all cases the supernatant was mostly depleted of the co-IP’ed protein (Figures 1A–F). From these results we conclude that zebra finch FoxP1/2/4 homologs are able to form homo- (Figures 1A–C) and hetero-dimers (Figures 1D–F) in cell lines.

To examine whether FoxP1/2/4 proteins also interact in neurons of the zebra finch forebrain, we first characterized commercial antibodies against FoxP1/2/4 zebra finch proteins by over-expressing them individually in HeK293 cells and performing Western Blots. Specific antibodies show a band of about 80 kDa molecular weight in each case. Each antibody only recognized one FoxP (Figures 2A–C, top panel), and not the other two. The faint bands in the FoxP1 and FoxP2 lanes in panel c do not correspond to cross reactivity of FoxP4 antibody against FoxP1 and FoxP2 protein, which have a slightly different molecular weight, as can be seen in the FLAG tagged versions (middle panel). The subsequent detection with FLAG (and actin antibodies, lower panel) also demonstrates that over-expressed proteins were present in the protein lysate in similar quantity (Figures 2A–C). From these results we conclude that antibodies are specific for the different zebra finch FoxPs.

We performed co-IP assays on protein lysates of adult male zebra finch forebrain using the specific antibodies described above. FoxP1 monoclonal antibody was not able to pull down the non-denatured, native brain protein (data not shown), and therefore we used FoxP2 and FoxP4 polyclonal antibodies to pull down protein complexes, and all three antibodies for Western blot detection. As a negative control we used IgG antibodies of the same species in which the specific FoxP antibodies were raised. After IP with FoxP2 or FoxP4 antibodies, we detected co-IP’ed FoxP1 (Figures 3A,B). After IP with FoxP2 we detected co-IP’ed FoxP4 (Figure 3C). In contrast, no protein of the expected size was precipitated with IgG controls (Figures 3A–C). From these results we conclude that FoxP1/2/4 form hetero-dimers in forebrain neurons in vivo. To our knowledge this is the first report showing that FoxP1/2/4 proteins can form hetero-dimers in the brain.

Since the majority of FoxP expressing cells in Area X and the surrounding striatum express FoxP1/2/4 (Mendoza et al., 2015), and since it is known that FOXP3 is able to homo-oligomerize and hetero-associate in cell lines (Li et al., 2007; Song et al., 2012), we wanted to assess whether all three neurally expressed FoxP proteins could bind in a single protein complex. To test this we performed double co-IP from HeK293 cell lysates co-transfected with full length FoxP1-FLAG, FoxP2-Myc and FoxP4-V5 tagged zebra finch proteins (Figure 4A schematic). If FoxP1/2/4 were not able to hetero-associate in a complex after double co-IP we should only detect the two proteins that were immunoprecipitated, but if they hetero-associate in a complex we should be able to detect all three FoxPs. We first immunoprecipitated with Myc, to pull down FoxP2 and all proteins bound to it. After the first immunoprecipitation we found that all three proteins had co-IP’ed (Figure 4B). The fact that we detected FoxP1-FLAG and FoxP4-V5 after immunoprecipitating the Myc-taggd FoxP2 could be due to pulled down hetero-dimers of FoxP2 with FoxP1 or with FoxP4. Alternatively, FoxP1/2/4 could have been simultaneously pulled down by immunoprecipitating Myc-tagged FoxP2. To assess this, a second, sequential IP with FLAG was carried out to detect FoxP1-FLAG after the first IP. After the second immunoprecipitation we detected not only FoxP1-FLAG but also co-IP’ed FoxP4-V5 (Figure 4, upper blot).

Not all protein was depleted from the first supernatant, one possible explanation being that FoxP1 homodimers and FoxP4 homodimers were not precipitated by Myc antibodies detecting FoxP2-Myc. Leftover FoxP2 in the supernatant additionally indicates that not all FoxP2 was precipitated by the Myc antibody, likely because of insufficient antibody concentration to deplete all FoxP2 protein and its associated interaction partners. No protein was detected after the second immunoprecipitation in the second supernatant (Figure 4, S2).

In summary, these data suggest that FoxP1/2/4 can form a complex including all three proteins, a result which has not been reported before. In addition to this hetero-oligomerization, we also found homo- and hetero-dimerization of all FoxPs (Figures 1, 3).

To assess whether FoxP1/2/4 can hetero-oligomerize in the zebra finch song system we used nuclear protein extracts of Area X. Double IP using specific antibodies detected the three proteins in the co-immunoprecipitated fraction suggesting that FoxP1/2/4 can also hetero-associate in vivo in Area X (Figure 5).

We assessed whether the homotypic and heterotypic interaction of zebra finch FoxP proteins lead to differential regulatory activity. To do so we used the SV40 promoter driving Luciferase (Photinus pyralis synthetic protein). The SV40 promoter has a putative core consensus sequence for binding (TATTTRT) human and murine Foxp1/2/4 (Wang et al., 2003; Vernes et al., 2006). For the human and mouse proteins it was shown in luciferase promoter reporter assays that each FoxP can repress the SV40 promoter individually (Wang et al., 2003; Li et al., 2004a; Vernes et al., 2006). Likewise, we found that zebra finch FoxP proteins when expressed individually in HeLa cells significantly repressed the transcriptional activity under the control of the SV40 promoter (Figure 6; One way ANOVA; F = 28.79; DF = 7; n = 8; p < 0.0001; Tukey’s Multiple comparison Test; compared to empty vector: FoxP1 p < 0.0001, FoxP2 p = 0.013, and FoxP4 p < 0.0001), in the same range as reported for mice and human FOXP1/2/4. Also, like its mammalian homologs, zebra finch FoxP4 repressed transcription stronger than FoxP1 or FoxP2 did, and FoxP2 was a weak repressor of the SV40 promoter (Wang et al., 2003; Li et al., 2004a; Vernes et al., 2006). Cells expressing combinations of the different FoxP subfamily members also showed a significant repression of the SV40 promoter (compared to empty vector: FoxP1/2 p < 0.0001; FoxP1/4 p < 0.0001; FoxP2/4 p < 0.0001 and FoxP1/2/4 p = 0.0001). There were no significant differences between cells expressing FoxP1 or FoxP4 alone or in any of the combinations tested. We did observe a stronger transcriptional repression in cells expressing FoxP2 in combination with another FoxP subfamily member than when FoxP2 was expressed alone (compared to FoxP2 alone: FoxP1/2 p = 0.016; FoxP2/4 p = 0.0005 and FoxP1/2/4 p < 0.0001). However, we cannot rule out the formation and action of homodimers of FoxPs in those cells. Taken together, FoxP target genes in cells that express only FoxP2 may be less strongly regulated than those that are co-expressed with FoxP1 and/or FoxP4.

To explore whether FoxP proteins not only regulate SV40 but also neurally relevant promoters we chose VLDLR for a number of reasons: (a) it is a direct target of FoxP2 in humans, mice, and zebra finches (Spiteri et al., 2007; Vernes et al., 2007; Adam et al., 2016); (b) both FoxP2 and VLDLR promote dendrite and dendritic spine development in various species (Niu et al., 2004, 2008; Schulz et al., 2010; Vernes et al., 2011); and (c) in songbirds, FoxP2 and VLDLR are expressed in Area X and the basal ganglia (Balthazart et al., 2008; Hilliard et al., 2012) and are co-regulated in different contexts (singing, age and by molecular manipulations; Adam et al., 2016). Most of the work on target genes in the FoxP subfamily that are expressed in the brain have focused on FoxP2, and it is not known how much overlap there is in the binding sites for FoxP1 and FoxP4 and whether they can bind to the same promoter motives. To investigate this we used the VLDLR-oligonucleotide that FoxP2 can bind to in EMSA (Adam et al., 2016) and tested whether FoxP1/4 also binds to this probe. The protein extract of HeK293 cells transfected with empty vector (mock extract) presented a non specific shift (n.s.; Figures 7A,B, first lane). In the presence of FoxP1 or FoxP4 a specific shift of the DNA was observed, indicating that both proteins are able to bind to the VLDLR oligonucleotide (Figure 7A,B, second lane). Addition of a specific competitor diminished the intensity of this band (Figure 7A,B, third lane). Adding a V5 antibody to the FoxP/oligonucleotide mix resulted in a further shift, strengthening the notion that that FoxP1 or FoxP4 generated the shift (Figure 7A,B, fourth lane). From these results we conclude that FoxP1 and FoxP4 can bind to the same VLDLR region that FoxP2 binds to, suggesting that all three FoxPs share a common target gene.

To test whether binding of FoxP1/2/4 to the VLDLR promoter leads to an alteration of transcription, we performed luciferase assays using the pGL4-VLDLR vector described previously (Adam et al., 2016). There were significant differences in activation of the VLDLR promoter depending on the specific combinations of FoxP proteins (One way ANOVA; F = 26.09; DF = 7 and n = 8; p < 0.0001; Figure 7C). Tukey’s Multiple comparison Test revealed that when compared to empty vector, FoxP1 was a significantly better activator of the VLDLR promoter than FoxP2 or FoxP4 (FoxP1 p < 0.0001, FoxP2 p < 0.007 and FoxP4 p = 0.0024). However, FoxP2 and FoxP4 were less effective and not significantly different from each other in their ability to activate the VLDLR promoter (p = 0.999). Combinations of FoxP1/2/4 did activate the VLDLR promoter to a similar degree (compared to empty vector: FoxP1/2 p < 0.0001; FoxP1/4 p < 0.0001; FoxP2/4 p = 0.0003 and FoxP1/2/4 p < 0.0001). Comparing the activation of FoxP2 alone to the FoxP1/2 and FoxP1/2/4 combination (p = 0.0078 and p = 0.012 respectively) we found a significant difference on the VLDLR promoter but not with the FoxP2/4 combination (p = 0.88). There were no differences between the individual activation by FoxP4 and the combinations with FoxP1 (p = 0.10) or FoxP2 (p = 0.98), but there was a significant difference to the FoxP1/2/4 combination (p = 0.034). Taken together FoxP1/2/4 are not only repressors, but are able to bind to the VLDLR promoter and activate it differently depending on the co-expression with other FoxPs.

To further test the regulatory ability of zebra finch FoxPs on a target gene relevant in Area X, we chose CNTNAP2 because: (a) CNTNAP2 is associated with speech disorders, as well as ASD, dyslexia and intellectual disability (Rodenas-Cuadrado et al., 2014) some of which are also associated with FoxP mutations (MacDermot et al., 2005; Vargha-Khadem et al., 2005; Shriberg et al., 2006; Zeesman et al., 2006; Lennon et al., 2007; Pariani et al., 2009; Vernes et al., 2009; Carr et al., 2010; Hamdan et al., 2010; Horn et al., 2010; Bowers and Konopka, 2012; Rice et al., 2012; Žilina et al., 2012; Chien et al., 2013; Le Fevre et al., 2013; Palumbo et al., 2013; Toma et al., 2013); (b) CNTNAP2 is a direct target of Foxp2 in mice and of FOXP1 in humans (Vernes et al., 2007; O’Roak et al., 2011); (c) FoxP1/2/4 and CNTNAP2 are expressed in Area X and the basal ganglia in zebra finches (Haesler et al., 2004; Panaitof et al., 2010; Condro and White, 2014; Mendoza et al., 2015); and (d) FoxP2 binds to the CNTNAP2 promoter region in zebra finches and the expression of both is correlated in adult undirected singing and in non-singing juveniles as well as after FoxP2 lentiviral knockdown (Adam et al., in review; personal communication).

The assembled genome region of the zebra finch CNTNAP2 promoter contained a sequence gap located ~0.75 kb upstream of the start codon of the first exon of CNTNAP2 (Figure 8A). We cloned and sequenced across the gap, identifying a 462bp fragment, which had a high GC-content of up to 79.50% and was within the CpG island. The CpG island was located ~0.83–1.2 kb upstream of the star codon of the first exon. A TATA box was identified ~20 bp of the transcription start site (TSS). This region had 11 FoxP2 binding sites (Figure 8A, red shapes; Pierrou et al., 1994; Stroud et al., 2006; Nelson et al., 2013), four located in the promoter region and 7 in the 5′ UTR region. We were able to close the gap in the genome enabling us to further investigate the previously incomplete CNTNAP2 promoter sequence and use it for luciferase assays.

We used an oligonucleotide probe from a sequence in the 5′ UTR region of the zebra finch CNTNAP2 (Figure 8A) to test whether FoxP1/2/4 bind to it. This oligomer was identified in a parallel study on FoxP2 and CNTNAP2 interactions interactions (Adam et al., in review). FoxP1 was able to bind to the CNTNAP2 oligo and resulted in two shifted bands, consistent with the possibility that it binds in two of the three FoxP2 binding sites (Pierrou et al., 1994; Enard et al., 2009; Nelson et al., 2013) contained in the oligo (Figure 8B, “Materials and Methods” Section). We also confirmed the results of Adam et al. (in review), showing that FoxP2 also binds (Figure 8C). Interestingly, FoxP4 did not result in a DNA shift in the EMSA (Figure 8D).

To investigate whether binding of FoxP1 and FoxP2 to the CNTNAP2 promoter leads to an alteration of transcription, we performed luciferase assays using the pGL4-CNTNAP2 vector (Figure 8E). When tested individually, FoxP1 activated the CNTNAP2 promoter and FoxP2 repressed it. As expected from the lack of binding ability of FoxP4 to the CNTNAP2 promoter oligo (Figure 8D) FoxP4 failed to regulate the CNTNAP2 promoter in the Luciferase assay (Figure 8E; One way ANOVA; F = 21.66; DF = 7 and n = 8; p < 0.0001; Tukey’s Multiple comparison Test; compared to empty vector: FoxP1 p = 0.005, FoxP2 p < 0.0001 and FoxP4 p = 0.91). Moreover, in combination, the different FoxPs also failed to regulate expression driven by the CNTNAP2 promoter, (compared to empty vector: FoxP1/2 p = 0.26; FoxP1/4 p = 0.28; FoxP2/4 p = 1 and FoxP1/2/4 p = 0.22). Interestingly, combinations of FoxP4 with either FoxP1 or FoxP2 did not result in repression or activation, in contrast to FoxP1 or FoxP2 alone, suggesting that hetero-dimerization might affect the regulation of FoxPs. Taken together we show that FoxP1 and FoxP2 can bind to and regulate the CNTNAP2 promoter, with opposing activities. CNTNAP2 is an example of how FoxPs may tune the regulation of different targets genes via homo- and hetero-dimerization.