FOXP transcription factors in vertebrate brain development, function, and disorders

Abstract

FOXP transcription factors are an evolutionarily ancient protein subfamily coordinating the development of several organ systems in the vertebrate body. Association of their genes with neurodevelopmental disorders has sparked particular interest in their expression patterns and functions in the brain. Here, FOXP1, FOXP2, and FOXP4 are expressed in distinct cell type-specific spatiotemporal patterns in multiple regions, including the cortex, hippocampus, amygdala, basal ganglia, thalamus, and cerebellum. These varied sites and timepoints of expression have complicated efforts to link FOXP1 and FOXP2 mutations to their respective developmental disorders, the former affecting global neural functions and the latter specifically affecting speech and language. However, the use of animal models, particularly those with brain region- and cell type-specific manipulations, has greatly advanced our understanding of how FOXP expression patterns could underlie disorder-related phenotypes. While many questions remain regarding FOXP expression and function in the brain, studies to date have illuminated the roles of these transcription factors in vertebrate brain development and have greatly informed our understanding of human development and disorders.

Graphical Abstract

FOXP transcription factors have distinct expression patterns in the brain, and manipulations in animal models are disentangling how their spatiotemporal expression is linked to neurodevelopmental disorder etiology.

Introduction

Note: In this review, we use the following nomenclature: for human, FOXP1 (in italics) for the gene and FOXP1 for the protein; for rodent, Foxp1 (in italics) for the gene and FOXP1 for the protein; for zebra finch, FoxP1 (in italics) for the gene and FOXP1 for the protein; for other or combined species, FOXP1 (in italics) for the gene and FOXP1 for the protein.

The forkhead box P (FOXP) protein subfamily belongs to the FOX family of transcription factors which coordinate essential developmental processes within various organ systems, including pulmonary, cardiac, nervous, and immune systems (Lu, Li, Yang, & Morrisey, 2002; Shu, Yang, Zhang, Lu, & Morrisey, 2001). FOXP1, FOXP2, and FOXP4 are highly expressed in the nervous system and regulate molecular pathways required for proper brain development and function (Ferland, Cherry, Preware, Morrisey, & Walsh, 2003; Takahashi, Liu, Hirokawa, & Takahashi, 2008). FOXP1 and FOXP2, in particular, are strongly linked to human neurodevelopmental disorders (NDDs), including autism spectrum disorder (ASD), intellectual disability (ID), and speech and language disorder (Lai, Fisher, Hurst, Vargha-Khadem, & Monaco, 2001; Meerschaut et al., 2017; Siper et al., 2017). Multiple brain regions are implicated in the etiology of these disorders, and more recent studies are homing in on specific circuits and cell types underlying vulnerability to NDDs (Chang, Gilman, Chiang, Sanders, & Vitkup, 2015; Xu, Wells, O'Brien, Nehorai, & Dougherty, 2014). Because FOXP transcription factors show distinct spatiotemporal expression patterns within the developing brain, understanding their molecular functions within their respective cell types can lend important insights into neural circuits disrupted in NDDs.

Vertebrate FOXP DNA binding domains (DBDs), named winged helix or forkhead after the Drosophila gene fork head, are highly conserved with that of Drosophila FOXP, indicating an evolutionarily ancient origin for this protein subfamily (Shu et al., 2001). Additional shared structural features of FOXP1/2/4 include a polyglutamine tract, a zinc finger domain, and a leucine zipper domain (Figure 1). Unlike other FOX transcription factors, FOXP subfamily members can homo- and heterodimerize through their leucine zippers to bind DNA at a consensus forkhead sequence (S. Li, Weidenfeld, & Morrisey, 2004; Stroud et al., 2006; Wang, Lin, Li, & Tucker, 2003). In vivo evidence supports these FOXP interactions, including coimmunoprecipitation of these proteins from zebra finch brain (Mendoza & Scharff, 2017) and the similarity in gene dysregulation between Foxp1 and Foxp2 mutant mouse brains (Araujo et al., 2015). Moreover, in vitro assays suggest that the various combinations of FOXP dimers may differentially regulate downstream target genes (Mendoza & Scharff, 2017; Sin, Li, & Crawford, 2015). In addition to these intra-subfamily interactions, FOXP proteins can associate directly or indirectly with numerous other transcriptional regulators, as indicated by coimmunoprecipitation, bioluminescence resonance energy transfer, and proximity ligation assays; these interactors include other transcription factors (CTBP1/2, NFIA/B, NKX2-1, NR2F1/2, POT1, SATB1/2, SOX5, TBR1, YY1, ZMYM2) and components of the nucleosome remodeling and deacetylase complex (Chokas et al., 2010; Deriziotis et al., 2014; Estruch et al., 2018; Hickey, Berto, & Konopka, 2019; S. Li, Weidenfeld, et al., 2004; Sakai et al., 2011; Tanabe, Fujita, & Momoi, 2011; Zhou et al., 2008). In addition to these diverse FOXP protein-protein interactions, differential splicing of FOXP transcripts may provide further molecular regulation of neuronal development and function, as demonstrated by studies of a FOXP2 isoform lacking the DBD (Bruce & Margolis, 2002; Burkett et al., 2018; Teramitsu & White, 2006; Vernes et al., 2006). Altogether these studies highlight a need to dissect shared and distinct FOXP expression patterns in the brain to understand their coordination of brain development.

Figure 1. Shared structural domains of human FOXP proteins.

Shared protein domains of human FOXP1, FOXP2, and FOXP4 are shown using information from UniProtKB. FH: forkhead, LZ: leucine zipper, polyQ: polyglutamine tract, ZF: zinc finger.

In this review, we provide a detailed overview of the known spatiotemporal expression patterns of FOXP genes across vertebrate species. We then describe the distinct NDDs caused by mutations in FOXP1 and FOXP2, and we point to evidence suggesting FOXP4 as a candidate gene for NDD. Finally, we summarize animal models of FOXP function in the brain, including brain region- and cell type-specific gene manipulations that relate their distinct expression patterns to NDD-relevant phenotypes.

FOXP EXPRESSION PATTERNS IN THE BRAIN

In the developing and adult mouse, FOXP proteins are expressed in multiple tissues throughout the body, including heart, lung, kidney, spleen, spinal cord, and brain (Lu et al., 2002; Morikawa, Hisaoka, & Senba, 2009; Shu et al., 2001; Tamura, Morikawa, Iwanishi, Hisaoka, & Senba, 2003; Zhao et al., 2015). Given the focus of this review on neurodevelopmental roles of these transcription factors, we will mainly detail their spatiotemporal expression patterns within the brain.

Expression in the cortex

The mammalian neocortex (referred to as ‘cortex’ in this review) is a dorsal telencephalon-derived brain structure arising from the ventricular zone (VZ), where radial glia divide to eventually produce postmitotic neurons of the cortical plate (CP) and later the mature six-layered cortex (Greig et al., 2013). In mice, faint mRNA signals for Foxp1/2 appear in the VZ by embryonic day (E) 12.5 (Ferland et al., 2003), but studies conflict over their protein expression here, with some reporting expression (Braccioli et al., 2017; Pearson et al., 2018; Tsui, Vessey, Tomita, Kaplan, & Miller, 2013) and others reporting absence (Ferland et al., 2003; Kast, Lanjewar, Smith, & Levitt, 2019). By E16.5, both are expressed in the CP at the protein level (Ferland et al., 2003). In the developing rat cortex, Foxp4 mRNA appears by E14 in the medial VZ/CP and lateral CP and is broadly expressed in early postnatal cortex (Takahashi, Liu, Hirokawa, et al., 2008). While rat Foxp4 mRNA diminishes postnatally, FOXP1 and FOXP2 are maintained in projection neuron subpopulations in the mature mouse cortex (Ferland et al., 2003; Hisaoka, Nakamura, Senba, & Morikawa, 2010; Takahashi, Liu, Hirokawa, et al., 2008). FOXP1 is mainly expressed in intratelencephalic projection neurons (ITPNs) in layers (L) 3-5 and in some corticothalamic projection neurons (CThPNs) in L6a (Hisaoka et al., 2010). In contrast, FOXP2 is highly enriched in L6 CThPNs as well as in a subpopulation of L5 pyramidal tract neurons (Campbell, Reep, Stoll, Ophir, & Phelps, 2009; Hisaoka et al., 2010; Kast et al., 2019; Sorensen et al., 2015). Early postnatal mouse cortex also contains a small population of FOXP2+ L6 corticocortical neurons that diminishes over development (Kast et al., 2019). Interestingly, coexpression of FOXP1/2 in L6a increases over postnatal development, the functional significance of which remains unknown (Hisaoka et al., 2010). These cortical FOXP expression patterns are reflected at the transcriptomic level in a recent single-cell RNA-seq study from adult mouse cortex (Saunders et al., 2018) (Figure 2A). These data confirm upper- and lower-layer expression of Foxp1 and enrichment of Foxp2 in L6, while also showing broad low-level expression of Foxp4 transcript in adult mouse brain.

Figure 2. Cell-type specific expression of FOXP transcription factors in forebrain regions.

(A) Visualization of adult mouse single-cell RNA-seq data from (Saunders et al., 2018). Visualization of Foxp1, Foxp2, and Foxp4 expression within the frontal and posterior cortex, striatum, and hippocampus using the DropViz tool (http://dropviz.org). Expression is shown in black with enriched clusters shaded in color. Endo: endothelial cells, L: layer, SPN: spiny projection neuron, DG: dentate gyrus, Sub: subiculum, Entor: entorhinal cortex. (B) Adult human cortical scRNA-seq data showing the expression of FOXP1, FOXP2, and FOXP4 within the medial temporal gyrus from the Allen Brain Atlas tool (https://celltypes.brain-map.org/rnaseq/human). Red indicates high expression, black indicates no expression. Ex: excitatory neurons, Int: interneurons, L: layer.

Additional studies have examined cortical expression patterns of FOXP1/2 in other vertebrates. The enrichment of FOXP1 in upper layers and FOXP2 in lower layers is generally conserved across vocal learning bats, monkeys, and humans, but vocal learning bats show additional L2-4 FOXP2 expression (Rodenas-Cuadrado et al., 2018; Takahashi, Liu, Oishi, et al., 2008; Teramitsu, Kudo, London, Geschwind, & White, 2004) (Figure 2B). In zebra finch pallial regions homologous to the mammalian cortex, FOXP1 is highly expressed in the hyper-, meso-, and nidopallium as well as the song nuclei HVC (proper name) and robustus arcopallii (RA), while FOXP2 is expressed at very low levels in the hyper-, meso-, and nidopallium and the nucleus lateralis magnocellularis nidopallii anterioris (lMAN) (Haesler et al., 2004; Mendoza et al., 2015; Teramitsu et al., 2004). In the adult zebra finch, FOXP4 is broadly expressed throughout the pallium, contrasting with the transient expression of Foxp4 in rat cortex (Mendoza et al., 2015).

In summary, FOXP1 and FOXP2 show mostly complementary cell type-specific expression patterns in the vertebrate cortex, with enrichment of FOXP1 in ITPNs and FOXP2 in CThPNs, while FOXP4 shows broad cortical expression that is transient in mammals.

Expression in the hippocampal formation

The hippocampal formation is a limbic system structure arising from the medial telencephalon and includes the dentate gyrus, subiculum, and hippocampus, which is further subdivided into CA1-4 subfields. In the rat hippocampus, Foxp4 is expressed at E14 while Foxp1 appears later by E20 (Takahashi, Liu, Hirokawa, et al., 2008). At this later stage in rodent development, Foxp4 is broadly expressed in CA1-4 and the subiculum, while FOXP1 is more limited to CA1 and the subiculum (Ferland et al., 2003; Takahashi, Liu, Hirokawa, et al., 2008). As in the cortex, hippocampal Foxp4 expression decreases postnatally while FOXP1 is maintained into adulthood (Ferland et al., 2003; Takahashi, Liu, Hirokawa, & Takahashi, 2003; Takahashi, Liu, Hirokawa, et al., 2008). Foxp1 mRNA also appears in the adult mouse CA2/3 and DG (Saunders et al., 2018) (Figure 2A). In the rodent hippocampal formation, Foxp2 is not detected at high levels at any stage (Ferland et al., 2003; Saunders et al., 2018; Takahashi et al., 2003; Takahashi, Liu, Hirokawa, et al., 2008) (Figure 2A).

Other vertebrates show varying degrees of hippocampal FOXP1/2 expression. While vocal learning bats and monkeys show conserved FOXP1 expression in CA1 and subiculum, postnatal monkeys show additional FOXP1 expression in CA3, and one vocal learning bat species shows abundant FOXP2 in CA1 (Rodenas-Cuadrado et al., 2018; Takahashi, Liu, Oishi, et al., 2008). Reports of human hippocampal FOXP2 expression are mixed, with one study failing to detect mRNA by in situ hybridization and another reporting detection by microarray (Lai, Gerrelli, Monaco, Fisher, & Copp, 2003; Wilcke et al., 2011). Curiously, in the adult zebra finch hippocampal formation, neither FOXP1 nor FOXP2 appear to be expressed, but FOXP4 is expressed at moderate levels (Mendoza et al., 2015).

In summary, mammals generally show abundant FOXP1 expression in CA1 and subiculum, variable FOXP2 expression, and transient but broad FOXP4 expression in the hippocampal formation. In the adult zebra finch, only FOXP4 is detectably expressed in this brain region.

Expression in the amygdala

The amygdala is a limbic system structure derived from the telencephalon and composed of multiple nuclei, and the extended amygdala encompasses portions of the amygdala as well as the bed nucleus of the stria terminalis (BST). In postnatal rodent main and extended amygdala, Foxp genes show distinct expression patterns: FOXP1 is sparsely expressed; FOXP2 is expressed highly in the intercalated nucleus (ITC) and the anterior nucleus of the medial BST, moderately in the basomedial and medial amygdalar nuclei, and sparsely in the basolateral nucleus and ventral nucleus of the lateral BST; Foxp4 mRNA is moderately expressed in select cells of the basolateral nucleus (Campbell et al., 2009; Ferland et al., 2003; Kaoru et al., 2010; Takahashi, Liu, Hirokawa, et al., 2008). In the mature rodent amygdala, FOXP1/2 is maintained whereas Foxp4 diminishes (Campbell et al., 2009; Ferland et al., 2003; Takahashi, Liu, Hirokawa, et al., 2008).

As for other vertebrates, FOXP2 protein expression has been described in the zebra finch extended amygdala (Vicario, Mendoza, Abellan, Scharff, & Medina, 2017). In the central extended amygdala, FOXP2 is abundant in ITC-like and peri-intrapeduncular nuclear areas, and sparsely expressed in the capsular central amygdala and the oval central nucleus. In the medial extended amygdala, FOXP2 is found in specific subdivisions of the medial amygdala and medial BST (Vicario et al., 2017). In monkeys, FOXP2 but not FOXP1 mRNA is expressed in the ITC (Kaoru et al., 2010).

In summary, vertebrates show well-conserved FOXP2 expression in the amygdalar ITC. While rodents show mostly non-overlapping expression of Foxp1/2/4 in the amygdala, FOXP1/4 expression patterns in other vertebrates remain poorly described.

Expression in the basal ganglia

The basal ganglia are a diverse group of nuclei derived from multiple embryonic origins. In the forebrain, the medial and lateral ganglionic eminences (MGE and LGE) and the preoptic domain give rise to the striatum, globus pallidus, ventral pallidum, and subthalamic nucleus. In rodents, Foxp1/2/4 appear in the LGE around E12-E14, and FOXP1/2 remain expressed in the adult striatum while Foxp4 diminishes postnatally (Ferland et al., 2003; Lai et al., 2003; Precious et al., 2016; Saunders et al., 2018; Takahashi et al., 2003; Takahashi, Liu, Hirokawa, et al., 2008) (Figure 2A). The striatum is organized by neurochemical zones called striosome and matrix, and in postnatal rodents, Foxp1 is equally expressed in both, Foxp2 is enriched in striosomes, and Foxp4 is enriched in matrix (Takahashi et al., 2003; Takahashi, Liu, Hirokawa, et al., 2008). Monkeys also show these striosome-matrix expression patterns of FOXP1/2 (Takahashi, Liu, Oishi, et al., 2008). Striatal neurons are also organized by projection pathways, and mouse FOXP1 is equally highly expressed in direct and indirect pathway spiny projection neurons (dSPNs and iSPNs), while FOXP2 is expressed at high levels in dSPNs and low levels in iSPNs (Fong, Kuo, Wu, Chen, & Liu, 2018; Saunders et al., 2018; Vernes et al., 2011) (Figure 2A). Both are also expressed in a newly described distinct subpopulation of “eccentric” SPNs (eSPNs), whose functional role in the striatum remain to be elucidated (Saunders et al., 2018) (Figure 2A). Within the ventral striatum of rodents and monkeys, FOXP1 is expressed in both the nucleus accumbens shell and core, whereas some authors have detected FOXP2 more prominently in the shell than in the core (Campbell et al., 2009; Takahashi et al., 2003; Takahashi, Liu, Oishi, et al., 2008). FOXP1/2 are also highly expressed in the developing human striatum (Lai et al., 2003; Teramitsu et al., 2004). In contrast with the high striatal expression of these genes, FOXP1 is absent and FOXP2 is lowly expressed in the globus pallidus of mice and humans (Ferland et al., 2003; Lai et al., 2003; Teramitsu et al., 2004). The basal ganglia also include a dopaminergic midbrain-derived component called the substantia nigra, where in rodents, Foxp1/2 but not Foxp4 are expressed during development and in adulthood, with stronger expression of Foxp2 than Foxp1 (Ferland et al., 2003; Lai et al., 2003; Takahashi, Liu, Hirokawa, et al., 2008).

Developing and adult zebra finches show similar FoxP1/2 expression patterns in the basal ganglia, with high expression of both in the striatum, no FoxP1 and little FoxP2 in the globus pallidus, and enriched FoxP2 in the substantia nigra (Haesler et al., 2004; Teramitsu et al., 2004). In contrast with rodents, FOXP4 expression persists in the striatum of adult zebra finches (Mendoza et al., 2015). In the striatal song nucleus Area X, FOXP1 is enriched compared with the surrounding striatum throughout the lifespan, whereas FOXP2/4 are enriched in Area X during the sensorimotor song learning period (Haesler et al., 2004; Mendoza et al., 2015; Teramitsu et al., 2004). FOXP2 is also decreased in Area X during adult undirected singing (J. E. Miller et al., 2008; Teramitsu & White, 2006). In both juvenile and adult Area X, a subset of DARPP-32+ SPNs expresses FOXP2, similar to the enrichment of FOXP2 in the dSPN subset of mouse striatal neurons (Adam, Mendoza, Kobalz, Wohlgemuth, & Scharff, 2016; Rochefort, He, Scotto-Lomassese, & Scharff, 2007).

In summary, vertebrates show conserved FOXP1/2 expression patterns within the basal ganglia: broad FOXP1 and restricted FOXP2 within the striatum, no FOXP1 and low FOXP2 within the globus pallidus, and enriched FOXP2 within the substantia nigra. In contrast, rodent Foxp4 is limited to developing striatum while zebra finch FOXP4 is expressed into adulthood.

Expression in the thalamus

The thalamus is a diencephalon-derived subcortical structure with a complex array of nuclei. In rodents, FOXP1 is weakly expressed in early thalamic development but later intensifies in a few nuclei, such as the paraventricular and various posterior nuclei (Ferland et al., 2003; Takahashi, Liu, Hirokawa, et al., 2008). In contrast, as early as E12, mouse FOXP2 shows graded expression in the posterior thalamic primordium, and by E16 this gradient is lost but FOXP2 remains in several thalamic nuclei (Ebisu, Iwai-Takekoshi, Fujita-Jimbo, Momoi, & Kawasaki, 2016). These include paraventricular, lateral posterior, habenular, medial and lateral geniculate, and some dorsal nuclei (Ferland et al., 2003; Lai et al., 2003). In the rat, Foxp4 mRNA is expressed in the developing dorsal and ventral thalamus, and notably, its expression in the dorsal thalamus complements that of Foxp2, with Foxp4 higher in proliferating cells than differentiating cells and vice-versa for Foxp2 (Takahashi, Liu, Hirokawa, et al., 2008). However, this Foxp4 signal diminishes by late embryonic development (Takahashi, Liu, Hirokawa, et al., 2008). In perinatal monkeys, FOXP1/2 are expressed in centromedial and -lateral, ventral posteromedial and -lateral, and lateral geniculate nuclei, and FOXP2 is additionally expressed in mediodorsal and medial habenular nuclei (Takahashi, Liu, Hirokawa, et al., 2008). Similarly, in human and zebra finch, FOXP1 and FOXP2 are both expressed in the thalamus, but FOXP2 is expressed in additional sensorimotor nuclei and more strongly expressed overall (Lai et al., 2003; Teramitsu et al., 2004).

Expression in the cerebellum

The cerebellum is a major hindbrain structure consisting of a tightly folded cortex and deep cerebellar nuclei (DCN) embedded in the white matter underneath. In mice, FOXP1 is highly expressed in the DCN and its mRNA has recently been described in a subset of Purkinje cells (PCs) of the cerebellar cortex at E13.5 (Ferland et al., 2003; Wizeman, Guo, Wilion, & Li, 2019). Foxp2 mRNA is also expressed in this subset of PCs at E13.5, and FOXP2 protein is expressed in all PCs and a DCN subset by E17.5 as well as a subset of PCs and DCN neurons by adulthood (Ferland et al., 2003; H. Fujita & Sugihara, 2012; Lai et al., 2003; Wizeman et al., 2019). In adult mice, there are FOXP2-positive, FOXP2-negative, and mixed FOXP2-expressing PC regions following a transverse striping pattern along the cerebellum (H. Fujita & Sugihara, 2012). FOXP4 is expressed in the cerebellar primordium at E12.5 and is maintained in all PCs in adulthood (Tam, Leung, Tong, & Kwan, 2011). Zebra finch cerebellum shows the same pattern of FoxP1 absence and FoxP2 presence in PCs, but FoxP4 expression in this brain region has not yet been described (Haesler et al., 2004; Teramitsu et al., 2004). In human fetal brain, FOXP2 is expressed in the alar plate of the cerebellar primordium and later in the PC-containing piriform layer of the cerebellum (Lai et al., 2003). While not located within the cerebellum, the inferior olive of the brainstem provides excitatory synaptic inputs onto PC dendrites and shows high FOXP2 expression across vertebrates (Ferland et al., 2003; H. Fujita & Sugihara, 2012; Haesler et al., 2004; Lai et al., 2003).

Expression in other neural regions

FOXP genes are expressed in other sites throughout the developing and adult central nervous system. In rodent olfactory regions, FOXP1/2 are expressed in the anterior olfactory nucleus and olfactory tubercle, while FOXP2/4 are expressed in migratory zone interneurons of the developing olfactory bulb (Ferland et al., 2003; Takahashi, Liu, Hirokawa, et al., 2008). Regions expressing both FOXP1/2 in the mouse brain also include the hypothalamus and superior colliculus; additional FOXP1 sites include the pontine nuclei and septum, while FOXP2 is enriched in the septal nucleus, inferior colliculus, lateral lemniscus nucleus, zona incerta, and ventral tegmental area (Campbell et al., 2009; Ferland et al., 2003; Lai et al., 2003). FOXP2-expressing regions described in human fetal brain include the medullary raphe, medulla oblongata, hypothalamus, and red nucleus (Lai et al., 2003; Teramitsu et al., 2004). Zebra finch FoxP2 is expressed in similar regions as mammalian FoxP2, such as the hypothalamus and ventral tegmental area, and is generally more enriched in midbrain and hindbrain nuclei than FoxP1 (Haesler et al., 2004; Teramitsu et al., 2004).

While the main focus of our review is on FOXP expression and function in the brain, it is important to note that these proteins are also highly expressed in the developing and mature spinal cord (Morikawa, Hisaoka, et al., 2009; Morikawa, Komori, Hisaoka, & Senba, 2009; Tamura et al., 2003) and have been shown in animal models to coordinate spinal motor neuron migration, differentiation, and connectivity (Dasen, De Camilli, Wang, Tucker, & Jessell, 2008; Palmesino et al., 2010; Rousso, Gaber, Wellik, Morrisey, & Novitch, 2008; Rousso et al., 2012). Thus, deficits in FOXP function within both the brain and spinal cord could jointly contribute to speech and motor phenotypes seen in FOXP-related NDDs.

Expression over human brain development

The BrainSpan Atlas was developed by a multi-site consortium to provide gene expression data over the full course of human brain development (M. Li et al., 2018; J. A. Miller et al., 2014), making it a valuable tool for examining FOXP expression in the human brain (Figure 3). Several noteworthy points arise from these data. FOXP expression peaks in the prenatal period compared to postnatal, consistent with the roles of their proteins as transcription factors orchestrating neurodevelopment. Expression of one or more FOXP is postnatally maintained in some brain regions, indicating potential roles in neuronal maintenance or function. Furthermore, FOXP transcript levels vary spatially and over development. Given that FOXP1/2 are particularly gene-dosage sensitive, as shown by their haploinsufficiency in human disorders (Lai et al., 2003; Meerschaut et al., 2017; Siper et al., 2017), these spatiotemporal differences in expression levels could suggest region-specific roles for each gene, as well as differential vulnerability of brain regions to their haploinsufficiency. One of the few brain regions where FOXP1/2/4 have shared, high expression is the developing striatum, a critical region for both fine and gross motor control that is active during human speech and language tasks (Liegeois et al., 2003). The full extent of cellular subpopulations expressing each gene, and to what degrees, in the human brain remain to be fully elucidated. Continued studies using single-cell sequencing techniques in human brain tissue and brain organoids (Velasco et al., 2019) will build upon these important findings.

Figure 3. Expression patterns of FOXPs over human brain development.

Using RNA-seq data from the BrainSpan Atlas of the Developing Human Brain (Li et al., 2018; Miller et al., 2014), log-transformed reads per kilobase of transcript per million mapped reads (RPKM) values were plotted at each timepoint for FOXP1, FOXP2, and FOXP4 within distinct brain regions. Dotted line indicates last prenatal timepoint.

Relevance to motor learning-related circuitry

Human lesion studies and experiments in mice and songbirds have elucidated the conserved neural pathways underlying vocalization and other forms of motor learning. While mice differ from humans and avian vocal learners in that their vocalizations are largely innate (Hammerschmidt et al., 2012; Mahrt, Perkel, Tong, Rubel, & Portfors, 2013), they share similar vocal production and motor learning circuits with these species, as well as some limited ability for vocal modification (Arriaga, Zhou, & Jarvis, 2012). Mammalian and avian brains differ in their structural organization, yet their homologous vocalization circuits all involve connections between the cortex/pallium, basal ganglia, thalamus, and cerebellum, regions with high FOXP1/2 expression (Ferland et al., 2003; Konopka & Roberts, 2016) (Figure 4). Considering the tendency for FOXP1/2 to be expressed in projection neurons of these brain regions, these genes are positioned to play a vital role in the development and function of these long-range circuits to enable speech and other forms of sensorimotor learning.

Figure 4. FOXP1 and FOXP2 expression patterns in mammalian and zebra finch brain.

Expression patterns of FOXP1 and FOXP2 in mammalian brain (left) and zebra finch brain (right) with major motor- and vocalization-related circuits highlighted. CB: cerebellum, CTX: cortex, DLM: medial nucleus of dorsolateral thalamus, DM: dorsomedial nucleus of midbrain, DTZ: dorsal thalamic zone, HP: hippocampus, lMAN: lateral magnocellular nucleus of anterior nidopallium, M: mesopallium, nXIIts: tracheosyringeal subdivision of the hypoglossal nucleus, RA: arcopallial vocal nucleus, STR: striatum, THAL: thalamus.

Altogether these expression studies have revealed key regions and developmental stages for FOXP actions in the brain. One important takeaway is the distinct nature of their expression patterns—FOXP4 is broad yet transient, while FOXP1/2 are maintained into adulthood in mostly discrete neuronal subpopulations. Still, some neuronal subtypes in the mouse brain, such as L6 CThPNs, dSPNs, and developing PCs, express multiple Foxp genes. As neuronal subtype identification continues with the aid of single-cell RNA-seq, FOXP coexpression patterns will become more refined, particularly in mid- and hindbrain regions where their cellular overlap remains unclear. Furthermore, cell type-specific splice variants of FOXP transcripts will also be identified to understand this additional mode of regulating neural development. Finally, protein expression, protein-protein interaction, and protein-DNA interaction experiments will then be needed to confirm FOXP protein coexpression in these cell types and assess combinatorial gene regulation by FOXP homo- and heterotypic interactions.

ASSOCIATION OF FOXP GENES WITH NEURODEVELOPMENTAL DISORDERS

FOXP2 mutations predominantly impair speech and language

FOXP2 was the first gene associated with speech and language after clinicians recognized an autosomal dominant inheritance pattern of a speech disorder within the large multigenerational ‘KE’ family (Lai et al., 2001). Affected family members carried a point mutation at a highly conserved residue (R553H) within the forkhead DBD, which impairs subcellular localization and DNA-binding ability of FOXP2 and may act as a dominant negative (Lai et al., 2001; Vernes et al., 2006). Core features of their disorder included childhood apraxia of speech and orofacial dyspraxia; in other words, they showed difficulty performing sequential orofacial movements, both linguistic (e.g. consonant clusters in “spoon” or “blue”) and nonlinguistic (e.g. “close the lips, then open the mouth, then protrude the tongue”) (Alcock, Passingham, Watkins, & Vargha-Khadem, 2000; Hurst, Baraitser, Auger, Graham, & Norell, 1990; Vargha-Khadem, Watkins, Alcock, Fletcher, & Passingham, 1995; Watkins, Dronkers, & Vargha-Khadem, 2002). Affected members also struggled with aspects of grammar, such as pluralization, derivational morphology, and tense production (Gopnik & Crago, 1991; Vargha-Khadem et al., 1995; Watkins, Dronkers, et al., 2002). They additionally showed impairment of the phonological loop, a working memory component specific to speech-based information (Schulze, Vargha-Khadem, & Mishkin, 2017). Further studies of patients with speech and language disorders have identified more than a dozen other pathogenic variants within FOXP2, solidifying its role in proper development of these capabilities (Morgan, Fisher, Scheffer, & Hildebrand, 2017). It is important to note that these known FOXP2 mutations are heterozygous and generally predicted to be loss-of-function changes, although dominant negative effects of the mutant proteins may also contribute (Mizutani et al., 2007; Tsui et al., 2013; Vernes et al., 2006).

The discovery of a major role for FOXP2 in speech and language prompted investigations into brain regions affected by its mutation and thus critical for these abilities. A structural magnetic resonance imaging (MRI) dataset was collected from both affected and unaffected KE family members and analyzed using several approaches, some involving additional control subjects. Across two or more analyses, affected KE members showed unilateral or bilateral gray matter decreases in cortex (inferior frontal gyrus, precentral gyrus, supplementary motor area, temporal pole), caudate nucleus, and cerebellum (lobules VIIa Crus I and VIIb-VIIIb); they also showed gray matter increases in other cortical areas (Argyropoulos et al., 2018; Belton, Salmond, Watkins, Vargha-Khadem, & Gadian, 2003; Vargha-Khadem et al., 1998; Watkins, Vargha-Khadem, et al., 2002) (Figure 5A). Beyond the KE family, a patient (A-II) with a de novo intragenic FOXP2 deletion showed volume reductions in the globus pallidus, caudate nucleus, thalamus, and hippocampus compared with 26 controls (Liegeois et al., 2016). The authors of this study noted that the inability to detect cortical or cerebellar changes was likely a technical artefact of the single-case study design (Liegeois et al., 2016).

Figure 5. Structural and functional brain abnormalities in affected KE family members.

(A) Brain regions containing sites of altered gray matter in KE family members affected by FOXP2 mutation. Regions altered in two or more structural MRI studies are shown. (B) Brain regions containing sites of altered neural activation in PET or fMRI studies of affected KE family members.

Functional brain imaging has also offered insights into how FOXP2 mutation affects brain activity during language tasks. Positron emission tomography (PET) and functional MRI (fMRI) studies of the KE family have shown differential activation of cortical, basal ganglia, and cerebellar regions during word repetition, non-word repetition, silent verb generation, and spoken verb generation (Argyropoulos et al., 2018; Liegeois et al., 2003; Liegeois, Morgan, Connelly, & Vargha-Khadem, 2011; Vargha-Khadem et al., 1998) (Figure 5B). Many under- or overactive brain regions in affected KE members also showed gray matter alterations in the structural MRI studies, including the inferior frontal gyrus, sensory and motor cortices, temporal gyri and pole, caudate nucleus, and cerebellar lobule VIIa Crus I (Argyropoulos et al., 2018; Belton et al., 2003; Vargha-Khadem et al., 1998; Watkins, Vargha-Khadem, et al., 2002). In addition, fMRI of patient A-II during non-word repetition showed underactivation of the inferior frontal gyrus, but no differences between A-II and controls survived multiple comparisons correction (Liegeois et al., 2016).

Altogether these human brain imaging studies indicate that FOXP2 mutation most prominently alters the structure and function of the cortex, basal ganglia, and cerebellum, which are major sites of FOXP2 expression in humans and other vertebrates. Future imaging analyses should include connectivity measures such as diffusion tenor imaging, especially given the role of Foxp2 in neurite outgrowth in animal models as discussed later in this review. Such studies would indicate whether the speech and language deficits found in FOXP2 patients can be attributed to miswiring of neural circuits during development.

Additional studies have investigated FOXP2 in the etiology of other NDDs involving language and communication deficits. Early studies with small cohort sizes directly queried the role of FOXP2 variants in ASD, and some found association while others did not (Chien et al., 2011; Gauthier et al., 2003; Gong et al., 2004; Laroche et al., 2008; H. Li, Yamagata, Mori, & Momoi, 2005; Marui et al., 2005; Newbury et al., 2002; Richler, Reichert, Buxbaum, & McInnes, 2006; Toma et al., 2013; Wassink et al., 2002). A larger study involving over 2500 parent-proband trios identified FOXP2 among ASD risk loci in a homozygous haplotype mapping approach (Casey et al., 2012). Another trio exome sequencing study identified de novo FOXP2 variants in two individuals with ASD and speech and language disorder (Deciphering Developmental Disorders, 2015; Reuter et al., 2017). Furthermore, recent data from the largest ASD exome sequencing cohort to date (>35,000 total individuals) place FOXP2 among 78 significant ASD-associated risk genes (Satterstrom et al., 2019). Additionally, a large (>50,000 total individuals) genome-wide association study of attention deficit/hyperactivity disorder (ADHD) identified an intron of FOXP2 among 12 loci associated with this disorder (Demontis et al., 2019). While further work is needed to clarify how FOXP2 variation may contribute to these disorders, the high genetic overlap between ASD and ADHD (Ghirardi et al., 2017) suggests that FOXP2 could function upstream of biological pathways pertinent to both NDDs.

FOXP1 mutations cause a severe global neurodevelopmental disorder

After association of FOXP2 with a developmental speech and language disorder in 2001 (Lai et al., 2001), case reports linking its closest paralog FOXP1 (64% total protein sequence similarity, 89% in the forkhead domain) to NDDs began surfacing in the literature (Carr et al., 2010; Pariani, Spencer, Graham, & Rimoin, 2009). While these initial studies identified large deletions within FOXP1 in children with severe motor and speech delays, each had limitations precluding a direct link between FOXP1 mutations and NDDs. Given its high similarity to FOXP2, researchers were surprised that no pathogenic mutations in FOXP1 were identified in individuals with verbal dyspraxia (Vernes, MacDermot, Monaco, & Fisher, 2009). This spurred a broader search for FOXP1 mutations in patients with other NDDs with strong language impairments, such as ASD and ID. These efforts identified de novo FOXP1 mutations in two patients diagnosed with ASD and severe language impairment (Hamdan et al., 2010). Another report directly linked FOXP1 mutations to ID by finding three patients with large deletions affecting only FOXP1 within a large cohort of individuals with unexplained ID (Horn et al., 2010). Together, these initial findings implicated FOXP1 in regulating broader neurodevelopmental processes underlying cognitive ability, compared to the more selective deficits in speech and language seen with FOXP2 mutations.

In 2011, the first whole-exome sequencing study of individuals with sporadic ASD and their families found a de novo mutation in FOXP1 in one patient with a high ASD severity score, language delay, and ID (O'Roak et al., 2011). Subsequently, several large-scale high-throughput sequencing studies have uncovered that de novo, likely gene-disrupting mutations in FOXP1 are among the most significantly recurrent mutations in ASD or ID/NDD cohorts (Coe et al., 2019; Coe et al., 2014; Iossifov et al., 2014; Stessman et al., 2017). To date, over 40 pathogenic variants have been documented, and distinctive neurodevelopmental phenotypes associated with FOXP1 mutations have emerged from the many published case reports (Le Fevre et al., 2013; Mutlu-Albayrak & Karaer, 2019; Palumbo et al., 2013; Sollis et al., 2016; Song, Makino, Noguchi, & Arinami, 2015; Thevenon et al., 2014; Urreizti et al., 2018; Vuillaume et al., 2018; Yamamoto-Shimojima, Okamoto, Matsumura, Okazaki, & Yamamoto, 2019). As with FOXP2 mutations, it is again important to note that these FOXP1 mutations may result in haploinsufficiency or dominant negative effects of mutant proteins, and further functional work is needed to disentangle the molecular mechanisms leading to patient phenotypes.

Meta-analyses of molecular and clinical data from FOXP1 patients have revealed a recognizable “FOXP1 syndrome” (Meerschaut et al., 2017; Siper et al., 2017). Core features included developmental delay, fine and gross motor coordination deficits, speech delay, articulation problems, ASD symptoms, and mild to moderate ID. Moreover, all patients had clinically significant ADHD symptoms, such as inattention, hyperactivity, and impulsivity (Meerschaut et al., 2017; Siper et al., 2017). Other medical features included endocrine, gastrointestinal, sleep, and sinopulmonary problems (Siper et al., 2017). Combining reports from both studies, 61% of individuals with FOXP1 mutations had brain structural abnormalities, few of which were shared across patients other than ventricular abnormalities (Meerschaut et al., 2017; Siper et al., 2017). Other structural abnormalities included cerebral/cerebellar atrophy, cortical and subcortical white matter alterations, and arachnoid cysts (Meerschaut et al., 2017; Siper et al., 2017). Notably, patients with a FOXP1 point mutation equivalent to a FOXP2 point mutation causing speech and language disorder still presented with these broader developmental delays and deficits (Sollis et al., 2017). Thus, despite the high degree of sequence similarity between FOXP1 and FOXP2, mutations in their genes typically result in distinct NDDs. How these phenotypic differences arise at the circuit, cellular, and molecular level is a subject of current explorations into FoxP1/2 function in the brain, including those described in this review.

FOXP4 is a candidate gene for neurodevelopmental disorder

In 2016, whole exome sequencing of consanguineous Arab families uncovered a recessive, homozygous frameshift mutation in FOXP4 (Charng et al., 2016). The affected individual showed developmental delay with laryngeal hypoplasia, feeding difficulties, and ventricular septal defects. However, this individual also carried two homozygous mutations in LRRC1 and ZNF514. While additional studies will be needed to confirm a disease association, this study provides the first indication that mutations in FOXP4 may be linked to NDDs.

ANIMAL MODELS OF FOXP FUNCTION

Expression patterns of FOXP genes and clinical phenotyping of patients with FOXP mutations have indicated key brain regions and cell types for the molecular actions of these transcription factors. A crucial approach for linking molecular functions to behavioral phenotypes is the manipulation of gene expression and function in animal models, which can be conducted in a brain region- or cell type-specific manner. Here we describe animal models of FOXP1/2/4 function with regards to molecular, neurophysiological, brain morphological, and behavioral phenotypes.

Animal models of FOXP1 function

Whole-body Foxp1 loss-of-function

Foxp1 mutant mice were initially generated to study this gene in non-neural tissues. An investigation of Foxp1 in heart function used homologous recombination to replace the forkhead domain with a neomycin cassette (Foxp1-FH-KO) (Wang et al., 2004), while a study of immune B cell development generated a second Foxp1 mutant line by targeting the N-terminal two-thirds of the protein, including the forkhead domain (Foxp1^−/−) (Hu et al., 2006). Because homozygous Foxp1-FH-KO mice die at E14.5 due to impaired cardiac development (Wang et al., 2004), heterozygous Foxp1 knockout (Foxp1^+/−) mice have been instrumental for understanding Foxp1 functions in postnatal neurodevelopment and behavior (Araujo et al., 2015). Neonatal Foxp1^+/− mice gain weight and perform gross motor functions normally but show abnormalities in ultrasonic vocalizations (USVs). In adulthood they exhibit hyperactivity and impaired grip strength but show normal motor learning in the rotarod assay. Brain region-specific Western blotting and RNA-seq conducted on adult Foxp1^+/− mice revealed FOXP1 reductions and subsequent dysregulation of ASD-associated genes in the hippocampus and striatum. Curiously, FOXP1 protein levels and overall gene expression were unchanged in the cortex, suggesting compensatory Foxp1 upregulation in this brain region. This study also identified a cell type-specific role for Foxp1 in striatal SPNs, whereby reduction of Foxp1 increased the excitability of iSPNs but had no effect on dSPNs.

To conduct cell type-specific or temporal deletions of Foxp1, mice with a conditional allele were generated using homologous recombination to flank exons 11-12 with loxP sites for Cre-mediated deletion (Foxp1^flox/flox) (Feng et al., 2010). As described below, conditional homozygous deletions generated using these mice have been powerful in revealing mutant phenotypes in distinct cell types throughout the brain. While it is important to consider that these manipulations are distinct from the heterozygosity of patient-specific mutations in FOXP1, they could represent a first step towards the development of circuit-specific targeted ASD therapeutics.

Brain-wide Foxp1 deletion

Brain-specific conditional knockout (cKO) of Foxp1 was performed by crossing Foxp1^flox/flox mice with Nestin-Cre mice, resulting in loss of Foxp1 from neuronal and glial precursors throughout the brain (Bacon et al., 2015; Frohlich, Rafiullah, Schmitt, Abele, & Rappold, 2017; Tronche et al., 1999). These mice are viable but gain weight more slowly than controls, and like Foxp1^+/− mice, they exhibit altered neonatal USVs (Bacon et al., 2015; Frohlich et al., 2017). In adulthood they demonstrate severe behavioral deficits, including hyperactivity, repetitive rearing and jumping, impaired learning and memory, social retreat, absent nest-building, and reduced sensorimotor integration (Bacon et al., 2015). Morphologically, their brains show large reductions in striatal area and dispersed neuronal organization in hippocampal CA1, and the neurons within these regions have altered dendritic branching (Bacon et al., 2015). Hippocampal CA1 neurons of these mice also have decreased excitability and larger excitatory postsynaptic current amplitudes (Bacon et al., 2015). These analyses prove a requirement for Foxp1 for normal striatal and hippocampal development and CA1 function, but evaluations of the cortex were absent despite high cortical expression of Foxp1.

Cortico-hippocampal Foxp1 deletion

The whole-brain Foxp1 cKO study established that Foxp1 plays a critical role in mammalian brain development and led researchers to ask which specific circuits and cell types regulate the phenotypes observed in these mice. To answer this question, two studies specifically deleted Foxp1 from cortical and hippocampal excitatory neurons using Emx1-Cre, which induces recombination at E10.5 in progenitors and projection neurons derived from the dorsal telencephalon (Araujo et al., 2017; Gorski et al., 2002; Usui, Araujo, et al., 2017). Cortico-hippocampal Foxp1 cKO mice produce fewer postnatal vocalizations and show cortical lamination defects (Usui, Araujo, et al., 2017). Gene expression profiling of these mice revealed that differentially expressed genes (DEGs) regulated by FOXP1 were associated with synapses and synaptic transmission, and that cortical DEGs were enriched for known ASD-associated genes (Usui, Araujo, et al., 2017). In adulthood, cortico-hippocampal Foxp1 cKOs showed significant deficits in spatial memory, motor learning, activity levels, social interaction, nest building, and USVs (Araujo et al., 2017). RNA-seq of cortical and hippocampal tissue from adult cKOs also found genes enriched in synaptic transmission and disease-relevant pathways. Confirming these molecular findings, cortico-hippocampal Foxp1 cKOs had significant deficits in long-term potentiation in the hippocampal CA1 region (Araujo et al., 2017). These studies were the first to dissect the region-specific role of Foxp1 during brain development at the molecular, functional, and behavioral levels.

Other cortical Foxp1 manipulations

Knockdown and overexpression studies of Foxp1 in cortical cells have offered additional insights into its functions in this region. In utero electroporation (IUE) of a Foxp1-targeting short hairpin RNA (shRNA) at E14.5 impairs cellular migration, as evidenced by accumulation of cells in the intermediate zone and fewer cells in the CP above (Braccioli et al., 2017; X. Li et al., 2015). Analysis of mature cortex after Foxp1 knockdown further supported cortical migration impairment via the presence of upper-layer cells in lower layers (X. Li et al., 2015). However, these studies conflicted on whether Foxp1 knockdown impairs neuronal differentiation, as one study claimed no change in pH3+ dividing cells or TBR2+ intermediate progenitors, while the other saw more TBR2+ progenitors and fewer SATB2+ neurons (Braccioli et al., 2017; X. Li et al., 2015). Additionally, neuronal morphology analyses after knockdown suggested impairments in multipolar-to-bipolar transition and dendritic outgrowth (X. Li et al., 2015). A recent study has modelled the effects of FOXP1 patient mutations on cortical development, with overexpression of FOXP1-R521X phenocopying Foxp1 knockdown in cortical migration and neuronal morphology deficits (X. Li et al., 2018).

Animal models of FOXP2 function

Whole-body Foxp2 loss-of-function

Following the association of FOXP2 with speech and language, several whole-body loss-of-function Foxp2 mouse lines were generated. The first line was generated by replacement of exons 12-13 with a neomycin cassette via homologous recombination, thus deleting part of the forkhead domain (Foxp2-ex12/13-KO) (Shu et al., 2005). Then two mouse lines were generated to model the KE family mutation, one of which resulted from a knockin strategy via homologous recombination (Foxp2-R552H-KI) (E. Fujita et al., 2008) and the other from an N-ethyl-N-nitrosourea (ENU) mutagenesis screen (Foxp2-R552H-ENU) (Groszer et al., 2008). Two additional mutations were generated from the ENU screen, one near the KE mutation (Foxp2-N549K) and the other similar to the R328X mutation in a family with speech and language disorder (Foxp2-S321X) (Groszer et al., 2008). The R552H and N549K mutations do not affect FOXP2 expression levels, but S321X causes absence of protein and reduction of mRNA, presumably due to nonsense-mediated decay (E. Fujita et al., 2008; Groszer et al., 2008). Last, another Foxp2 knockout line was generated by Cre-loxP deletion of exon 7, resulting in a similar reduction of mRNA and absence of protein (Foxp2-ex7-KO) (Enard et al., 2009).

Severe developmental abnormalities appear in mice homozygous for the above mutations, including delayed weight gain, gross motor impairments, reduced isolation USVs, reduced cerebellum size, and death by 3-4 weeks after birth (E. Fujita et al., 2008; Gaub, Groszer, Fisher, & Ehret, 2010; Groszer et al., 2008; Shu et al., 2005). Foxp2-N549K homozygotes show comparable but milder developmental and cerebellar volume deficits, and they survive for 3-5 months (Groszer et al., 2008). Other than volume, cerebellar Purkinje phenotypes vary by mouse line, with Foxp2-ex12/13-KO and Foxp2-R552H-KI mice showing disorganization and reduced dendritic arborization, and Foxp2-R552H-ENU mice showing normal histoarchitecture (E. Fujita et al., 2008; Groszer et al., 2008; Shu et al., 2005). Foxp2-ex7-KO mice show normal striosome-matrix formation in the striatum but fewer excitatory synapses and postsynaptic currents in SPNs (Chen et al., 2016). Other brain morphological phenotypes in homozygous Foxp2 mutants include impaired survival of amygdalar intercalated cells, abnormal thalamic patterning, reduced posterior thalamocortical projections, and altered cortical barrel formation (Ebisu et al., 2016; Kuerbitz et al., 2017). These studies reveal the requirement for Foxp2 for normal development and survival and are consistent with the absence of patients homozygous for FOXP2 mutations in the literature.

Mice heterozygous for Foxp2 mutations, on the other hand, develop mostly normally and survive into adulthood. Different mutant lines show slight variations in motor development and USVs, presumably due to differences in mutant generation methods and/or genetic background strains. For example, mice created through homologous recombination (Foxp2-ex12/13-KO, Foxp2-R552H-KI) show small decreases in postnatal weight gain and motor ability, while mice from the ENU screen (Foxp2-R552H-ENU, Foxp2-S321X) do not (E. Fujita et al., 2008; Groszer et al., 2008; Shu et al., 2005). Most heterozygote lines show neonatal USV reductions (Foxp2-ex7-KO, Foxp2-ex12/13-KO, Foxp2-R552H-KI, Foxp2-S321X), although one study of Foxp2-R552H-ENU heterozygotes did not find differences (Chen et al., 2016; E. Fujita et al., 2008; Gaub et al., 2010; Groszer et al., 2008; Shu et al., 2005). Foxp2-ex12/13-KO heterozygotes continue to show USV abnormalities from juvenile development into adulthood, with fewer courtship calls and altered bouts (Castellucci, McGinley, & McCormick, 2016). Adult Foxp2-R552H-ENU and Foxp2-S321X heterozygotes also exhibit courtship USV alterations in acoustic structure or social modulation of syntax (Chabout et al., 2016; Gaub, Fisher, & Ehret, 2016). These results generally implicate Foxp2 heterozygosity in vocalization deficits, consistent with the speech impairments found in patients heterozygous for FOXP2 mutations.

Additional behavioral assessments have been carried out in adult Foxp2 heterozygous mice. The most consistent finding among heterozygote lines (Foxp2-ex7-KO, Foxp2-R552H-ENU, Foxp2-S321X) is impairment on the rotarod assay (Enard et al., 2009; French et al., 2012; Groszer et al., 2008). Foxp2-R552H-ENU heterozygotes also show difficulties with a tilted running-wheel assay, another assessment of motor-skill learning (Groszer et al., 2008). In addition to these impairments, Foxp2-R552H-ENU and Foxp2-S321X mice exhibit impaired auditory-motor association learning (Kurt, Fisher, & Ehret, 2012). In contrast, Foxp2-ex12/13-KO heterozygotes did not have spatial learning deficits as tested in the Morris water maze, nor did Foxp2-R552H-ENU mice show working memory deficits or perseverative behaviors via spontaneous alternation in a T-maze (Groszer et al., 2008; Shu et al., 2005). Foxp2-ex7-KO heterozygotes did show increased exploration in a modified hole board assay, while Foxp2-R552H-ENU mice showed normal locomotion, anxiety, and grooming behaviors (Enard et al., 2009; Groszer et al., 2008). In summary, mice heterozygous for Foxp2 mutations commonly show alterations in motor-skill and auditory-motor association learning, while other forms of learning and baseline motor behaviors appear normal.

Neurophysiological and neurochemical alterations accompany the motor learning changes in Foxp2 heterozygotes. One brain region strongly implicated in these deficits is the striatum, which shows increased baseline firing rates in awake Foxp2-R552H-ENU mice and abnormally modulated activity during the rotarod assay (French et al., 2012). Slice electrophysiology studies also indicate absence of striatal long-term depression (LTD) in Foxp2-R552H-ENU mice, as well as decreased excitatory and increased inhibitory currents in dSPNs of Foxp2-S321X mice (Groszer et al., 2008; van Rhijn, Fisher, Vernes, & Nadif Kasri, 2018). Cerebellar slices from Foxp2-R552H-ENU mice show increased paired-pulse facilitation of parallel-fiber-Purkinje cell synapses, indicating changes in short-term plasticity (Groszer et al., 2008). Foxp2-R552H-ENU mice also show altered brainstem responses to auditory tones while Foxp2-S321X do not (Kurt, Groszer, Fisher, & Ehret, 2009). Analysis of neurotransmitter levels in Foxp2-ex7-KO heterozygotes found increased dopamine in the globus pallidus and nucleus accumbens, increased serotonin in the nucleus accumbens, and decreased GABA in the frontal cortex, suggesting altered chemical modulation of neurons in these brain regions (Enard et al., 2009). Altogether these studies indicate a requirement for Foxp2 for normal striatal, cerebellar, and auditory brainstem physiology and corticostriatal neurotransmitter levels.

In summary, heterozygous Foxp2 mutant mice have offered important insights into the etiology of speech and language disorders. They highlight motor-skill learning, auditory-motor association learning, and innate vocalization as key abilities regulated by Foxp2. They also reveal some of the neurophysiological and neurochemical abnormalities potentially underlying these disorders. Strongly implicated brain regions include the striatum and cerebellum, which agrees with the altered morphology and function of these regions in the KE family. To further investigate region- and cell type-specific contributions of Foxp2 to brain function, Foxp2^flox/flox mice were generated with loxP sites flanking exons 12-14, allowing for Cre-mediated deletion of the forkhead domain and subsequent loss of FOXP2 protein expression (French et al., 2007). Foxp2 cKO in various cell types, as well as other region-specific Foxp2 manipulations, are described below.

Cortex-specific Foxp2 manipulations

Developmental manipulations of cortical Foxp2 expression have been performed to ascertain the role of this gene in neurogenesis, neuronal migration, cortical morphology, and behavior. In one study, Foxp2 knockdown by IUE of shRNA at E13/14 impaired the transition of radial glia into intermediate progenitors and neurons, and also impaired the migration of neurons into the upper layers of the cortex (Tsui et al., 2013). However, crosses of Foxp2^flox/flox mice with cortical Cre-driver mice have generated cKO mice with normal cortical size, neuronal density, layering, and projections (Co, Hickey, Kulkarni, Harper, & Konopka, 2019; French et al., 2018; Kast et al., 2019; Medvedeva et al., 2018). These discrepancies could result from molecular compensation by other Foxp genes upon genetic Foxp2 deletion, but notably, Foxp1/4 transcripts were not upregulated in cortical neurons in an RNA-seq study of cortical Foxp2 cKO mice (Co et al., 2019).

Despite grossly normal cortical development, cortical Foxp2 cKO mice show behavioral deficits. In a study of motor-skill learning, Emx1-Cre cortical Foxp2 cKO mice performed normally during the rotarod assay, operant lever-pressing, and unperturbed sessions on the ErasmusLadder, another motor-skill learning test (French et al., 2018). However, during perturbed (i.e. more difficult) sessions of the ErasmusLadder, cortical cKO mice made more missteps than controls (French et al., 2018). ASD-related behaviors have also been investigated in cortical cKO mice. In one study, cKO mice generated using Neurod6/Nex-Cre, which causes recombination around E11.5 in postmitotic projection neurons of the dorsal telencephalon, spent less time socializing with conspecifics and engaged in less following, close contact, and social dominance behavior (Medvedeva et al., 2018). Another study found impaired behavioral flexibility in Emx1-Cre Foxp2 cKO mice using a reversal learning task in the water Y-maze (Co et al., 2019). These behavioral changes were not due to generalized anxiety or hyperactivity as assayed by open field, elevated plus maze, or locomotor activity tests (Co et al., 2019; Medvedeva et al., 2018). While adult Neurod6/Nex-Cre Foxp2 cKO mice showed subtle, social context-dependent alterations in USVs, mice with Foxp2 deletion by Emx1-Cre vocalized normally save for a slight decrease in neonatal call loudness (Co et al., 2019; Medvedeva et al., 2018). These discrepancies could be attributed to methodological differences between the two studies, such as the slight timing difference between Cre lines, the use of superovulated females in the Neurod6/Nex-Cre study to elicit male USVs, or differences in the categorization of USV call types. Nonetheless, these behavioral changes were accompanied by several gene expression alterations in cortical neurons, including downregulation of the ASD gene Mint2, dopamine signaling genes Drd1 and Ppp1r1b, and cytoskeletal genes (Co et al., 2019; Medvedeva et al., 2018).

Together, these studies indicate that cortical Foxp2 is required for motor-skill learning, proper social behavior, and behavioral flexibility. They also suggest a role for Foxp2 in cortical modulation of USVs, the ethological effects of which remain undetermined. Furthermore, the lack of major cortical development phenotypes in cortical Foxp2 cKO mice prompt further investigation of circuit and synaptic functions for Foxp2 in this brain region.

Striatum-specific FoxP2 manipulations

Striatal manipulations of Foxp2 expression highlight important roles in motor function. To assess striatal Foxp2 in motor-skill learning, striatal Foxp2 cKO mice were generated using Rgs9-Cre, which is expressed at postnatal day (P) 8 in most SPNs (French et al., 2018). Like cortical cKO mice, striatal cKO mice developed normally, performed normally on the rotarod, and made more missteps during perturbed ErasmusLadder sessions than controls. However, in addition, they made fewer presses during high-speed operant lever-pressing, and they executed presses more variably. This variability was not seen in cortical or cerebellar Foxp2 cKO mice, indicating a specific role for striatal Foxp2 in controlling motor variability. Another study performed viral Foxp2 knockdown in the adult striatum to assess its roles in Huntington’s disease-related phenotypes, which include motor impairments (Hachigian et al., 2017). This knockdown decreased the latency for mice to fall off the rotarod, and also reduced vertical activity, rearing, and climbing in an open field. Thus, these studies indicate that striatal Foxp2 is essential for motor function.

Proper FoxP2 expression in zebra finch Area X is also required for song learning and production. Lentiviral FoxP2 knockdown in juvenile Area X causes incomplete and inaccurate copying of tutor songs and high vocal variability over development (Haesler et al., 2007; Murugan, Harward, Scharff, & Mooney, 2013), an effect potentially mediated through decreased spine density on SPNs (Schulz, Haesler, Scharff, & Rochefort, 2010). FoxP2 overexpression in juvenile Area X also impairs tutor song copying (Heston & White, 2015). In adult zebra finches, FoxP2 knockdown abolishes the decrease in song variability that normally occurs during directed singing to a female (Murugan et al., 2013). This behavior change is accompanied by decreased dopamine D1 receptor and DARPP-32 expression in Area X, similar to the Drd1 and Ppp1r1b downregulation in Foxp2 cKO mouse cortex (Co et al., 2019; Murugan et al., 2013). FoxP2 overexpression in adult Area X does not affect vocal variability, but upon deafening, overexpression hastens song deterioration and alters song variability, indicating a role for FoxP2 in maintaining song stability (Day, Hobbs, Heston, & White, 2019).

In summary, these studies show a critical, conserved role for striatal FoxP2 in motor-skill learning and motor variability. Further work is needed to clarify the role of striatal Foxp2 in mouse vocalizations and other NDD-relevant behaviors. Additionally, given that Rgs9-Cre acts postnatally, studies of the effects of embryonic striatal Foxp2 deletion are needed to fully understand its functions in striatal development.

Thalamus-specific Foxp2 manipulation

Whole-body homozygous Foxp2-R552H-KI mice show altered thalamic patterning, with reduced posterior nuclei and expanded intermediate nuclei (Ebisu et al., 2016). To prove a cell-autonomous origin for this phenotype, the authors also performed IUE knockdown of Foxp2 in the thalamic primordium at E11, and they found a similar result of downregulated posterior nuclei markers and upregulated intermediate nuclei markers, confirming the role of Foxp2 in posterior thalamic development. No other published studies have performed thalamus-specific manipulations of Foxp2; therefore, much remains to be explored regarding Foxp2 function in this core region of vocalization and sensorimotor learning circuitry.

Cerebellum-specific Foxp2 manipulations

To assess the contribution of Purkinje cell Foxp2 to motor-skill learning, Purkinje Foxp2 cKO mice were generated using Pcp2/L7-Cre, which is expressed by P9 in all Purkinje cells of the cerebellum (French et al., 2018). This resulted in complete loss of FOXP2 from these cells without affecting expression in DCN neurons. While Purkinje cKO mice showed normal performance on the rotarod, they had fewer normal and high-speed lever presses and more missteps during both unperturbed and perturbed sessions of the ErasmusLadder (French et al., 2018). Extracellular recordings of Purkinje cells in head-fixed mice walking on a wheel revealed increased intrinsic activity and decreased correlation of activity with phases of walking (French et al., 2018). Other cerebellar manipulations of Foxp2 highlight its importance for mouse motor function and vocal communication. IUE of a Foxp2-targeting shRNA in the developing cerebellum caused major deficits in neonatal motor behaviors and isolation USV production (Usui, Co, et al., 2017). This knockdown also caused defects in Purkinje dendritic arborization reminiscent of those seen in Foxp2 mutant homozygotes. In addition, Purkinje cell-specific Foxp2 rescue in Foxp2-R552H-KI heterozygotes restored the relative proportions of whistle-, short-, and click-type calls to wild-type levels, suggesting that cerebellar Foxp2 rescue may be sufficient for restoring USV deficits in Foxp2 mutant mice (Fujita-Jimbo & Momoi, 2014).

These Purkinje cell-specific experiments point to the cerebellum as a key site in the brain for Foxp2 function in mice, consistent with the major cerebellar abnormalities seen in Foxp2 mutant homozygotes. Roles of FOXP2 in other cells of the cerebellar circuit, such as the DCN and inferior olive, remain unknown, nor have the downstream molecular pathways of FOXP2 in any cerebellar cell type been fully elucidated.

Summarized in Figure 6, studies of Foxp1/2 using mouse models have identified important roles for these transcription factors in the development and function of various neural circuits, with regards to vocalization, motor, social, and cognitive behaviors.

Figure 6. Summary of reported behavioral findings in Foxp1 and Foxp2 mouse experimental systems.

Behavior deficits in whole-brain Nestin-Cre; Foxp1 cKO mice (A), Foxp1 heterozygous mice (B), Emx1-Cre; Foxp1 cKO mice (C), Foxp2 mutant or heterozygous mice (D), Neurod6/Nex-Cre; Foxp2 cKO mice, (F) Emx1-Cre; Foxp2 cKO mice, Rgs9-Cre; Foxp2 cKO mice (G), and Pcp2/L7-Cre; Foxp2 cKO mice (H). Highlighted areas within brain schematics indicate brain regions with targeted deletion of Foxp1 or Foxp2 within mouse strains. *(D) represents different mouse strains that have whole-body, heterozygous loss-of-function mutations in Foxp2 that include the following strains: Foxp2-R552H-KI, Foxp2-R552H-ENU, Foxp2-S31X, and Foxp2-ex7-KO mice.

Animal models of FOXP4 function

Compared to FOXP1/2, the role of FOXP4 in neurodevelopment has been understudied. Foxp4 KO mice, originally generated to study lung and immune system development, are embryonic lethal between E10.5-E12.5 due to severe cardiac defects, earlier than the Foxp1 KO lethality at E14.5 (S. Li, Zhou, Lu, & Morrisey, 2004; Rousso et al., 2012). A subsequent Foxp4 cKO strain was developed but has not yet been used to study the role of Foxp4 in neurodevelopment (S. Li et al., 2012). Two studies to date have examined the contribution of Foxp4 to neural development. The first examined the role of Foxp4 in the cerebellum via knockdown in mouse organotypic cerebellar slices at P10, a period of rapid dendritic growth in PCs, and found a significant reduction of PC dendritic arborization (Tam et al., 2011). Similar cerebellar defects were observed with Foxp2 knockdown in mouse cerebellar slices and in vivo (French et al., 2007; E. Fujita et al., 2008; Shu et al., 2005; Usui, Co, et al., 2017), suggesting that Foxp2/4 may co-regulate pathways critical for PC development. The second study of FOXP4 function examined the role of FOXP2/4 in chick and mouse neural tube development (Rousso et al., 2012). In both models, reduction of FOXP2/4 led to a variety of neural tube defects driven by enhanced neural progenitor maintenance. FOXP4 repressed N-cadherin, a key component of adherens junctions between neural progenitor cells, and promoted neural differentiation by opposing the progenitor maintenance activity of SOX2. Foxp4-null mice also exhibited gross morphological defects, including holoprosencephaly, spina bifida, and occasional notochord and floor plate duplications (Rousso et al., 2012). These studies suggest that FOXP4 is critical for global developmental processes important for survival, which might explain the rarity of FOXP4 mutations in humans.

Conclusion

Nearly two decades of research examining the expression of FOXP proteins in the brain and their role in vertebrate brain development has provided important insights into the molecular pathways and brain circuits underlying cognitive behaviors. The spatial and temporal expression patterns of FOXP1, FOXP2, and FOXP4 vary within distinct brain regions and cell-types, with few brain regions highly expressing all three factors (the striatum being a notable region of overlap). These transcription factors also share highly concordant expression patterns across species. Future studies will need to better elucidate the cell-type specificity of FOXP1/2/4 expression within certain brain regions over development. The finding that FOXP proteins form both hetero- and homodimers to regulate transcriptional activity raises interesting questions about how these proteins might cooperate to regulate pathways in the developing brain. Moreover, FOXP1 and FOXP2 have been directly linked to specific neurodevelopmental disorders that affect speech and language ability. Given both their unique patterning in the brain and their link to disease, a major undertaking in the FOXP field has been to understand how disruption of FOXP1/2 expression in distinct brain circuits in animal experimental systems contributes to phenotypes seen in human patients. Deletion of Foxp1 throughout the mouse brain or within distinct brain regions, such as the cortex and hippocampus, replicates several phenotypes observed in human patients with FOXP1 mutations, including impairments in intellectual ability, gross motor learning, and vocal behavior. Manipulation of FOXP2 function or expression levels in distinct regions of the mouse and bird brains leads to common defects in vocal and motor learning behaviors reminiscent of the speech deficits found in human patients with FOXP2 mutations. Targeting distinct circuits to rescue behavioral deficits in animal models will be an important future direction for the field.