Dynamic FoxP2 levels in male zebra finches are linked to morphology of adult-born Area X medium spiny neurons

The transcription factor FOXP2 is crucial for the formation and function of cortico-striatal circuits. FOXP2 mutations are associated with specific speech and language impairments. In songbirds, experimentally altered FoxP2 expression levels in the striatal song nucleus Area X impair vocal learning and song production. Overall FoxP2 protein levels in Area X are low in adult zebra finches and decrease further with singing. However, some Area X medium spiny neurons (MSNs) express FoxP2 at high levels (FoxP2high MSNs) and singing does not change this. Because Area X receives many new neurons throughout adulthood, we hypothesized that the FoxP2high MSNs are newly recruited neurons, not yet integrated into the local Area X circuitry and thus not active during singing. Contrary to our expectation, FoxP2 protein levels did not predict whether new MSNs were active during singing, assayed via immediate early gene expression. However, new FoxP2high MSNs had more complex dendrites, higher spine density and more mushroom spines than new FoxP2low MSNs. In addition, FoxP2 expression levels correlated positively with nucleus size of new MSNs. Together, our data suggest that dynamic FoxP2 levels in new MSNs shape their morphology during maturation and their incorporation into a neural circuit that enables the maintenance and social modulation of adult birdsong.

The forkhead box P2 transcription factor (FOXP2) is linked to speech and language disorders. Heterozygous FOXP2 mutations in humans affect both the coordination of fine orofacial movements and language perception [1][2][3] . Because songbirds -like humans -need to learn most of their communicative vocalizations, they offer a unique model to study the role of FoxP2 (for nomenclature FOXP2/FoxP2 see Methods) for vocal learning and for the maintenance of learned vocalizations as adults 4 . Studying the relationship between FoxP2 and vocal learning in songbirds may inform the neurogenetic mechanism underlying the speech deficits in patients carrying FOXP2 mutations for the following reasons. The FoxP2 protein coding sequence is highly conserved between humans and songbirds as are the brain expression patterns, notably in the cerebellum and striatum [5][6][7] . Moreover, genetic manipulations of FoxP2 expression levels in the striatal song nucleus Area X during the critical phase of song learning lead to inaccurate and incomplete imitation of the tutor's song and more variable vocal production 3,[8][9][10] . This phenotype bears similarities to the specific speech deficits called developmental verbal dyspraxia, DVD (or childhood apraxia of speech), that patients carrying FOXP2 mutations suffer from. The core-phenotype of DVD consists of altered precision, consistency and sequencing of movements underlying speech in the absence of neuromuscular deficits 11 . In addition, altered FoxP2 levels in adult Area X affect the dopaminergic modulation of corticostriatal signaling important to song variability and affect song maintenance 12,13 , stressing the fact that tight regulation of FoxP2 expression is a prerequisite for correct neural transmission in differentiated neural circuits. Additional effects of Foxp2 and its disruption on the embryonic development and the function of neural circuits have been described in mice [14][15][16][17][18][19][20][21][22][23] . Further evidence for the biological relevance of tight regulation of FoxP2 expression levels comes from the following studies. FoxP2 expression levels in Area X transiently increase during song learning but are lower in adults 7,24 . Singing decreases overall FoxP2 levels in Area X but not in the surrounding striatum and the degree of FoxP2 down regulation correlates with the amount of produced song [25][26][27] . This relationship is missing in deafened birds, pointing to an important role of auditory feedback for singing-driven FoxP2 down regulation 27 . How does singing affect FoxP2 expression at the cellular level? Medium spiny neurons (MSNs), the most abundant cell type in the avian striatum, predominantly express FoxP2 at low levels (FoxP2 low ) while a subset expresses FoxP2 at very high levels (FoxP2 high ). Both subtypes are not equally affected by singing; the density of FoxP2 high MSNs is not measurably different after singing, contrary to the decreasing density of FoxP2 low MSNs 24 . The authors hypothesized that the difference might be due to the neuronal age. Adult Area X constantly receives new MSNs that originate at the ventricular zone [28][29][30][31][32][33] . FoxP2 high MSNs colocalize more frequently with a marker for new neurons than FoxP2 low MSNs 24 . Recently we showed that new MSNs mature and participate in singing activity -as measured by immediate early gene activation -within a timeframe of six weeks 34 . Whether FoxP2 influences not only the function but also the integration of new neurons into existing circuits is still an open question. Based on the results of Thompson et al. (ref. 21 ) we hypothesized that FoxP2 high MSNs are newly recruited into Area X and need to become FoxP2 low MSNs before they can participate in singing. To test this, we labelled neuronal progenitors in adult zebra finches. At different time points after these cells had migrated into Area X, we quantified their expression levels of FoxP2 and whether they also expressed the immediate early gene expression EGR-1 after singing.
In rodents, Foxp2 expression is associated with neurite outgrowth, neuronal morphology and synapse formation in cortico-striatal circuits 18,19,[35][36][37] . Foxp2 expression levels vary in striatal MNSs and these differences may be relevant for the morphology of striatal MSNs. Dopamine receptor 1 (D1) expressing MSNs 38 express Foxp2 at higher levels than dopamine receptor 2 (D2) expressing MSNs 23,35 . These differences in Foxp2 levels may be linked to anatomical differences between D1 and D2 MSNs 35,39 . Furthermore, in mice carrying humanized Foxp2 alleles (Foxp2 hum/hum mice), Foxp2 high MSNs are more numerous in the dorsal striatum and their MSNs have longer dendrites than wildtype mice 36,37 . Based on the latter results we hypothesized that FoxP2 levels of new MSNs in adult songbirds correlate with their neural morphology. To test this, we virally labelled neural progenitors and analyzed their FoxP2 expression, dendrite complexity and spine density after migration into Area X.

Results
Dynamic FoxP2 levels in new MSNs in Area X. To assess FoxP2 protein levels in individual newborn neurons we labelled progenitor cells with Bromodeoxyuridine (BrdU) and detected BrdU+/FoxP2+ cells after 21, 31 and 42 days post BrdU injection (dpi) in Area X of adult male zebra finches (Fig. 1a,b). We found that FoxP2 expression in Area X was very variable, with some neurons expressing FoxP2 at particularly high levels and some at low levels ( Fig. 1c). At 21 dpi and at 31 dpi, the mean pixel intensities of all BrdU+/FoxP2+ cells formed a bimodal distribution, whereas at 42 dpi the distribution was unimodal and shifted to low FoxP2 expression (Fig. 1d). We classified all neurons that had expression intensities within the top 30% of the measured pixel intensity distribution as FoxP2 high neurons. Neurons within the bottom 30% of the measured pixel intensity distribution were considered as FoxP2 low . Because we were interested in the two extremes of the expression levels in this study, we did not analyze the new neurons with intermediate FoxP2 expression levels further (29.3% ± 9.4, SD, see Methods). At 21 dpi and at 31 dpi 36.17% ± 6.02 (SEM) and 34.91% ± 2.53 (SEM) percent of all BrdU+ cells were FoxP2 high neurons, respectively. At 42 dpi the percentage of FoxP2 high cells had significantly decreased to 12.95% ± 2.87 (SEM) on average (p = 0.0077, Kruskal-Wallis test, Fig. 1f,g). The percentage of FoxP2 low neurons increased significantly from 21 dpi (34.59 ± 5.86, SEM) and 31 dpi (34.06 ± 3.35, SEM) to 42 dpi (59.19 ± 4.02, SEM, p = 0.013, ANOVA, Fig. 1e,g). We also noticed that BrdU+ cells varied in their nucleus size. Quantification revealed that the distribution of the nucleus size was shifted towards bigger nuclei at 31 dpi (data not shown). BrdU+/FoxP2+ cells at 31 dpi had significantly bigger nuclei (7.37 µm ± 0.94 (SD) than BrdU+/FoxP2+ cell at 21 dpi (6.88 µm ± 0.9 (SD), p = 0.00025, chi-square = 75.358, df = 2) or 42 dpi (6.74 µm ± 0.59 (SD), p = 2 × 10 −6 , chi-square = 75.358, df = 2, data not shown). Interestingly there was a significant positive relationship with a low effect size between nucleus size and FoxP2 expression levels in all three experimental groups (21 dpi: r 2 = 0.086, p = 0.017, 31 dpi: r 2 = 0.083, p = 5.2 × 10-7, 42 dpi = r 2 = 0.039, p = 0.0012, Fig. 1h).

Singing induced activity of new MSNs is independent of FoxP2 levels. In adult zebra finches
FoxP2 expression levels in Area X are behaviourally regulated. Undirected singing leads to downregulation of FoxP2 mRNA and protein [24][25][26] . Undirected singing is also associated with expression of the immediate early gene EGR-1 in Area X, which is therefore often used as a molecular readout of the neuronal activity associated with undirected song [40][41][42][43][44] . We hypothesized that FoxP2 levels in new FoxP2 high neurons needed to be downregulated before activation by singing could occur, resulting in EGR-1 expression. Consequently, we did not expect to find BrdU+/FoxP2 high /EGR-1+ MSNs in Area X. To test this, we analyzed BrdU+/EGR-1 ± /FoxP2+ cells in birds that had sung before sacrifice after 21 dpi, 31 dpi and 42 dpi.
Contrary to our hypothesis we found BrdU+/FoxP2 high /EGR-1+ cells in Area X (Fig. 2a) in all groups, with large differences between birds. At 31 dpi, on average 41.45% ± 15.43 (SEM) of new neurons expressed FoxP2 high and were also activated by singing (BrdU+/FoxP2 high /EGR-1+) and 40.28% ± 13.34 (SEM) of the new neurons that expressed FoxP2 high were not activated by singing (BrdU+/FoxP2 high /EGR-1-) (Fig. 2b). At 42 dpi, the more birds had sung the fewer FoxP2 high new neurons were found, resulting in a significant negative relationship between the number of BrdU+/FoxP2 + /EGR-1+ neurons and the number of motifs sung before sacrifice (r 2 = 0.753, p = 0.025, Fig. 2c) which was not the case at 21 dpi (r 2 = 0.168, p = 0.493, data not shown) nor at 31 dpi (r 2 = 0.164, p = 0.425, data not shown).

FoxP2 levels in new MSNs influence dendritic arborization and spine density.
Because of the relationship of FoxP2 levels and nucleus size we wanted to further characterize the morphology of new neurons expressing different FoxP2 levels. We used a lentiviral approach to express Green Fluorescent Protein (GFP) in progenitor cells at their place of birth in the lateral wall of the lateral ventricle (Fig. 3a). After 31 and 42 dpi we found labeled neurons (GFP+) in Area X and the surrounding striatum (Fig. 3b). GFP+ neurons exhibited the typical morphology of medium spiny neurons with small somata and spiny dendrites (Fig. 3c) and expressed FoxP2 (Fig. 3d,e). First, we traced FoxP2 high and FoxP2 low new neurons at 31 dpi (Fig. 4a) and found that FoxP2 high neurons had more primary dendrites (p = 0.021, Mann Whitney test, Fig. 4b), a higher total branch length (p = 0.003, Mann Whitney test, Fig. 4e) and thicker dendrites than FoxP2 low neurons (p = 0.003, t-test, Fig. 4f). Second, we analyzed the extent of dendritic arborization of GFP+/FoxP2+ neurons using a Sholl analysis (see   www.nature.com/scientificreports www.nature.com/scientificreports/ Methods). At 31 dpi FoxP2 high neurons had more intersections at 20 µm distance from the soma (p = 0.024, paired t-test, Fig. 4a,d) and a higher number of maximal intersections (p = 0.019, t-test, Fig. 4c) than FoxP2 low neurons, reflecting more extensive dendritic arborizations in FoxP2 high than FoxP2 low neurons. Second, Third, we used a semi-automated quantification approach to assess the number of dendritic spines in GFP+/FoxP2+ neurons at 31 dpi. FoxP2 high neurons in Area X had more dendritic spines than FoxP2 low neurons (p = 0.034, paired t-test, Fig. 5a-c). Overall, FoxP2 high neurons had more mushroom spines than FoxP2 low neurons (p = 0.0186, Mann Whitney test, Fig. 5d). There was no difference in the number of stubby spines (p = 0.819, Mann Whitney test) or thin spines (p = 0.409, Mann Whitney test) between FoxP2 high and FoxP2 low neurons (Fig. 5d). Because of the difference in mushroom spine number, we assumed that FoxP2 expression levels might influence mushroom spine head size, too. However, quantification revealed no difference in mushroom spine head size between FoxP2 high and FoxP2 low neurons (p = 0.317, Mann Whitney test, Fig. 5e). In a last step we compared spine densities between new MSNs at 31 dpi and 42 dpi. Since at 42 dpi only few new neurons were FoxP2 high we included only FoxP2 low new neurons in this analysis (Fig. 5f). At 42 dpi, the spine density of FoxP2 low new neurons was higher than in FoxP2 low neurons at 31 dpi (p = 2.517 × 10 −4 , t-test, Fig. 5f). This elevated spine density was largely due to an increase in thin spines (p = 8.422 × 10 −4 , Mann Whitney test) and not mushroom spines (p = 0.39, t-test) or stubby spines (p = 0.119, Mann Whitney test).

Discussion
In the present study we investigated the dynamics of FoxP2 expression in adult-born MSNs in the striatal song nucleus Area X of adult male zebra finches. We show that the new MSN strongly expressed FoxP2 at their arrival in Area X from the ventricular zone (VZ) where they were born 21 days prior. During this stage and at intermediate maturation stages (31 days) one third of new MSNs expressed FoxP2 at high levels. At the late maturation stage (42 days) most new MSNs expressed FoxP2 at low levels (Fig. 6). Together with our previous data we conclude that reaching low FoxP2 levels is a sign that adult-born born MSN in Area X have reached maturity by 6 weeks after their generation in the VZ 34 . www.nature.com/scientificreports www.nature.com/scientificreports/ Since our previous work demonstrated that new MSNs participate in singing-associated neural activity in Area X 34 here we asked if this was linked to FoxP2 levels. After singing FoxP2 mRNA and protein levels are lower in Area X tissue 25-27 than when birds were silent whereas the expression of the immediate early gene EGR-1 increases linearly the more birds sing undirected song 40,41 . Analyzing expression levels in individual neurons revealed that only FoxP2 low MSNs, but not Foxp2 high MSNs seemed to be subject to singing induced FoxP2 downregulation 24 . We therefore hypothesized that FoxP2 high MSNs were not yet connected into the circuit and therefore not regulated by singing. Our data contradict this hypothesis. We show that the induction of the immediate early gene EGR-1 in MSNs after singing was equally likely in FoxP2 low and FoxP2 high MSNs, suggesting that both were functionally incorporated into the song circuit.
Previous work showed that the degree of FoxP2 downregulation correlates with the quantity of produced song and depends on auditory feedback, which is relayed to Area X via the cortical song nucleus HVC 27,45 . In our study, the relationship between FoxP2 downregulation and song quantity was present in new MSNs at 42 dpi but not earlier and we therefore suggest that new MSNs start to receive auditory feedback signals between 31 and 42 dpi.
What might cause the age-dependent FoxP2 downregulation in new MSNs? One possibility is that intrinsic mechanisms, depending more on cell age than on extracellular inputs, downregulate FoxP2 during maturation. Another possibility is that EGR-1 gradually decreases FoxP2 levels during every singing event. This latter scenario is consistent with the findings that the FoxP2 promoter contains EGR-1 binding sites 4,46 and that EGR-1 expression is crucial for functional integration of new neurons in the adult rodent hippocampus 47 .
We also tested if varying FoxP2 levels affect neuronal morphology in adult male zebra finches and analyzed dendrite complexity and spine density of virally labelled new FoxP2 low and FoxP2 high MSNs. We now show differences between the morphology of adult generated MSNs that express FoxP2 at high or low levels; high FoxP2 levels were associated with greater dendrite complexity and higher dendritic spine density in comparison to neurons with low FoxP2 levels (Fig. 6). Concerning spine density, our results are consistent with previous studies www.nature.com/scientificreports www.nature.com/scientificreports/ in juvenile zebra finches, since experimental FoxP2 knockdown decreased overall spine densities of new Area X MSNs 48 . Moreover, our results are in line with findings in mice that link Foxp2 to neuronal outgrowth and spine density in striatal neurons and their progenitors 18,19,[35][36][37] . Additionally, we find similarities between mice and birds on the level of spine types. We show that FoxP2 low new MSNs 31 dpi have fewer mushroom spines than FoxP2 high new MSNs. In mice, striatal spiny neurons of heterozygous Foxp2 knockdown mice show specifically a decrease of mushroom and branched spines whereas stubby and thin spines are not affected 19 . In birds and mammals, dendritic spines of striatal MSNs receive both glutamatergic dopaminergic input from cortical/pallial regions and the midbrain, respectively. In the case of new MSNs in Area X, we hypothesize that high FoxP2 expression levels during their maturation might increase their capacity for receiving both inputs.
What mechanisms might account for morphological differences between FoxP2 low and FoxP2 high MSNs? They may originate from differential target gene activation in FoxP2 low and FoxP2 high MSNs. FoxP2 has hundreds of downstream target genes of which many are part of networks associated with neurite development 35,[49][50][51][52] . One specific candidate is the myocyte enhancer factor 2C (Mef2c), a negative regulator of synaptogenesis 19 . Foxp2 specifically promotes corticostriatal synaptogenesis via the repression of Mef2c. Whether a similar mechanism shapes the integration of new MSNs into the avian corticostriatal network remains to be elucidated by future studies. In zebra finch Area X, two direct FoxP2 targets are associated with neuronal outgrowth; the very-low-density-lipoprotein receptor (VLDLR) and the contactin-associated protein-like 2 (CNTNAP2). Their expression correlates positively with FoxP2 in juveniles and in singing adults 53,54 . Thus, in new Area X neurons, VLDLR and CNTNAP2 would be expected to be highly activated during singing in FoxP2 high MSN but not in FoxP2 low MSNs and may thus generate the diverging MSNs morphology we found.
We would like to propose some speculations regarding possible functions of two MSNs subpopulations that differ in FoxP2 expression levels. We found that these populations differed in nucleus size, dendritic complexity and spine density during an early time period of their integration into Area X. We do not know if the observed morphological differences persist long-term because of a lack of markers that could distinguish former FoxP2 high MSNs from former FoxP2 low MSNs in later maturation phases. If these two subpopulations persist long-term, they might resemble striato-nigral and striato-pallidal MSNs of the direct and indirect pathway in mammals. These MSNs subtypes are morphologically and neurochemically different. Direct pathway MSNs express the dopamine receptor D1 and their dendrites are more complex than indirect pathway MSNs that express the dopamine receptor D2 39,55 . High Foxp2 levels in D1 MSNs and low Foxp2 in D2 MSNs have been proposed to be linked to this anatomical dichotomy 35 . The avian direct and indirect pathway through the basal ganglia however is not characterized by different MSN projections but rather by direct and indirect pallidal-like output neurons that project from Area X to the thalamus 56 . To date, it is not known if MSN subtypes exclusively synapse on either direct or indirect pallidal-like neurons 57 . Contrary to mammalian MSNs, more than half of the Area X MSNs express multiple dopamine receptors 58 so that they cannot be used as markers for indirect versus direct pathway neurons. Investigating potential avian MSNs subtypes and the developmental role FoxP2 plays in those will be of interest for future studies.
What might be the function of new MSNs in Area X and how is it affected by FoxP2 expression levels? Our previous study showed that once matured, new MSNs have similar characteristics as older, resident MSNs and are active during singing 34 . General MSNs function is feed forward inhibition within a cortico-striatal-thalamo-cortical loop during singing 59 . We hypothesize that new MSNs in adult zebra finches are entrained to produce a correct firing pattern in a plastic phase during their maturation and thus may counteract www.nature.com/scientificreports www.nature.com/scientificreports/ song drift, as has been suggested before 33,60 . This process might be in influenced by varying FoxP2 expression levels and the resulting morphological differences between putative subpopulations of new MSNs. FoxP2 high new neurons with a higher dendrite complexity and more dendritic spines might be more receptive to tuning than FoxP2 low new neurons. For further interpretations of our findings it will be crucial to gain additional knowledge on the microcircuitry of Area X and on the role neurons play for its function.
In summary, FoxP2 expression levels vary in adult-born MSN at different maturation times after they have been recruited to Area X. We show that the different FoxP2 expression levels correlate with neuronal morphology and spine density. Varying FoxP2 expression levels during a specific time window might permit different target gene activation important for correct incorporation and function of new MSNs in Area X.

Methods
FoxP2 nomenclature. We follow the nomenclature proposed by 61 , FOXP refers to the human gene, Foxp refers to the mouse gene and FoxP refers to all other species. FOXP, Foxp2 and FoxP2 correspond to the protein product.
Animals. 42 adult male zebra finches (Taeniopygia guttata, age >120 days) were used in the present study, bred and housed at the Department of Animal Behaviour at Freie Universität Berlin. The colony was kept under a 12:12 h light:dark-cycle with food and water ad libitum. All experiments were reviewed and approved by the veterinary department of the Freie Universität Berlin and by the ethics committee of the Regional Office for Health and Social Affairs Berlin and were performed in accordance with relevant guidelines and regulations. The permit numbers are G0116/13 and G0296/15.

Experiments.
We conduced three experiments. In the first, we analyzed FoxP2 expression levels of new neurons (BrdU+, see below) in Area X at three time points, e.g. at 21 days (5 birds, 166 neurons), 31 days (6 birds, 295 neurons) or 42 days (6 birds, 272 neurons) after BrdU injections (dpi). In the second experiment we analyzed FoxP2 expression levels and the expression of the early growth response protein 1 (EGR-1) at 21 dpi (6 birds, 127 neurons), 31 dpi (6 birds, 108 neurons) and at 42 dpi (6 birds 156 neurons). In the third experiment we analyzed FoxP2 expression levels, dendrite morphology and spine density of new neurons that were labelled via lentiviral infection. In total we analyzed 52 neurons of 4 zebra finches (13 ± 3 neurons/bird, mean ± SD) at 31 dpi and 23 neurons of 3 zebra finches at 42 dpi (7.6 ± 0.5 neurons/bird, mean ± SD).

BrdU injections. For the analysis of FoxP2 levels and EGR-1 expression in newborn neurons 35 birds
received BrdU (50 µg/g) via intramuscular injections in the mornings for 5 consecutive days. Birds were assigned to three groups with different survival times (21,31, and 42 days after BrdU injection, dpi).

Song monitoring. For FoxP2 expression level analysis after BrdU treatment or lentiviral injections, 17 birds
were isolated in sound attenuated chambers for one night before sacrifice. In the following morning, birds were kept from singing by the experimenter sitting nearby for 1.5 h after lights went on. For EGR-1 expression analysis in new neurons after singing, 18 birds were kept in sound attenuated chambers for three nights and were perfused in the morning of the 4th day 1.5 h after the lights went on. During those 1.5 h birds had to sing at least 150 motifs to be included in the subsequent analysis of EGR-1 expression. Vocalizations were continuously monitored via Sound Analysis Pro 62 .

Confocal imaging and image processing and quantification. Z-Stacks of BrdU+ or GFP+ cells
in Area X were obtained with a SP8 confocal microscope (Leica). For FoxP2 scanning, all microscope settings were kept constant. Scans of BrdU+ nuclei were performed using a 63x lens (digital zoom 2.0), an image size of 1024 × 1024 pixels and a z-stack size of 1 µm. Whole neurons (GFP+) were imaged using a 63x lens, an image size of 2042 × 2042 pixel and a z-stack size of 1 µm. Acquired images were processed using the Fiji software package 64  www.nature.com/scientificreports www.nature.com/scientificreports/ plugin was used to subtract background. We measured the mean pixel intensity of nuclear FoxP2 expression, by positioning a circle of 4 µm in diameter (12.56 µm 2 ) in the center of the BrdU+ nucleus. In total we analyzed the intensity of the FoxP2 expression dependent fluorescence of 166 BrdU+ cells at 21 dpi (n = 5), 295 BrdU+ cells at 31 dpi (n = 6) and 272 BrdU+ cells at 42 dpi (n = 6). FoxP2 high were defined as cells that fell into the top 30% of measured mean pixel intensities in one animal (i.e. if the highest mean pixel intensity in one animal was 240 we counted all BrdU+ cells that had a FoxP2 mean pixel intensity between 168 and 240 as FoxP2 high neurons). We decided on the 30% value because it covered the FoxP2 high expressing cells in the bimodal distribution of all FoxP2 intensities. We defined the neurons that fell into the bottom 30% of measured mean pixel intensities as FoxP2 low . As for the FoxP2 high cutoff, the bottom 30% contained the low-intensity peak of the bimodal density distribution. Because we particularly wanted to address the effect of high and low FoxP2 expression levels on neuronal properties, neurons with intermediate FoxP2 expression levels were not considered for further analysis. The Simple Neurite Tracer plugin (Fiji) was used to trace individual neurons and we analyzed their total branch length and number of primary dendrites. The traces were then used by the Sholl analysis plugin in Fiji 65 . We measured intersections of dendrites with concentric circles that were placed every 10 µm starting from the center of the soma. The maximal number of intersections per neuron was extracted from the Sholl analysis dataset. For dendritic spine analysis images were deconvolved using the Tikhonov-Miller algorithm in the DeconvolutionLab plugin in Fiji 66 . Prior to deconvolution an individual point spread function was generated for each image using the Born and Wolf 3D optical model in the PSF Generator plugin in Fiji 67 . Semiautomated dendritic spine counts were performed using the software NeuronStudio 68 that uses a spine classification algorithm. For spine classification, the default settings were used to classify spines as mushroom, stubby or thin spines: a head-to-neck ratio threshold of 1.1 µm, a height-to-width ratio threshold of 2.5 µm and a minimum mushroom head size of 0.35 µm. A spine is considered mushroom if the head-to-neck ratio is above the threshold and its head is larger than 0.35 µm. A spine is considered stubby if its head-to-neck ratio and its heights-to-width ratio are below threshold. In all other cases spines were classified as thin. On average, we analyzed spines densities on secondary dendrites along the length of 118 µm ± 1.92 (mean, SEM) per neuron. In total, we analyzed spines of 52 individual neurons of 4 animals in experimental group 31 dpi, and 23 neurons of 3 animals at 42 dpi. Additionally, we measured the dendrite diameter of 44 new neurons in 4 animals (8-12 neurons per animal) using the line measuring tool in Fiji 64 . We took 5 measures on each of 3 secondary dendrites per neuron (in total 15 measures per neuron). The experimenter was blind to FoxP2 levels of individual neurons during the whole quantification process, because cells were selected for quantification based on their BrdU+ fluorescence or their EGR-1 fluorescence and FoxP2 fluorescence in a different channel was quantified last. The datasets generated and analysed during the current study are available from the corresponding author on request.
Statistics. The software R was used to analyze data 69 . Significance level was p < 0.05 for all tests. Plots were generated using the ggplot package in R 70 . For the analysis of FoxP2 high neurons we used a Kruskal-Wallis test followed by a Dunn's test for pairwise comparison. For the analysis of FoxP2 low neurons we used ANOVA followed by a Bonferroni's multiple comparison test. The relationship of (a) FoxP2 levels and nuclear diameter as well as (b) FoxP2+ neurons and singing were determined using a linear regression analysis. Dendritic spine data (all spines) and Sholl data were analyzed using the paired Student's t-test. Data from the spine type analysis, mushroom head size, total dendrite length, number of primary dendrites and dendrite thickness were analyzed using the Mann Whitney test. Number of maximal intersections was analyzed using the Student's t-test. Choice of test was based on previous analysis for normality using the Shapiro-Wilk-test and variance analysis using the F-test or Levene's test.