Hox Genes: Choreographers in Neural Development, Architects of Circuit Organization

Summary

The neural circuits governing vital behaviors, such as respiration and locomotion, are comprised of discrete neuronal populations residing within the brainstem and spinal cord. Work over the past decade has provided a fairly comprehensive understanding of the developmental pathways that determine the identity of major neuronal classes within the neural tube. However, the steps through which neurons acquire the subtype diversities necessary for their incorporation into a particular circuit are still poorly defined. Studies on the specification of motor neurons indicate that the large family of Hox transcription factors has a key role in generating the subtypes required for selective muscle innervation. There is also emerging evidence that Hox genes function in multiple neuronal classes to shape synaptic specificity during development, suggesting a broader role in circuit assembly. This review highlights the functions and mechanisms of Hox gene networks, and their multifaceted roles during neuronal specification and connectivity.

Introduction

Nervous system development relies on the establishment of precise connections between neurons and their pre- and post-synaptic targets. In many cases the neural circuits that shape basic behaviors are defined during embryonic development, with little influence from spontaneous or sensory-evoked neuronal activity. These hard-wired programs can be linked to signaling systems operating over a narrow window during embryogenesis. A major outcome of these patterning systems is to establish specific profiles of transcription factors in neuronal progenitors and postmitotic cells, thus defining unique molecular signatures for the thousands of subtypes comprising the nervous systems of most animal species. Transcription factors orchestrate key aspects of circuit formation by deploying cell-specific programs that define the migration, projection pattern, and synaptic specificity of neuronal subtypes. A significant question is whether there are any coherent sets of developmental principles which link early progenitor identity to the incorporation of specific groups of cells into a neural circuit that controls a specific behavior.

Progress towards understanding the developmental basis of neural circuit assembly has emerged through studies on the signaling pathways which determine the identity of neuronal subtypes along the dorsoventral and rostrocaudal axes of the neural tube. The contribution of these systems to neural circuit formation has been most intensely studied along the dorsoventral axis, where each progenitor domain is specified by a unique profile of transcription factors and gives rise to distinct classes of postmitotic neurons (reviewed in Jessell, 2000; Shirasaki and Pfaff, 2002). Modern genetic tools have provided the means to assess how a single neuronal class defined by transcription factor expression contributes to circuits controlling basic motor functions (reviewed in Arber, 2012). In many cases removal of a single transcription factor affects circuits associated with multiple motor behaviors (Bouvier et al., 2010; Lanuza et al., 2004), suggesting additional positional inputs are necessary to facilitate the incorporation of a neuron into a specific circuit.

Recent studies indicate that genetic programs acting along the rostrocaudal axis provide an important means through which neuronal classes establish subtype identities. The role of rostrocaudal positional identity in neuronal specification has been examined in the context of spinal motor neuron (MN) development, where there is a clear segregation of neurons targeting specific muscles along the rostrocaudal axis of the spinal cord (Dasen and Jessell, 2009). In addition to MNs, some of the key neural circuits controlling basic motor behaviors appear to rely on rostrocaudal positional information, including the rhythmically active circuits that control walking and breathing (reviewed in Ballion et al., 2001; Kiehn and Kjaerulff, 1998).

An important family of transcription factors which endow neural cell types with positional identities along the rostrocaudal axis are encoded by genes within the Hox clusters. Hox genes are found in all animal species and have conserved roles in body patterning (reviewed in McGinnis and Krumlauf, 1992). In most vertebrates they are comprised of 39 genes distributed across four clusters, referred to as HoxA, HoxB, HoxC or HoxD (Figure 1A). Hox genes within a cluster are classified as belonging to one of thirteen paralog groups (Hox1-Hox13), and a single cluster contains only a subset of the thirteen groups. Each gene is characterized by the presence of a 60 amino acid region encoding the homeodomain that mediates DNA binding. The majority of Hox genes are expressed in the central nervous system (CNS) where they have critical functions in neuronal specification and target connectivity.

Figure 1. Hox Expression Patterns in the Hindbrain and Spinal Cord.

(A) In vertebrates 39 Hox genes are distributed across 4 clusters. Each Hox gene is expressed in discrete rostrocaudal domains within the hindbrain and spinal cord. Color coding of Hox genes represents expression domains along the rostrocaudal axis.

(B) In the hindbrain, Hox genes from paralog groups 1–5 are expressed and anterior expression limits correspond to rhombomere boundaries. Higher color intensity denotes higher expression. Hoxa1 expression is transient. Hindbrain motor nuclei develop within specific rhombomeres and are shown within their rhombomeres of origin. IV=trochlear, V=trigeminal, VI=abducens, VII=facial, IX=glossopharyngeal, X=vagus, XI=accessory, X=hypoglossal.

(C) In the spinal cord, expression of Hox4-Hox11 genes aligns with MN columnar and pool subtypes. PMC=phrenic motor column, LMC=lateral motor column, HMC=hypaxial motor column, PGC=preganglionic motor column, MMC=medial motor column. Although technically a pool, we define phrenic MNs as a column due to their unique trajectory and because they do not reside within a larger columnar group. Peripheral targets of motor columns are indicated. LMC MNs further diversify in ~50 motor pools targeting limb muscles at brachial and lumbar levels.

In this review we highlight the diverse roles of Hox transcription factors in nervous system development, and compare and contrast their functions in the hindbrain and spinal cord. We describe recent studies which have revealed novel strategies through which Hox proteins contribute to neuronal diversity, connectivity, and circuit assembly. Studies on the mechanisms of Hox gene regulation, and the pathways through which their downstream effectors are controlled, have also provided important clues into how motor neuron subtypes are organized and have evolved. Moreover, recent assessments of Hox gene function in the hindbrain and spinal cord suggest key roles for this class of transcription factors in neural circuit assembly.

Hox expression and function in the nervous system

The embryonic hindbrain and spinal cord generate the neural circuitry required for basic motor functions such as respiration and locomotion, as well as a diverse array of sensory modalities including nociception, proprioception, audition and balance. During development, the hindbrain is transiently segmented into eight distinct compartments, or rhombomeres (r1-r8), that give rise to the pons, medulla and cerebellum. While no physical barriers exist between rhombomeres, differential cell adhesion properties, which develop with a conserved 2-segment periodicity, prevent intermixing of cells between compartments (Guthrie et al., 1993; Wizenmann and Lumsden, 1997). Unlike the hindbrain, the spinal cord does not undergo a phase of overt segmentation; however neuronal cell types are organized based on their rostrocaudal position. Spinal MNs, for example, exhibit stereotypical clustering patterns and follow selective axonal trajectories depending on their rostrocaudal coordinates (Landmesser, 2001). The spinal cord can be classified into cervical, thoracic, lumbar and sacral regions, based on the segmental position of vertebrae, as well as the organization of motor and sensory nerve roots.

Hox gene expression in the hindbrain and spinal cord is spatially and temporally dynamic. In general Hox1-Hox5 paralog group genes are expressed in the hindbrain while Hox4-Hox11 genes are detected in the spinal cord (Figure 1A–1C). Some Hox genes are expressed over a narrow time window during early development, while others may persist to postnatal stages. Most of our understanding of Hox gene function in vertebrates derives from knockout studies in mice or manipulation of Hox activity in chick embryos. Since Hox genes are expressed broadly in the embryo, historically it has been difficult to assess the cell autonomy of their function in the CNS. Moreover, even when a Hox gene is expressed during an early narrow time window, effects of its deletion may manifest in later steps of circuit formation, confounding the interpretation of mouse knockouts. Additional complications arise from the compensation observed between Hox paralog groups and cross-regulatory interactions between Hox genes, which result in changes in expression of other Hox genes in Hox null animals. Despite these challenges, genetic models of Hox function have provided valuable insights into their roles in CNS development (Table 1). Here we outline some of the emerging concepts from these studies as they relate to neuronal specification and circuit formation.

Table 1.

Mutant	Nervous system phenotype	References
Hox1 genes
Hoxa1	defects in caudal rhombomere boundaries, severe reduction of r4 and absence of r5, VI nerve and motor nucleus absent, smaller VII nucleus and thinner nerve with ectopic exit points, lack of superior olivary complex, defects in IX and X sensory ganglia, appearance of r2-like cells in r3, ectopic cells integrated into a novel functional respiratory network	Lufkin et al., 1991, Chisaka et al., 1992, Carpenter et al.,1993, Mark et al., 1993, Barrow et al., 2000, Helmbacher et al., 1998, del Toro et al., 2001
Hoxb1	homeotic transformation of r4 to r2/r3, VII MNs acquire V identity, defects in VII nucleus migration and axonal pathfinding resulting in loss of motor nucleus and nerve, misspecification of contralateral vestibuloacoustic afferents, ectopic serotonergic neuron generation in r4, loss of first-order visceral sensory neurons in r4, defects in auditory circuit formation, loss of lateral vestibulospinal tract projections	Goddard et al., 1996, Studer et al., 1996, Gavalas et al., 2003, Pattyn et al., 2003, Gaufo et al., 2004, Di Bonito et al., 2013, Chen et al., 2012
Hoxa1/Hoxb1	lack of rhombomeres 4 and 5, defects in patterning of cranial nerves VII through XI	Rossel et al., 1999, Studer et al.,1998, Gavalas et al., 1998
Hoxd1	defects in nociceptor specification and nociceptive circuit formation	Guo et al., 2011
Hox2 genes
Hoxa2	no r1/r2 and r2/r3 boundaries, expansion of r1-derived cerebellar territory, disorganization and abnormal pathfinding of V motor neurons, reduction of VII nucleus and nerve, defects in dorsoventral neuronal specification, loss of vestibuloacoustic afferents, loss of somatic sensory neurons in r2 and severe reduction in r3, increased oligodendrocyte production in r2-r3, defects in somatosensory map formation and sound-localizing auditory circuits	Gavalas et al., 1997, Barrow et al., 2000, Davenne et al., 1999, Gaufo et al., 2004, Miguez et al., 2012, Oury et al., 2006, Di Bonito et al., 2013
Hoxb2	reduced VII nucleus-partial transformation to V identity, ectopic serotonergic neuron generation in r4, decreased oligodendrocyte production in r4, defects in auditory circuit formation	Barrow et al., 1996, Davenne et al., 1999, Gavalas et al., 2003, Pattyn et al., 2003, Miguez et al., 2012, Di Bonito et al., 2013
Hoxa2/Hoxb2	lack of interhombomeric boundaries from r1 to r4, lack of Evx1+ interneurons in r2 and r3, defects in dorsoventral neuronal specification	Davenne et al., 1999
Hoxa1/Hoxa2	complete lack of rhombomere boundaries	Barrow et al., 2000
Hox3 genes
Hoxa3	abnormal projections of IX and X ganglia	Manley et al., 1997, Watari et al., 2001
Hoxa3/Hoxb3	absence of VI nucleus, loss of first-order visceral sensory neurons in r5	Gaufo et al., 2003, Gaufo et al., 2004
Hoxa3/Hoxd3	generation of VII-like MNs that caudally migrate to r7	Gaufo et al., 2003
Hox5 genes
Hoxa5/Hoxc5	abnormal diaphragm innervation, reduced and disorganized PMC	Philippidou et al., 2012
Hox6 genes
Hoxa6/Hoxc6	reduction in brachial LMC	Lacombe et al., 2013
Hoxc6	reduction of Pea3+ pool, decreased innervation of CM muscle	Lacombe et al., 2013
Hox8 genes
Hoxb8	degeneration of the second spinal ganglion, abnormal dorsal horn neuronal distribution	van den Akker et al., 1999, Holstege et al., 2008
Hoxc8	reduction in brachial MNs, abnormal projections into the limb, reduction and abnormal migration of Pea3+ motor pool	Tiret et al., 1998, Vermot et al., 2005
Hox9 genes
Hoxc9	lack of PGC and HMC neurons, expansion of brachial LMC	Jung et al., 2010
Hox10 genes
Hoxc10	loss of lumbar MNs	Hostikka et al., 2009
Hoxa10/Hoxd10	loss and disorganization of lumbar MNs, defective hindlimb innervation	Wahba et al., 2001, Lin et al., 2003
Hoxc10/Hoxd10	acquirement of thoracic identities in lumbar MNs, abnormal hindlimb innervation	Wu et al., 2008
Hox13 genes
Hoxb13	caudally extended spinal cord, supernumerary DRGs, defective tail sensory innervation	Economides et al., 2003

Loss of hindbrain compartmental boundaries in Hox null mice

In the hindbrain, initiation of Hox gene expression precedes segmentation, with Hox1 paralogs, Hoxa1 and Hoxb1, expressed as early as e7.5, followed by paralog groups 2 and 3 at e8.5 (Tumpel et al., 2009). Mice lacking Hoxa1 exhibit defects in hindbrain segmentation (Chisaka et al., 1992; Lufkin et al., 1991), including the absence (Carpenter et al., 1993) or reduction (Mark et al., 1993) of r5 and the reduction in the size of r4, leading to the extension of r3 (Figure 2A). While Hoxb1 null mice do not manifest defects in early hindbrain patterning (Goddard et al., 1996; Studer et al., 1996), compound Hoxa1/Hoxb1 mutants exhibit a more severe phenotype than Hoxa1 mutants, and lack both r4 and r5 (Gavalas et al., 1998; Rossel and Capecchi, 1999; Studer et al., 1998). Hoxa2 mutants lack r1/r2 and r2/r3 borders (Barrow et al., 2000; Gavalas et al., 1997) leading to an expansion of the cerebellum and a corresponding reduction of pontine structures, while Hoxa2/Hoxb2 compound mutants show a more severe phenotype, lacking inter-rhombomeric boundaries between r1-r4 (Davenne et al., 1999). Hoxa1/Hoxa2 double mutants exhibit complete absence of rhombomere boundaries (Barrow et al., 2000). It therefore appears that at the initial stages of hindbrain development, the earliest expressed Hox1 and Hox2 genes are responsible for its compartmentalization.

Figure 2. Mutations in Hox Genes Cause Defects in MN Development, Migration and Axon Guidance.

(A) Mutations in Hox1-Hox3 genes result in misspecification, disorganization and abnormal projections of hindbrain MNs. The schematics are composites, showing both early segmentation defects and subsequent MN defects. Facial MNs (VII) caudally migrate to r6 between embryonic day (e)11-e14. The facial (VII) nucleus is absent in Hoxb1 mutants and reduced in Hoxa1, Hoxb2 and Hoxa2 mutants. Ectopic trigeminal nuclei are generated in Hoxa1, Hoxb1 and Hoxb2 mutant mice but are subsequently cleared by apoptosis. Trigeminal (V) axons are misrouted in Hoxa2 mutants and the abducens (VI) nucleus is absent in Hoxa1 and Hox3 mutants. In Hoxa1 mutants, “rx” denotes a hybrid region with no clear rhombomeric identity, in Hoxb1 mutants r4 is transformed to r2/r3-like and in Hox3 mutants r6 acquires an r4 identity.

(B) Mutations in Hox5-Hox10 genes result in transformation or reduction of distinct motor columns. Hox5 genes control PMC development, Hoxc9 determines thoracic MN identities while Hox6, Hoxc8 (at brachial levels) and Hox10 (at lumbar levels) genes define aspects of LMC identity. Hoxc6 and Hoxc8 are also required for the specification of pools defined by expression of Pea3 and Scip. In Foxp1−/− mice all Hox-dependent programs are disrupted with the exception of PMC specification. MMC neurons are not depicted as their development is thought to be Hox-independent.

Hox genes specify the identity and connectivity of cranial nerves

Following hindbrain segmentation, combinations of Hox genes act as determinants of neuronal identity within rhombomeres. In both the hindbrain and spinal cord, changes in neuronal identity and connectivity in the absence of Hox genes are exemplified by their effects on motor neurons (Figure 2A–2B).

In the hindbrain MNs are generated in specific rhombomeres and cluster in motor nuclei that can be classified as somatic, visceral (vm), or branchiomotor (bm) according to their target muscles (Guthrie, 2007). Somatic MNs are located ventrally and innervate body muscles derived from paraxial or prechordal mesoderm. Both vm and bm MNs migrate and project their axons dorsally, to innervate parasympathetic ganglia and muscles in the pharyngeal arches, respectively. Hindbrain somatic MNs are derived from a progenitor domain expressing the transcription factor Olig2, which also gives rise to all spinal MNs, while bm and vm MNs are derived from Nkx2.2+ progenitors that give rise to V3 interneurons in the spinal cord. Despite differences in their origins, all classes of hindbrain MNs require Hox genes for their development (Figure 2A).

In general, MNs that develop in regions of no or little overlap between Hox genes are the most susceptible to single gene mutations, likely due to lack of compensation by other paralog genes (Table 1, Figure 2A). For example, trigeminal (V) MNs develop in r2-r3, where Hoxa2 is either the only Hox gene expressed (r2) or co-expressed with Hoxb2 (r3). In the absence of Hoxa2, trigeminal MNs are disorganized and their axons are misrouted (Gavalas et al., 1997). Similarly, facial (VII) MNs are primarily determined by the activity of Hoxb1, which is selectively expressed in r4 (Popperl et al., 1995; Studer et al., 1994). In Hoxb1−/− mice, presumptive facial MNs acquire an r2/r3-like identity, displaying migration patterns and molecular signatures of trigeminal MNs, and fail to migrate caudally leading to the subsequent loss of the facial nerve (Gavalas et al., 2003; Goddard et al., 1996; Studer et al., 1996). Hoxb2−/− and Hoxa1−/− mice also show a severe reduction of the facial motor nucleus, which may be partly due to their requirement for transcriptional initiation and maintenance of Hoxb1 expression (Barrow and Capecchi, 1996; Davenne et al., 1999; Gavalas et al., 2003). Evidence that Hox genes act cell-autonomously in MNs is provided by misexpression studies in chick. Ectopic expression of Hoxa2 or Hoxb1 in r1 results in the generation of trigeminal- or facial-like MNs respectively (Jungbluth et al., 1999), while Hoxb1 misexpression in r2 leads to the transformation of trigeminal to facial neurons (Bell et al., 1999). Thus, despite cooperation between Hox genes in multiple contexts, certain Hox genes also act individually to dictate specific MN identities.

While certain motor nuclei rely on the activity of single Hox genes, others are specified by the combinatorial expression of several homologs. For example, the abducens (VI) nucleus, containing somatic MNs originating in r5, requires the collective activity of Hox3 group genes. These MNs are absent in Hoxa3/b3 double mutants and can be induced by ectopic Hoxa3 expression in chick (Gaufo et al., 2003; Guidato et al., 2003). The abducens nerve is also absent in Hoxa1−/− mice (Mark et al., 1993), despite the transient expression of the gene prior to MN differentiation (Murphy and Hill, 1991). MNs of the glossopharyngeal (IX) nerve are derived from r6 and also require the function of Hox3 genes for correct pathfinding (Manley and Capecchi, 1997; Watari et al., 2001). In the absence of Hox3 paralogs, there is also a derepression of Hoxb1 in r6, resulting in the ectopic generation and caudal migration of facial MNs (Gaufo et al., 2003). Thus, the generation of the appropriate MN subtypes in the hindbrain requires a complex interplay between multiple Hox genes.

Hox genes in neural crest cells: non-cell autonomous effects on bm axon guidance

Hindbrain branchiomotor neurons derive from multiple segments but their axons exit the hindbrain primarily from even-numbered rhombomere exit points; trigeminal (V) nerve exits from r2, facial nerve (VII) from r4 and glossopharyngeal (IX) nerve from r6 to innervate arches 1, 2 and 3 respectively. Hindbrain segmentation impinges on head development by generating neural crest cells that migrate into the pharyngeal arches. Interestingly, neural crest cells migrating to a pharyngeal arch are generated in the same rhombomeres as bm neurons innervating that arch, suggesting coordinated development of neurons and their targets (Trainor and Krumlauf, 2001). In Hoxa2 mutants, trigeminal MNs exit the hindbrain from r4 instead of r2, resulting in the innervation of pharyngeal arch 2 and not their correct target, arch 1. Since Hoxa2 is also a determinant of neural crest cells migrating from r4 to arch 2, Hoxa2 global deletion transforms the environment within arch 2 to arch 1-like (Gendron-Maguire et al., 1993; Prince and Lumsden, 1994; Rijli et al., 1993). This raises the question of whether the axon guidance defects observed are cell-autonomous for trigeminal MNs or a result of changes in environmental cues. At late stages of development, Hoxa2 mutants also show a reduction of the facial motor nucleus and nerve, due to a requirement for Hoxa2 in arch 2 neural crest cells (Briscoe and Wilkinson, 2004; Gavalas et al., 1997). Hoxb1 expression in neural crest cells populating the 2^nd arch is also required for correct pathfinding of facial MNs, underscoring the importance of coordinated Hox activities in neurons and their target regions (Arenkiel et al., 2004). While the precise roles of Hox genes in the neural crest are not fully resolved, recent analysis of HoxA and HoxB cluster conditional mutants, indicate a critical cell autonomous function in patterning of crest derivatives (Vieux-Rochas et al., 2013).

Hox genes in spinal MN specification

Unlike the relatively nuclear organization of MNs in the hindbrain, spinal MNs are organized into longitudinally arrayed columns and pools which span multiple segments (Figure 1C) (Landmesser, 2001). The Phrenic Motor Column (PMC), which is unique to mammals, is generated at rostral cervical levels and innervates the diaphragm muscle. At lower cervical (brachial) and lumbar levels, MNs of the lateral motor column (LMC) project axons towards the limbs, while preganglionic (PGC) and hypaxial (HMC) neurons at thoracic levels innervate the sympathetic ganglia and hypaxial muscles respectively. Finally, medial motor (MMC) neurons are found at all rostrocaudal levels and project dorsally to axial muscles. Forelimb and hindlimb LMC neurons further segregate in ~50 motor pools targeting specific muscles, and thus comprise a highly diverse MN population.

Like hindbrain MNs, the acquisition of columnar identity in the spinal cord can require the activity of a singular or multiple Hox genes (Table 1, Figure 2B). The most dramatic effect of a single Hox gene mutation is observed in Hoxc9 mutant mice in which thoracic motor columns acquire a brachial identity (Jung et al., 2010). Hoxc9 mutants lack both PGC and HMC neurons and the brachial LMC extends to the anterior boundary of the lumbar LMC. This is likely due to the predominant expression of Hoxc9 at thoracic levels and the lack of compensation by other paralog genes. The ability of a single Hox gene to determine neuronal identity in the thoracic region also reflects the low density of expressed Hox genes in this region relative to limb levels (Figure 1C) (Dasen et al., 2005). A more restricted role for Hox proteins is evident for Hox5 paralogs during the specification of phrenic MNs. While other Hox genes, such as Hoxc6, play an auxiliary role in setting the PMC boundaries, the absence of Hox5 genes results in a severe defect in diaphragm innervation and a dramatic reduction and disorganization of the PMC (Philippidou et al., 2012). As a result, Hox5 mutants perish at birth due to respiratory failure.

Hierarchical roles of Hox genes in spinal LMC and pool specification

The contributions of Hox genes to LMC specification reflect the varying strategies through which they contribute to MN diversification. Although Hox6 genes are expressed by the majority of brachial LMC neurons, Hox6 mutants still maintain an LMC, although reduced in size. Several Hox5-Hox8 paralogs can confer an LMC identity to MNs when ectopically expressed at thoracic levels of the spinal cord in chick embryos, suggesting that the early columnar identity of limb-innervating MNs is determined by redundant Hox inputs (Lacombe et al., 2013). At lumbar levels, Hox10 genes are major determinants of LMC identity and different combinations of Hox10 mutant alleles exhibit defects in hindlimb innervation and MN survival (Lin and Carpenter, 2003; Shah et al., 2004; Wahba et al., 2001; Wu et al., 2008). These defects are more severe than those of single Hox10 gene mutants (Hostikka et al., 2009; Wu et al., 2008) indicating collaborative roles for Hox10 genes in the lumbar spinal cord. Interestingly, thoracic specification programs are still suppressed in the brachial spinal cord of Hox6 mutants (Lacombe et al., 2013) while Hox10 mutants show a transformation of lumbar MNs to thoracic fates (Wu et al., 2008), indicating distinct mechanisms for organizing motor columns at these levels (Figure 2B).

Despite the converging actions of multiple Hox genes to specify LMC fate, individual Hox genes are required to further diversify LMC neurons into motor pools targeting specific muscles in the limb. The profile of Hox protein expression by motor pools is established through repressive interactions between Hox genes shortly after MNs are generated (Figure 3D). Combinatorial expression of Hox genes defines discrete transcriptional profiles for each pool and contributes to their clustering and peripheral synaptic specificity (Dasen et al., 2005). For example, both Hoxc8 and Hoxc6 are required for the specification of Pea3+ pools in the caudal brachial spinal cord (Figure 2B). Hoxc8 promotes specification of motor pools in the caudal half of the LMC, and in Hoxc8 mutants the Pea3+ pool is dramatically reduced and mispositioned (Vermot et al., 2005). Hoxc6 mutants also show a decrease of Pea3+ MNs and a severe reduction in the arborization of the cutaneous maximus muscle (Lacombe et al., 2013). Disorganization of motor pools innervating forelimb distal muscles is also observed in Hoxc8 mutants (Tiret et al., 1998). Ectopic expression or depletion of Hox4-Hox8 genes in the brachial spinal cord of chicks indicate that combinatorial Hox activity is critical for the establishment of pool fates (Dasen et al., 2005). Collectively, studies of Hox gene function in spinal cord reveal varying degrees of redundancy at the level of columnar identity, but highly specific roles in motor pool specification.

Figure 3. Regulation of Hox Gene Expression in the CNS.

(A) Gradients of FGF and RA establish initial patterns of Hox gene expression in the early embryo. Regions of RA signaling are inferred from expression of the Raldh2 gene in somitic mesoderm. Color coding of Hox genes denotes paralog groups regulated by indicated morphogens. RA induces primarily Hox1-Hox5 genes, FGF Hox6-Hox9, and Gdf11/FGF8 Hox10-Hox11 genes.

(B) Genes within a Hox cluster are sequentially activated along the rostrocaudal axis in a manner that is spatially and temporally linked to their chromosomal position. Colinear activation is linked to removal of repressive chromatin marks (H3K27me3). Figure on right shows Hox expression is nested in more caudal regions of the embryo.

(C) Anterior limits of Hox gene expression are established and maintained through the actions of Polycomb repressive complexes (PRCs). In stem cells, Hox genes are repressed by PRCs. At the progenitor phase RA and FGF act to clear PRC2 associated methylation marks from Hox genes. PRC1 activities are required to maintain Hox gene repression in postmitotic cells. PRC1 function may be distinct in MN progenitors (pMNs, indicated in gray) as depletion of PRC1-Bmi1 only affects Hox expression in post-mitotic MNs (Golden and Dasen, 2012). Cross-repressive effects of Hox genes are also indicated.

(D) Cross-repressive interactions of Hox genes define rostrocaudal boundaries and contribute to MN diversification. At thoracic levels Hoxc9 excludes the expression of multiple brachially-expressed Hox genes. At brachial levels, repressive interactions between multiple Hox genes determine motor pool fates. Motor pool identity appears to be specified prior to clustering, and individual MNs within a pool are indicated in yellow, green, and red.

(E) Feed-forward and auto-regulatory interactions amongst Hox1 and Hox2 paralogs ensure restricted expression of Hoxb1 in r4. Retinoic acid receptor (RAR) mediates the activation of Hoxa1 and Hoxb1. Hoxb1 is maintained through autoregulation.

Hox genes and MN generation in the hindbrain and spinal cord

While Hox genes have central roles in MN subtype specification in the hindbrain and spinal cord, their mode of action differs in each region. The acquisition of basic features of MN identity appears to be Hox-independent in the spinal cord, established primarily as a result of dorsoventral patterning systems (reviewed in Jessell, 2000). To date, no Hox mutants have been reported in which spinal MN progenitors are affected, although Hox mutations may affect the survival or relative numbers of MNs present at brachial and thoracic levels (Dasen et al., 2008; Jung et al., 2010; Tiret et al., 1998). In contrast, manipulating Hox expression in the hindbrain can alter the distribution of neuronal classes specified along the dorsoventral axis (Davenne et al., 1999; Gaufo et al., 2000; Pattyn et al., 2003). In Hoxa3/b3 compound mutants there is a reduction of the Olig2+ MN progenitor zone and an expansion of the V2 interneuron progenitor domain (Gaufo et al., 2003). Hoxa3 overexpression in r1-r4 results in the generation of somatic MNs at the expense of V2 interneurons (Guidato et al., 2003) and both Hoxa2 and Hoxb1 can produce ectopic branchiomotor neurons when overexpressed in r1 (Jungbluth et al., 1999). It therefore appears that Hox genes in the hindbrain can impinge on dorsoventral fate specification programs, while no such role has yet been described in the spinal cord.

The basis for this difference may lie in the distinct temporal and spatial profiles of Hox genes in each region. In the hindbrain, Hox proteins are present in neural progenitors and become restricted within dorsoventral domains of specific rhombomeres (Davenne et al., 1999). This dynamic pattern allows Hox genes to impinge on multiple fate decisions and provides an additional layer for diversifying neuronal populations (Davenne et al., 1999; Gaufo et al., 2000). Sustained expression of Hoxb1 in r4 mediates the prolonged generation of visceral motor neurons from a progenitor domain that switches to producing serotonergic neurons in all other rhombomeres, resulting in variations of neuronal populations within the hindbrain (Pattyn et al., 2003). In the spinal cord, Hox protein expression is predominant in postmitotic neurons, likely precluding them from influencing dorsoventral signaling pathways (Dasen et al., 2003). This difference may ultimately reflect the increased complexity of neural networks residing in the brainstem and the need for additional diversification strategies.

Hox genes and specification of non-MN populations in the nervous system

While the study of Hox genes in MNs has provided a valuable system to discern their functions, it is becoming apparent that other CNS populations rely on Hox genes for their specification (Table 1). In the hindbrain, Hox mutants also display defects in the formation of sensory ganglia. Hoxb1 and Hoxa3/b3 are required for the specification of first-order visceral sensory neurons in r4 and r5, respectively, while Hoxa2 loss-of-function results in complete elimination of somatic sensory neurons in r2 and a severe reduction in r3 (Gaufo et al., 2004). Serotonergic neuron generation also relies on Hox-dependent programs (Pattyn et al., 2003) while Hox2 paralog genes control oligodendrocyte production (Miguez et al., 2012). At spinal levels, Hoxb8 has been implicated in the organization of dorsal horn neurons that relay nociceptive stimuli at lumbar levels, survival of the second spinal ganglion and the specification of noradrenergic sympathetic neurons of the autonomic nervous system (Holstege et al., 2008; Huber et al., 2012; van den Akker et al., 1999) while Hoxb13 acts to define the caudal boundary of the spinal cord (Economides et al., 2003).

Thus, analogous to their function in body patterning along the rostrocaudal axis, Hox genes act to establish segmental boundaries and regional identity within the developing hindbrain and spinal cord. A failure to establish the correct pattern of Hox protein expression in the nervous system results in changes in neuronal identity that ultimately lead to defects in axon guidance and circuit formation.

Orchestrating Hox expression in the CNS

Neuronal subtype specification in the hindbrain and spinal cord relies on Hox-dependent regionalization of progenitor and postmitotic cells along the rostrocaudal axis. Understanding how Hox gene profiles are established and maintained in the CNS is therefore critical in revealing how neural circuits are organized. Hox expression is both temporally and spatially dynamic during CNS development, involving mechanisms that are shared and distinct between the hindbrain and spinal cord.

In general, the pattern of Hox gene expression along the rostrocaudal axis is directly correlated with its position within the cluster, a principle termed spatial colinearity (Kmita and Duboule, 2003). Hox genes located at the 3’ end of a cluster are activated earlier and at more rostral levels of the neuraxis, while 5’ genes are activated later and more caudally (Figure 3A–3B). Sequential activation of vertebrate Hox genes contrasts with the initiation of Hox gene expression in Drosophila, where segments form essentially in unison, and Hox genes are activated through the actions of segmentation transcription factors (Gellon and McGinnis, 1998). Nevertheless, in both vertebrates and invertebrates, the pattern of Hox expression along the rostrocaudal axis is linked to its position within a cluster, indicating that spatial colinearity is conserved amongst diverse species.

In broad terms, the establishment of Hox expression in the CNS and other tissues is defined over multiple temporally distinct phases (Figure 3). Induction of Hox gene expression occurs during axis extension, as stem cell-like populations emerge from the node and generate neuronal progenitors. Growth of the tail bud is associated with the progressive removal of repressive chromatin marks from Hox loci, and the appearance of chromatin marks indicative of gene activation (Soshnikova and Duboule, 2009). The sequential activation of genes within a Hox cluster is mediated by morphogens acting in a graded manner along the rostrocaudal axis. This initial pattern of Hox gene expression is subsequently modified through auto- and cross-regulatory interactions between Hox proteins and Hox genes. In general, while the initial inductive phase involves the actions of morphogens in neural progenitors, refinement and maintenance of Hox patterns occurs at or near the time neurons become postmitotic. Below we review the mechanisms associated with each of these phases of Hox gene regulation in the hindbrain and spinal cord.

Morphogens and Hox gene colinearity

As with patterning along the dorsoventral axis of the neural tube, the initial profile of Hox gene expression involves the activities of secreted morphogens acting on neural progenitors in a graded fashion (Figure 3A). The signaling pathways regulating Hox expression are linked to the same patterning cues involved in neural induction, and many of the signaling molecules involved in rostrocaudal patterning, notably Retinoic Acid (RA) and Fibroblast Growth Factors (FGFs), also play key roles in establishing progenitor identity along the dorsoventral (DV) axis (Diez del Corral et al., 2003; Novitch et al., 2003). However, while DV patterning systems activate expression of transcription factors that are largely restricted to neural progenitors, Hox genes are expressed in both progenitors and postmitotic cells, and are subject to distinct modes of regulation within these two cellular states.

During axis elongation, progenitor cells are exposed to graded levels of signaling molecules leading to the progressive activation of Hox genes located at more distal parts of the cluster. RA and FGFs exert central roles in patterning initial Hox expression along the neuraxis, with additional signaling systems acting to modulate these profiles (Figure 3A–3B) (Bel-Vialar et al., 2002; Liu et al., 2001). RA provided by somites adjacent to the neural tube patterns the caudal hindbrain and rostral spinal cord. The role of RA in Hox regulation has been most extensively studied in the hindbrain where it acts as a posteriorizing signal. Exposure of hindbrain progenitors to elevated RA in chick leads to an expansion of caudal rhombomeres at the expense of rostral, while inhibition of RA expands rostral and depletes caudal rhombomeres (Marshall et al., 1992). Depletion of RA signaling, through a mutation in the gene encoding the RA-synthesizing enzyme retinaldehyde dehydrogenase-2 (Raldh2), causes a loss of caudal Hox gene expression and rhombomere identity (Niederreither et al., 2000). RA promotes the expression of Hox genes through direct binding of retinoic acid receptors to regulatory elements in Hox genes, which have been characterized in Hox1 and Hox4 genes (reviewed in Alexander et al., 2009). RA also has an important role in patterning Hox expression in the spinal cord where it regulates expression of Hox genes associated with rostral cervical levels (Liu et al., 2001).

FGF signaling has a key role in establishing the patterns of Hox4-Hox10 gene expression in the spinal cord (Figure 3A–3B). Studies in chick and embryonic stem cell-derived neuronal progenitors have shown that increasing the levels of FGF can induce Hox genes with a progressively more posterior character (Liu et al., 2001; Peljto et al., 2010). Similarly, elevation of FGF signaling in vivo induces a rostral shift of Hox expression and transforms the identities of MN subtypes to a more caudal fate (Bel-Vialar et al., 2002; Dasen et al., 2003; Dasen et al., 2005). The effects of FGF expression in the spinal cord are mediated by Cdx homeodomain factors, as FGF can induce Cdx expression, and Cdx proteins are sufficient to induce expression of caudal Hox genes in the rostral neural tube (Bel-Vialar et al., 2002). Moreover, depletion of Cdx proteins in zebrafish confers a hindbrain identity to the spinal cord (Skromne et al., 2007), indicating that the FGF-Cdx-Hox network has a general role in distinguishing hindbrain from spinal cord neuronal identity.

FGFs also function in concert with other signaling systems to orchestrate patterns of Hox expression in the neural tube (Figure 3A). At rostral levels FGF acts with RA to establish expression of Hox6-Hox8 genes in brachial MNs. At more posterior levels, FGFs act with growth differentiation factor-11 (Gdf11) to initiate expression of Hox10 genes at lumbar levels (Liu et al., 2001). Wnt signaling also has an obligate role in the regulation of Hox induction, where it specifies spinal identity and the responsiveness of progenitors to RA and FGF (Nordstrom et al., 2006). The establishment of graded signaling systems also relies on interactions between its primary components. For example, FGF is capable of repressing expression of Raldh2, thus contributing to establishing the rostrocaudal gradient of RA signaling (Diez del Corral and Storey, 2004).

Post-transcriptional regulation of Hox gene expression

The pattern of Hox gene expression induced by morphogens in neural progenitors is characterized by well-defined anterior boundaries, with posterior expression that often extends to the tail bud (Figure 3B). Thus there is extensive overlap in Hox gene expression in progenitors at caudal levels. The specific activity of Hox genes in regions of nested expression has been argued to be facilitated by the ability of the more posterior Hox gene to suppress the activities of the anterior; a phenomenon termed posterior dominance/prevalence (Duboule and Morata, 1994). While Hox genes are transcribed in spinal progenitors, in many cases Hox proteins are not observed, becoming detectable only at the time neurons differentiate (Dasen et al., 2003). Thus the significance of posterior dominance in spinal progenitors is uncertain. While the mechanisms underlying the delay between Hox transcription and translation are not known, several studies indicate that Hox genes are post-transcriptionally regulated by both miRNAs and through translational control (Kondrashov et al., 2011; Yekta et al., 2004). These actions may serve to allow morphogens to prefigure Hox transcription but prevent Hox proteins from precociously activating genes that are not necessary until neurons differentiate.

Cross-regulatory interactions and the establishment of Hox boundaries

Although nested patterns of Hox expression are observed in neural progenitors, at the time of differentiation clear posterior boundaries become apparent. The establishment of caudal boundaries has been best studied in the context of spinal MNs, where posterior limits of Hox expression coincide with the position of specific columnar and pool subtypes (Figure 3C). For example, the posterior boundary of Hoxc6 and Hoxc9 demarks the caudal limit of the forelimb LMC and thoracic PGC neurons, respectively, and the positional boundaries of forelimb MN pools are similarly defined by Hox expression (Dasen et al., 2003; Dasen et al., 2005). Boundary formation is a consequence of repressive effects of Hox proteins on Hox genes (Figure 3D). The mechanisms mediating Hox repressive interactions have been studied in detail for the Hoxc9 protein, and are facilitated by direct interactions between Hoxc9 and Hox genomic sequences (Jung et al., 2010). In Hoxc9 mutants, multiple genes in the Hox4-Hox8 paralog groups are derepressed at thoracic levels leading to the transformation of PGC neurons to an LMC fate (Jung et al., 2010). Within LMC neurons, cross repressive interactions amongst Hox genes contribute to the intrasegmental diversification of motor pools (Figure 3D) (Dasen et al., 2005). Similar cross-regulatory interactions appear to operate in the hindbrain, where Hox3 paralogs are required for excluding Hoxb1 expression from r6 (Gaufo et al., 2003). However, the hindbrain appears to be less reliant on Hox cross-repressive interactions, possibly reflecting a greater dependence on molecular-based boundaries in the spinal cord, in the absence of overt segmentation.

In the hindbrain feed-forward and autoregulatory mechanisms also act to refine and maintain Hox expression, exemplified by the regulatory network that specifies the identity of rhombomere 4 (Figure 3E). Hoxa1 and Hoxb1 are induced in response to RA signaling prior to segmentation (Dupe et al., 1997; Studer et al., 1998), and Hoxa1 is required to maintain expression of Hoxb1 in the presumptive r4 territory (Carpenter et al., 1993). This initial pattern is reinforced through an autoregulatory enhancer in the Hoxb1 gene, which consolidates the initial Hox input into stable Hoxb1 expression (Popperl et al., 1995). The pattern of Hoxb1 is further constrained to r4 through inhibitory interactions mediated by Krox20, which represses Hoxb1 in the adjacent rhombomeres (Garcia-Dominguez et al., 2006). Hoxb1 is subsequently required to maintain expression of Hox2 paralogs in r4. The combination of positive, inhibitory, and feedforward inputs is likely to be typical of the regulatory networks that confine Hox gene expression in the CNS.

Polycomb proteins and the refinement and maintenance of Hox boundaries

Classic studies in Drosophila indicate that the maintenance of Hox segmental boundaries is mediated by members of the Polycomb group family. Because of their critical functions in embryonic and neural stem cells, the mechanisms of Polycomb action have been intensely studied (Schuettengruber and Cavalli, 2009). In the CNS Polycomb proteins appear to be needed for both exclusion of Hox expression at specific rostrocaudal positions and maintained repression in differentiated cells. In both ES cells and embryos, Hox clusters are characterized by a broad distribution of the repressive histone mark H3K27me3 (trimethyl-lysine-27 on histone H3), a chromatin modification associated with Polycomb repressive complex (PRC) activities (Figure 3B–3C). During tail bud extension H3K27me3 marks are removed from Hox clusters (Soshnikova and Duboule, 2009). Activation of Hox genes along the rostrocaudal axis correlates with the presence of the H3K4me3 chromatin mark deposited by the Trithorax complex. While it is assumed that removal of repressive marks and gain of activation marks is an obligate step in Hox gene activation, the precise role of these modifications at Hox loci in neural progenitors is yet to be resolved.

In vertebrates the Polycomb group encompasses a highly diverse collection of proteins with a variety of subunit compositions. Polycomb proteins form two distinct complexes; PRC2 which deposits the H3K27me3 mark, and PRC1 which recognizes H3K27me3 and mediates repression through ubiquitin ligase activity and chromatin compaction. Recent studies have challenged this canonical hierarchical view of PRC function. In stem cells lacking subunits required for the enzymatic activity of PRC2, PRC1 subunits still localize to Hox loci, suggesting PRC1 can be recruited independently of H3K27me3 marks (Tavares et al., 2012). The subunit composition of PRC1 is also highly diverse, and at least six PRC1 complexes have been recognized, based on differential incorporation of Polycomb group RING finger (PCGF) family proteins (Gao et al., 2012). Some of these alternate PRC1 complexes lack the Cbx subunit which recognizes H3K27me3, and localize to Hox loci in cell culture models. Further investigation into the developmental roles of PRC proteins will be required to fully resolve how this network controls Hox gene expression in the CNS.

Recent studies have assessed the roles of PRC2 and PRC1 in CNS development. The enzymatic component of PRC2, Ezh2, is necessary for maintaining Hox expression in the hindbrain. Neural-specific Ezh2 mutants are characterized by defects in pontine neuronal migration, partly as a consequence of aberrant Hox gene expression (Di Meglio et al., 2013). In contrast, removal of Ezh2 from MN progenitors has no noticeable effect of Hox expression in the spinal cord (Golden and Dasen, 2012), suggesting a distinct strategy for PRC-mediated Hox repression. While the overall pattern of PRC2-mediated H3K27me3 occupancy at Hox loci appears to be determined at the progenitor phase, PRC1 function has been shown to be critical for maintaining Hox boundaries in postmitotic cells (Golden and Dasen, 2012). Depletion of the PRC1 component Bmi1 (PCGF4) at forelimb levels leads to ectopic expression of Hoxc9, and the conversion of LMC neurons to a thoracic PGC fate. Conversely, elevation of Bmi1 represses Hoxc9 at thoracic levels and converts PGC neurons to an LMC fate. These observations suggest that Hox repression may be maintained in MNs by distinct PRC1 activity levels along the rostrocaudal axis.

Given the importance of chromatin modifications and secreted morphogens in regulating Hox expression, what is the relationship between these pathways? A recent study in ES cell-derived MNs indicates a direct link; treating cells with RA depletes PRC2-associated marks from rostral Hox genes, while FGF in combination with Cdx2 can remove H3K27me3 from caudal Hox genes (Mazzoni et al, 2013). In contrast, studies in chick embryos suggest that PRC1 acts independently of the FGF8-mediated effects on chromatin status. Both elevation of FGF8 signaling and depletion of Bmi1 at brachial levels leads to identical molecular phenotypes: ectopic expression of Hoxc9, loss of Hoxc6, and a conversion of LMC neurons to a PGC fate (Dasen et al., 2003; Golden and Dasen, 2012). However, while FGF8 induces Hoxc9 in progenitors and postmitotic MNs, depletion of Bmi1 derepresses Hoxc9 in early postmitotic MNs, indicating that PRC1 functions in differentiated cells to maintain appropriate Hox expression patterns (Figure 3C). Collectively, these studies in stem cell and embryonic MNs indicate that PRC2 and PRC1 act at distinct phases to establish and maintain the chromatin landscape of Hox loci.

Maintenance of Hox expression in the nervous system

Connections in the nervous system are established over the course of embryonic and postnatal development, raising the question of how long Hox actions are required within neurons. Arguably, the need for maintaining Hox expression in the CNS will rely on whether direct Hox effectors are continuously expressed and whether late expressed cell determinants require Hox function for their activities. In forelimb-innervating LMC neurons, Hoxc6 is only transiently expressed suggesting it only needs to be maintained over the short window necessary for it to deploy its columnar and pool program (Lacombe et al., 2013). In respiratory PMC neurons, Hox5 protein expression is extended to late embryonic phases indicating that its maintenance is required throughout phrenic MN ontogeny (Philippidou et al., 2012). In the hindbrain, Hox genes are expressed up to early postnatal stages in several nuclei including the ventral cochlear nucleus and the superior olivary complex (Geisen et al., 2008; Narita and Rijli, 2009). Thus the temporal profile of Hox gene expression is likely to be as important as its spatial profile for Hox function. Understanding the significance of Hox temporal regulation will require clearer knowledge about the specific pathways downstream of Hox proteins.

Mechanisms of Hox protein function in the CNS

While significant progress has been made in defining the regulation of Hox gene expression in the CNS, the mechanisms by which they deploy cell type specific gene programs are less well understood. Attempts to define the specificity of Hox protein function have been thwarted by two major challenges. First, Hox proteins contain conserved DNA binding domains that recognize very similar binding motifs (Noyes et al., 2008). Secondly, Hox proteins are broadly distributed amongst multiple classes of neurons, raising the question of how their cell type specificity is achieved. Recent work indicates that much of the specificity of Hox gene function is conferred by the cofactors they associate with.

Hox specificity can be conferred through TALE cofactor interactions

Hox proteins bind AT-rich hexamer sequences through homeodomains which are conserved among Hox paralogs (Figure 4A) (Gehring et al., 1994; Noyes et al., 2008). Thus, it has been difficult to determine how different Hox proteins accomplish specific functions, given the low selectivity in their binding motif. Hox proteins display weak affinity for DNA in vitro and high affinity binding typically requires cooperative interaction with the TALE (three amino acid loop extension) class homeodomain proteins Pbx and Meis (reviewed in Mann et al., 2009; Moens and Selleri, 2006). Cooperative binding of Hox/TALE proteins expands the size of the DNA recognition sequence, therefore limiting the number of potential Hox targets (Figure 4A–4B). Interactions between Hox proteins and TALE cofactors additionally enhance the binding selectivity of Hox paralogs, likely by altering their structure to facilitate contact of the N-terminal part of the homeodomain with specific DNA binding sites (Joshi et al., 2007; Slattery et al., 2011). The divergence of Hox sequences outside of the homeodomain may likewise influence Hox specificity through interactions with additional cofactors and collaborators.

Figure 4. Binding Specificity and Cofactors Mediate Hox Activity in the CNS.

(A) Binding of Hox proteins to Pbx cofactors enhances binding specificity to target sequences. Binding preferences for the indicated Hox paralogs groups are inferred from Drosophila homologs (Slattery et al., 2011). Interaction of Meis proteins with Hox/Pbx dimers can occur in the absence of Meis binding sites.

(B) Cofactor interactions determine the specificity of transcriptional outputs in Hox target gene regulation. In Drosophila, Engrailed (En) allows the Hox protein AbdA to repress the distalless (Dll) gene. Foxp1 appears to act as a collaborative factor for many LMC-specific genes. The Foxp1/Hox interactions are speculative. At the Hoxb1 gene, Meis proteins displace corepressors from Hox/Pbx dimers and recruit coactivators.

(C) Hox proteins display distinct specificities in determining MN subtype identities.

Due to their role as cofactors, TALE gene mutants often exhibit phenotypes similar to Hox mutants in the CNS. Mutations in zebrafish Pbx genes result in the absence of hindbrain segmentation and rhombomere specification (Popperl et al., 2000; Waskiewicz et al., 2002). In Pbx2/Pbx4 zebrafish mutants the hindbrain reverts to the r1 “Hoxless” ground state, suggesting that all Hox function in hindbrain patterning is mediated through Pbx proteins (Waskiewicz et al., 2002). However, mouse single Pbx mutants do not exhibit gross patterning defects in the hindbrain, which could be a reflection of more extensive redundancy among the different paralogs. Pbx4 zebrafish mutants also show defects in the migration of the facial nucleus and the axonal pathfinding of the trigeminal nerve, although some of these phenotypes may be non-cell-autonomous (Cooper et al., 2003). In mice, Pbx3 mutation leads to perinatal death due to central respiratory failure (Rhee et al., 2004). While the phenotype of Pbx3−/− mice can be partially explained by its interaction with the Tlx homeodomain transcription factor, Rnx (Shirasawa et al., 2000), the involvement of multiple Hox genes in respiratory circuit formation may be indicative of multiple roles for Pbx cofactors in this process.

Meis cofactors have also been implicated in hindbrain patterning and segmentation in zebrafish (Choe et al., 2002; Deflorian et al., 2004; Vlachakis et al., 2001; Waskiewicz et al., 2001), although CNS defects in Meis mouse mutants have not been reported (Azcoitia et al., 2005; Hisa et al., 2004). Interestingly an important function of Meis proteins in the hindbrain is to displace corepressors from Pbx proteins and recruit coactivators (Figure 4B) (Choe et al., 2009). Therefore, it is possible that the identity of the TALE cofactor incorporated into a Hox complex may play an additional role in gating its activity. In mice there are 4 Pbx and 5 Meis homologs and whether these have different affinities for different Hox proteins or other cofactors remains to be determined.

Not all Hox binding to target sequences is dependent on TALE cofactor interactions, and it appears that there is a bias towards repressing, as opposed to activating, gene targets when these cofactors are absent (Mann et al., 2009). Studies in Drosophila indicate that monomeric binding of Hox proteins represses transcription (Galant et al., 2002). Consistent with this idea, mutation of the Pbx-interaction motif in the Hoxc6 protein eliminates its cooperativity with Pbx3 but preserves its ability to repress the Hoxc9 gene (Lacombe et al., 2013). However Hox/TALE interactions can also mediate repression of target genes. For example, the Drosophila homeodomain protein Engrailed (En) is a Hox cofactor that mediates repression of the target gene Distalless in a TALE-interaction dependent manner (Figure 4B) (Gebelein et al., 2004). These observations indicate that the output of Hox/TALE interactions may rely on the additional cofactors it interacts with, rather than intrinsic transcriptional activities. It should also be noted that TALE proteins can play roles independent of Hox proteins, exemplified by the role of Drosophila Homothorax in antennae development, a tissue lacking any Hox gene expression (Casares and Mann, 1998).

Hox activities shared and unique amongst paralogs

A common theme from studies of Hox specificity in the hindbrain and spinal cord is the regulation of both converging and diverging pathways by Hox proteins. Several Hox functions appear to be shared between multiple Hox proteins, most frequently paralogs, while others are unique to a single Hox protein (Figure 4C). For example, in spinal MNs multiple genes in Hox5-Hox8 paralog groups contribute to LMC identity at brachial levels, while Hoxc9 uniquely confers thoracic PGC identity (Jung et al., 2010; Lacombe et al., 2013). Among paralog groups, Hoxa5 and Hoxc5 cooperate to control PMC development (Philippidou et al., 2012). In the hindbrain, expressing Hoxa1 from the Hoxb1 locus can rescue facial nerve defects in Hoxb1 mutant mice, despite diverse functions of the two proteins in vivo (Tvrdik and Capecchi, 2006). Conversely, individual Hox proteins, such as Hoxc6 and Hoxc8 control unique aspects of MN pool identities (Dasen et al., 2005; Lacombe et al., 2013).

At the level of target gene regulation, Hox proteins collectively determine the migration, survival and guidance of neuronal subtypes, possibly by activating common or related downstream targets. The acquirement of different identities and trajectories upon ectopic expression of Hox proteins indicates there is also divergence of targets that confer specific neuronal identity. It would then appear that multiple Hox proteins could share common effectors but also that each Hox protein would have a unique set of targets to further fine-tune neuronal identity. The convergence of Hox proteins on a common set of targets could also provide a mechanism through which variations in Hox protein level or activity could determine the relative expression of a particular gene. In MN axon guidance for example, the relative rather than absolute expression levels of guidance receptors such as RET, GFRa1 and ephrins appears to be important for correct pathfinding (Bonanomi et al., 2012). Ephrins are among the known Hox target genes and therefore it is plausible that differential efficacies of Hox-mediated gene activation may regulate ephrin levels.

Effectors of Hox protein activities in the CNS

While the pathways acting downstream of Hox genes are at present not well-defined, several studies in the hindbrain and spinal cord indicate that Hox proteins deploy a variety of intermediate transcription factors that in turn activate signaling pathways which contribute to neuronal connectivity. In some contexts Hox proteins may bypass intermediate targets and directly activate specificity determinants.

Transcription factors functioning downstream of Hox proteins

Several factors acting downstream of Hox genes have been identified that mediate subsets of Hox functions in the CNS (Figure 5). In the hindbrain, mutations in Hox1 and Hox2 genes affect the expression of factors involved in cell type specification, such as Phox2b, Nkx proteins and Pax6 (Davenne et al., 1999; Gaufo et al., 2000; Pattyn et al., 2003). Phox2b, a determinant of cranial bm and vm neuron specification, is a direct target of Hoxb1 and Hoxb2 (Samad et al., 2004). GATA2 and GATA3 act downstream of Hoxb1 to control facial MN migration and contralateral vestibuloacoustic efferent neuron projections in r4 (Pata et al., 1999). The interneuron determinant Evx1 is regulated by Hox2 paralogs while Hox3 genes confer somatic MN identity by upregulating Olig2 and Hb9 (Davenne et al., 1999; Gaufo et al., 2003; Guidato et al., 2003).

Figure 5. Hox Effectors Control Aspects of Subtype-Specific Neuronal Identity.

(A) LMC neurons are disorganized and project haphazardly in Foxp1−/− mice, due to the loss of adhesion and guidance molecules, such as EphA4.

(B) Hoxc6−/− and Hoxc8−/− mice have a reduced and disorganized Pea3+ pool, due to loss of cadherin expression.

(C) MN-specific Hox5 deletion results in defects in multiple aspects of PMC development, likely due to downregulation of target effectors such as ALCAM and PTN.

(D) Tangentially migrating pontine neurons express Robo2 which prevents premature ventral migration by responding to Slit2/3 repulsive signals secreted from the facial motor nucleus (VII). In the absence of Hox2 genes, both Robo2 and Slit2/3 are downregulated resulting in abnormal pontine neuron migration.

(E) Hoxa2 inactivation perturbs anteroventral cochlear neuron (AVCN) axonal pathfinding to the medial nucleus of the trapezoid body (MNTB) in the superior olive, resulting in decreased contralateral and increased ipsilateral targeting of MNTB due to the downregulation of Rig1/Robo3, the main axon guidance receptor required for midline crossing.

In the spinal cord, a major target effector of Hox proteins in MNs is the gene encoding the transcription factor Foxp1 (Dasen et al., 2008; Rousso et al., 2008). Limb-level Hox proteins induce high levels of Foxp1 to specify LMC fates, whereas Hoxc9 induces low levels of Foxp1 at thoracic levels to specify PGC fates. Hox-dependent regulation of Foxp1 levels determines columnar identity, as elevation of Foxp1 at thoracic levels can convert PGC and HMC neurons to an LMC fate. Once induced, Foxp1 also acts as an accessory factor for Hox genes and is required for the expression of all LMC motor pool determinants (Dasen et al., 2008). In Foxp1 mutants there is a loss of columnar and pool identities and a randomization of axonal projections and cell body position, likely reflecting loss of downstream effectors such as ephrins and cadherins (Figure 5A–5B). Within LMC neurons Hox proteins also coordinate the expression of multiple pool-restricted transcription factors. The transcription factors Nkx6.1, Pea3 and Scip are deployed downstream of Hox genes and act to specify facets of identity within LMC pools. In the absence of Nkx6.1, axons from lumbar MN pools fail to innervate their appropriate muscle targets (De Marco Garcia and Jessell, 2008), while Pea3 mutation leads to defects in MN clustering and intramuscular branching (Figure 5B) (Livet et al., 2002). The extent of cooperativity between Hox genes and these downstream factors also remains to be determined and will heavily depend on the identification of the final effector targets in these cascades.

While the diversification of LMC neurons relies on multiple Hox genes and downstream effectors, some inroads into deciphering target gene regulation can be made by analyzing populations that rely on a single Hox paralog group. The development of phrenic MNs is determined by the Hoxa5 and Hoxc5 genes (Philippidou et al., 2012). In Hox5 mutant mice multiple aspects of PMC identity are compromised, including cell body clustering, axon guidance, intramuscular branching and survival (Figure 5C). Late removal of Hox5 genes from postmitotic MNs demonstrated that prolonged Hox5 expression is necessary for maintenance of certain target genes, such as the trophic factor pleiotrophin (PTN). PTN is under direct regulation by Hoxa5 (Chen et al., 2005), indicating that Hox proteins can act to induce effector molecules directly and not exclusively through intermediate factors. PMC neurons also express the transcription factor Scip (Pou3f1); however expression of Scip appears to be incapable of conferring a PMC identity in the absence of Hox5 genes. Since Pou proteins have been shown to bind DNA collaboratively with Hox proteins (Di Rocco et al., 2001), it is possible that Hox5 and Scip have a synergistic rather than a linear relationship in regulating PMC specific genes.

Guidance, Adhesion, and Migratory Pathways regulated by Hox proteins

While Hox genes control a diverse array of cellular processes, there is evidence indicating they can achieve specific outcomes by regulating the same effector classes in different contexts. One class of Hox target effectors belong to the Eph/Ephrin receptor and ligand system. During early stages of hindbrain compartmentalization Eph/ephrin-mediated attractive and repulsive interactions act to maintain rhombomere boundaries, with receptors and ligands expressed in alternate rhombomeres. One Eph receptor, EphA2, is under direct regulation of Hox1 paralogs (Chen and Ruley, 1998) while EphA4 and EphA7 are downregulated in r2 and r3 of Hoxa2−/− mice respectively (Gavalas et al., 1997; Taneja et al., 1996), indicating that regulation of Eph receptors by Hox proteins contributes to rhombomere segregation. Hoxa2 may also employ the Eph/ephrin signaling system in defining topographic connectivity of the rostral principal (PrV) nucleus to the thalamus by regulating expression of EphA4 and EphA7 (Oury et al., 2006).

In spinal MNs, Eph/ephrin signaling has a critical role in the guidance of LMC axons, in particular their initial choice between a ventral and dorsal trajectory at the base of the limb bud. Hox proteins in LMC neurons regulate expression of Raldh2, which provides a MN-derived source of RA that induces expression of the transcription factor Lhx1 (Figure 5A). Lhx1 in turn stimulates EphA4 expression which is necessary for defining the dorsal trajectory of lateral LMC neurons (Kania and Jessell, 2003). While regulation of EphA4 in MNs is indirect, Hox genes could also have a more direct role, as multiple Eph and Ephrin family members are expressed and function in several MN subpopulations (Feng et al., 2000; Iwamasa et al., 1999).

Besides the ephrins, other axon guidance systems have been shown to be regulated by Hox proteins. The Robo/Slit receptor/ligand system mediates repulsive cell interactions. In Hox2 mutants, Robo2 is downregulated in migrating pontine neurons, contributing to their abnormal migration and premature attraction towards the midline (Figure 5D). Interestingly, in these mutants Slit2/3 are also downregulated in the facial motor nucleus, the source of repulsive cues, exacerbating the phenotype (Geisen et al., 2008). Hoxa2 also directly regulates the guidance receptor Rig1/Robo3 which controls the trajectory of a subset of cochlear neurons in the hindbrain (Figure 5E) (Di Bonito et al., 2013). Robo/Slit signaling contributes to establishing appropriate patterns of intramuscular branching of spinal MNs (Jaworski and Tessier-Lavigne, 2012), however whether this system is under Hox regulation in the spinal cord has not been explored. Another receptor mediating repulsive responses to netrin, Unc5b, appears to be repressed by Hox5 proteins during pontine neuron migration (Di Meglio et al., 2013).

The disorganization of motor columns and pools observed in Hox mutants suggests regulation of cell adhesion properties by Hox proteins. Expression of the cadherin family of adhesion molecules, which mediate cell body organization of spinal MNs, are lost in Foxp1−/− mice (Dasen et al., 2008; Demireva et al., 2011). In Hoxa1 mutants, cadherin6 expression is lost in r4 to r6 in the hindbrain, likely pointing to a role in rhombomere segregation (Inoue et al., 1997). Collectively these studies indicate that Hox genes have a significant role in shaping neuronal organization and synaptic specificity in the CNS, through regulating the expression of large families of guidance and cell adhesion molecules.

Hox Genes and the Diversity of Nervous Systems

The reliance on a Hox-based program for neuronal diversification presumably has allowed for a certain degree of flexibility in the ability of motor and sensory systems to adapt, as best evident in the variations in complexity and organization of vertebrate motor neurons (Fetcho, 1987, 1992). The prominent role of Hox genes in neuronal specification in vertebrates raises the question of whether they define a conserved mechanism for generating neuronal diversity along the rostrocaudal axis. There is emerging evidence that in addition to their more global roles in segmental patterning, Hox genes have essential roles in determining the identity of neuronal subtypes in invertebrates.

Hox genes in the development of the fly nervous system

Studies in the Drosophila CNS demonstrate that Hox genes determine both the segment-specific distribution and subtype identity of neural populations in the embryonic and larval CNS. In contrast to the vertebrate CNS, where progenitors typically give rise to a single or only a few distinct neuronal classes, Drosophila embryos use a lineage-based strategy to generate diversity. As neural stem cells (neuroblasts) asymmetrically divide they give rise to several classes of neuronal types, the identity of which depends on a temporal cascade of transcription factors which are transiently expressed in neuroblast lineages (reviewed in Skeath and Thor, 2003). The identity and distribution of these neurons depends in part on how temporally acting transcription factors intersect with rostrocaudal patterning cues.

The integration of temporal and rostrocaudal programs has been best studied in the specification of neuropeptide producing cells generated in abdominal and thoracic segments of the fly embryo (Figure 6A). For example the three thoracic segments (T1-T3) of the ventral nerve cord produce peptidergic neurons defined by the expression of the transcription factor Apterous. Generation of Apterous+ neurons relies on Hox activity, as these neurons are lost in mutants for the Hox gene Antp (Karlsson et al., 2010). Conversely misexpression of Antp in a mutant background lacking posterior Hox genes gives rise to Apterous neurons in all segments. The conversion requires coexpression of the “temporal” transcription factor Grainy head, underscoring the importance of integrating Hox activities with the temporal specification code.

Figure 6. Hox Genes Contribute to Neuronal Identity in the CNS of Drosophila.

(A) Hox genes are essential for the generation of several types of peptidergic neurons in the ventral nerve cord of the fly embryo. Hox expression is shown relative to thoracic (T1-T3) and abdominal (A1-A8) segments. Peptidergic neurons are identified by expression of Apterous (Ap) and Capability (Capa) as well as other markers in Va and dMP2 populations. The distribution of these neurons is affected in Hox mutant embryos.

(B) Graded Antp activities contribute to the innervation of leg muscles in the fly. Elevation or depletion of Antp affects the branching patterns of motor axons at proximal and distal regions of the femur.

Peptidergic neuron specification also depends on Hox-dependent programs that impact neuronal number, through termination of cell cycle progression or ablation of neurons that have already been generated. The absence of Apterous neurons in abdominal segments is not due to a cell fate specification program, but rather the fact that the three posterior Hox genes, Ubx, Abd-A and Abd-B, terminate cell cycle in the progenitors generating Apterous neurons. The Hox gene Abd-A contributes to the distribution of neurons within abdominal segments through induction of apoptosis (Bello et al., 2003), while Abd-B contributes to the specification of dMP2 neurons by preventing cell death (Miguel-Aliaga and Thor, 2004). Abd-B also promotes apoptosis of Va neurons to restrict their distribution to rostral abdominal segments (Suska et al., 2011). Thus Hox genes can pattern the nervous system through mechanisms that are not solely dependent on activation of a set of cell-type specific determinants.

Hox genes are also critical in the development of the neuromuscular network involved in locomotion. The embryonic nervous system of the fly generates motor circuits required for basic peristaltic movements that are distinct in specific rostrocaudal segments. Hox gene mutations have been shown to lead to transformation in these segment-specific patterns of motor activity (Dixit et al., 2008). Although flies and vertebrates use similar transcription factors to generate MNs as a class (Figure 7A) (Landgraf and Thor, 2006), there is little evidence that Hox genes contribute to MN subtype diversity in the fly embryonic nervous system. Thus the locomotor defects observed in Hox mutant embryos likely reflect alterations in other neuronal or non-neuronal cell types.

Figure 7. A Model for the Emergence of Motor Neuron Diversity in Vertebrates.

(A) Ancestral MNs were characterized by the expression of a core set of transcription factors including Hb9, Isl1/2, and Lhx3, which are conserved in MNs of many invertebrates.

(B) Agnathan species lacked paired appendages and likely contained MNs required for the innervation of dorsal and ventral muscles at all segments. For simplicity we indicate these populations as “MMC” and “HMC” neurons, although they likely lacked a columnar organization. The specification of ventrally projecting MNs may have required the exclusion of Lhx3, which defines dorsally projecting MMC neurons in tetrapods.

(C) Zebrafish pectoral fins are innervated by MN populations that appear to be generated in registry with Hox6 and Hox9 expression domains. MMC- and HMC-like MNs are not organized into columns.

(D) Tetrapod MNs display a columnar organization. Foxp1 likely played a key role in establishing both the columnar organization and multiple aspects of MN specification.

(E) In mammals Hox5 genes specify PMC MNs in a Foxp1-independent manner.

During the larval stage of fly development the nervous system generates additional neuronal populations to allow for the emergence of adult motor behaviors such as walking and flying. There is also evidence that Hox genes have important roles in the innervation of appendages such as the leg. Motor neurons innervating the leg are found in segments T1-T3, and within each segment nine neuroblast lineages give rise to ~50 MNs innervating 14 leg muscles (Baek and Mann, 2009; Truman et al., 2004). In contrast to vertebrates, the cell bodies of these MNs are not somatotopically organized, but rather their target specificity can be linked to birth order. One lineage, LinA gives rise to 28 of the 50 leg MNs. Early born neurons in LinA project their axons proximally in the leg, while later born neurons project distally, suggesting there is a temporal relationship between birth order and innervation pattern (Figure 6B).

Analyses of Hox gene mutants indicate an essential function in controlling the survival and identity of leg MNs (Figure 6B). Interestingly, Antp expression is graded in LinA MNs, with high levels found in late born MNs and decreasing levels in early born progeny (Baek et al., 2013). Genetic analysis indicates that Antp acts as dose-dependent determinant of MN connectivity. Loss of Antp leads to a reduction in axonal branches at distal muscles while elevation of Antp generates additional distal branches, and fewer proximal. These results indicate that graded actions of Hox proteins may influence the connectivity of neurons with target cells.

Evolution of vertebrate spinal MN diversity

While studies in fly suggest conservation in the Hox-based program for neuronal diversification, it is yet to be determined how this program is varied in related species that exhibit distinct motor behaviors, such as walking, flying and swimming. Comparisons of MN organization between modern species may shed light into how the vertebrate motor system evolved (Figure 7).

The common ancestors of tetrapods and fish lacked limbs, raising the question of how the Hox-dependent program for appendicular muscle innervation emerged in vertebrates. Cephalochordates (e.g. amphioxus) and agnathan vertebrates (e.g. lamprey) exhibit colinear expression of Hox genes in the spinal cord (Schubert et al., 2006; Takio et al., 2007) although they lack limbs and other regionally-restricted targets supplied by Hox-dependent MNs. Analyses in zebrafish provide some clues into the mechanisms through which the limb innervation program likely emerged. MNs innervating pectoral fins localize to both the hindbrain and spinal cord (Ma et al., 2010), suggesting that the program for limb innervation originated through a Hox-dependent program that was initially linked to movement of the head (Figure 7C). Despite the altered position of fin-innervating MNs relative to tetrapods, they still retain an alignment with Hox expression domains in the CNS, such that the anterior boundary of Hox9 expression defines the caudal extent of pectoral fin- innervating populations (Prince et al., 1998). Thus the boundary between limb-level Hox and Hox9 proteins appears to represent an early mechanism which segregated limb and non-limb MN identity. In contrast, MNs innervating the pelvic fin are not matched to Hox10 expression in the spinal cord, suggesting alternative Hox programs may be at work (Murata et al., 2010). Moreover, the discrete columnar grouping of MNs present in tetrapods has not been observed in fish species (Menelaou and McLean, 2012; Thorsen and Hale, 2007), indicating that if a Hox-dependent MN program is present, it is configured in such a way that only a subset of the programs present in tetrapods is activated.

Although forelimb and hindlimb LMC neurons are specified by distinct sets of Hox genes in tetrapods, they deploy nearly identical molecular programs initially; they express high levels of Foxp1, Raldh2, and display similar LIM-HD codes. A primary target of Hox gene activity at limb levels is the induction of the transcription factor Foxp1 (Figure 7D), which is required for all subsequent aspects of LMC and PGC neuronal differentiation (Dasen et al., 2008). In Foxp1 mutants MNs revert to a molecular identity similar to the hypaxial motor column (HMC) neurons that normally reside in thoracic spinal cord. These observations suggest that HMC neurons represent the ancestral MN population from which Hox-dependent MNs emerged, and was likely present at all rostrocaudal levels in more rudimentary vertebrates (Figure 7B).

Changes in Hox inputs to HMC-like MNs populations appear to have continued to contribute to MN diversity. The phrenic motor column is unique to mammals, as the diaphragm muscle is not present in other tetrapod classes such as birds, amphibians, and lizards (Figure 7D–7E). In the absence of Hox5 genes in mice PMC neurons are lost and animals perish due to respiratory failure (Philippidou et al., 2012). Interestingly, amongst brachially expressed Hox proteins, Hoxa5 shows a diminished ability for LMC induction (Lacombe et al., 2013), indicating that it may have evolved specifically to regulate PMC specific gene targets. In the absence of Foxp1, Hox5⁺ motor neurons at rostral cervical levels acquire molecular features and projection characteristics of PMC neurons. Thus, like LMC neurons, PMC neurons appear to have emerged from an HMC-like population that excluded LMC Hox determinants and acquired sensitivity to Hox5 genes.

Hox genes as substrates for adaptability in motor systems

How might the Hox-based system for generating MN subtypes have been used as an adaptive strategy in the vertebrate lineage? A highly varied attribute in vertebrates is the distance between the forelimb and the hindlimb, which is defined by the number of thoracic segments and can be as few as 4 in amphibians or as many as 300 in some snake species. One question is how LMC neurons are generated in registry with the position of the limbs. Genetic analysis in mice has revealed that while multiple Hox genes are involved in forelimb and hindlimb LMC specification, the single Hoxc9 gene specifies the position of thoracic MNs relative to the LMC (Jung et al., 2010). This strategy would in principle allow adaptability in the position of limb innervating MN populations through changing the expression of a single transcription factor. Changes in the pattern of Hoxc9 expression amongst vertebrates could therefore contribute to the alignment of motor columns with their peripheral targets.

Another important question is how the specification of Hox-dependent LMC pools has been implemented to articulate limb muscles in diverse species. Both forelimb and hindlimb LMC populations deploy an initially similar molecular program, but ultimately generate specific MN pools that are dedicated to innervating a single muscle in the limb (Dasen and Jessell, 2009). At this level the Hox-dependent program becomes more selective, and specific Hox proteins act by deploying pool-restricted programs such as activation of intermediate transcription factors. This idea is exemplified by the roles of Hoxc6 and Hoxc8, which while both capable of imposing an LMC identity to thoracic MNs, have distinct functions in specifying motor pools (Lacombe et al., 2013; Tiret et al., 1998; Vermot et al., 2005). Thus it appears that the Hox-dependent program of columnar differentiation uses a set of fairly permissive inputs early on (e.g activation of Foxp1), but more specific activities emerge during motor pool specification. When one considers the diversity of locomotor strategies, for example the use of the forelimb for walking versus wing muscles for flying, changes in Hox expression within LMC neurons could allow for alterations in the relative distribution of motor pools dedicated to innervating a muscle, without affecting their early columnar identity.

Hox genes and neural circuit formation

While connections between MNs and muscles are one of the first to be established during development, it is the subsequent connectivity with premotor interneurons and sensory neurons that drives the basic wiring of circuits in the hindbrain and spinal cord. Formation of these neuronal networks is perturbed in the absence of Hox genes (Figure 8), suggesting a more global role in circuit assembly.

Figure 8. Hox Genes in the Assembly of Neural Circuits.

(A) Multiple Hox genes mediate the assembly of respiratory networks in the hindbrain and spinal cord. The pontine respiratory group (PRG) relies on Hoxa2 expression, while Hoxa1 is involved in the development of the rhythmogenic parafacial respiratory group (pFRG) and Hox5 genes control phrenic MN identity.

(B) Hoxa2 is required for topographic projections from the trigeminal principal sensory nucleus (PrV) to the ventral posterior medial (VPM) nucleus of the thalamus. In the absence of Hoxa2, the maxillary (Mx) branch of the trigeminal nerve fails to arborize at the r3-derived domain of the PrV, which is ectopically targeted by the mandibular (Md) branch that normally innervates the r2-derived PrV. Projections from the r3-derived PrV are also mistargeted in the VPM and both the PrV and VPM lose their topographic organization.

(C) In the auditory system, r4-derived structures (red), such as the posteroventral cochlear nucleus (PVCN) and the ventral lateral lemniscus (VLL) form circuits that mediate sound perception while r2/r3 and r5-derived structures (green) such as the anteroventral cochlear nucleus (AVCN) and the superior olivary complex (SOC) are primarily involved in circuits encoding sound localization. Mutations in Hoxb1/Hoxb2 affect the sound perception pathway while Hoxa2 is involved in the wiring of the sound-localizing circuits. SOC is shown as one structure for simplicity.

(D) In mouse, but not chick, target-derived nerve growth factor (NGF) induces expression of Hoxd1 in nociceptors. In the absence of Hoxd1, both peripheral and central connectivity of mouse nociceptors is altered to resemble chick nociceptor properties.

Respiratory networks

The neural networks that control respiratory rhythm generation reside in the brainstem (Figure 8A). Two distinct nuclei, the Parafacial Respiratory group (pFRG/RTN) derived from r3-r4 and the pre-Bötzinger complex (pre-BötC), derived from caudal rhombomeres (r6-r8), are the primary respiratory rhythm generators in the medulla, and mutations affecting their development lead to perinatal death due to respiratory failure (Bouvier et al., 2010; Rose et al., 2009). The pre-BötC appears to be the dominant respiratory pattern generator, with FRG acting to entrain this rhythm and to initiate breathing during birth (Thoby-Brisson et al., 2009). Respiratory regions in the pons connect to and modulate medullary respiratory networks. These pontine respiratory structures are derived from rostral rhombomeres r1-r2. Hoxa2 inactivation leads to an expansion of r1 at the expense of r2 and an increase in inspiratory amplitude, while breathing frequency remains unaffected, consistent with a role of pontine circuits in respiratory functions other than rhythm generation (Chatonnet et al., 2007). This phenotype is distinct from that of Hoxa1 −/− mice which die shortly after birth from breathing defects (Carpenter et al., 1993; Chisaka et al., 1992; Lufkin et al., 1991; Mark et al., 1993). The Hoxa1 mutation does not affect the development of the caudally derived pre-BötC but these mice show a hypoplasia in the r4 region where the pFRG is located, uncovering a role of this nucleus in preventing apneas at birth. Further analysis of Hoxa1−/− mice revealed the integration of ectopic r3-r4 derived neurons into a functional rhythm generating network, demonstrating rewiring of neuronal circuits upon alterations in Hox gene expression (del Toro et al., 2001). Hox5 paralog genes control the development of the respiratory output phrenic MNs in the spinal cord (Philippidou et al., 2012), underscoring the importance of multiple Hox inputs in the wiring of respiratory circuits.

Formation of the somatosensory map

The transmission of sensory information from the periphery to cortical areas relies on high fidelity relay of sensory inputs to nuclei in the brainstem and thalamus. In the trigeminal pathway, sensory ganglia send projections to various facial areas and project centrally to distinct nuclei in the brainstem, such as the rostral principal (PrV) nucleus (Figure 8B). These nuclei in turn target nuclei in the thalamus, which connect to specific areas of the somatosensory cortex devoted to the representation of facial structures. There is a strict topographic organization of neuronal connections, such that each point in the periphery is mapped to a distinct area of the brainstem, thalamus and cortex. Hoxa2 has been shown to be a critical determinant in the formation of topographic somatosensory maps and acts at distinct phases to control multiple aspects of circuit connectivity (Oury et al., 2006). In the absence of Hoxa2, trigeminal sensory afferents inappropriately project to the cerebellum instead of terminating at the PrV nucleus. Late removal of Hoxa2 retains correct trigeminal pathfinding to the PrV, but the maxillary branch of the ganglion, carrying information from the whiskers, upper jaw and lip, fails to arborize. Examination of Hoxa2−/− mice at postnatal stages revealed that the topographic representation of inputs is eroded both in the PrV brainstem nucleus and in the thalamus and that PrV axons are mistargeted to other thalamic regions. The requirement for Hoxa2 at various stages of trigeminal circuit formation exemplifies the diverse temporal and spatial roles of Hox genes in CNS development.

Auditory circuits

An example of how multiple Hox genes can contribute to circuit formation comes from a recent study delineating the roles of Hoxb1, Hoxb2 and Hoxa2 in the assembly of auditory circuits (Figure 8C) (Di Bonito et al., 2013). Temporally controlled removal of these genes in the hindbrain bypassed early patterning defects and revealed novel roles in the specification and connectivity of auditory nuclei. While interpretation of Hox mutant phenotypes is confounded by the transcriptional interactions between these genes, the study sheds light on the temporally distinct steps of auditory circuit development. Afferent sensory fibers innervating inner and outer hair cells of the cochlea relay information to nuclei in the brainstem, which encompass the anteroventral (AVCN), posteroventral (PVCN) and dorsal (DCN) nuclei. Sensory information required for sound perception is transmitted from the cochlear nuclei through the lateral lemniscus complex and eventually reaches the auditory cortex through relay stations in the inferior colliculus of the midbrain and the medial geniculate nucleus of the thalamus. A parallel pathway emerging from cochlear nuclei and passing through nuclei of the superior olivary complex encodes sound localization.

The contribution of Hox paralogs to distinct circuit components aligns with their rhombomeric origin, with Hoxb1/b2 determining r4-derived structures and Hoxa2 being critical in r2-r3 structures. The PVCN is derived primarily from r4 while the AVCN is derived from r2/r3. In the absence of Hoxb1 or Hoxb2, r4-derived structures acquire an r2/r3-like identity, leading to abnormal specification and axonal targeting of PVCN neurons to AVCN targets. AVCN neurons normally project to a nucleus in the superior olive that is important for sound localization and deletion of Hoxa2 leads to defects in AVCN axon guidance, through downregulation of the guidance receptor Rig1/Robo3. The ventral lateral lemniscus, a target of PVCN neurons, is primarily derived from r4 and it is almost completely absent in Hoxb1 mutants. Hoxb1 and Hoxb2 are also required for the correct specification of olivocochlear motor neurons innervating cochlear hair cells. Absence of motor innervation in Hoxb1 and Hoxb2 mutant mice results in abnormal morphology and loss of cochlear hair cells, leading to an increase in auditory threshold and a hearing impairment in these mice in adulthood. The data collectively point to a role for Hoxb1/b2 and r4-derived structures in circuits controlling sound perception and amplification while Hoxa2 primarily determines specification and connectivity in an r3-dependent sound localization auditory circuit. Auditory circuits are a prime example of multiple Hox contribution in the assembly of a neuronal network.

Locomotor circuit assembly

The execution of coordinated movement in mammals is a complex behavior that relies on the appropriate formation of multiple neuronal networks throughout the CNS. Two key components of this behavior are the relay of proprioceptive information to MNs in the spinal cord and the transmission of cortical motor and sensory input via the brainstem to the cerebellum. Both of these processes have recently been linked to Hox function, as changes in Hox activity or expression lead to a rewiring of these networks. In LMC neurons, a major output of Hox activity relies on the high expression of the Hox cofactor Foxp1. In global Foxp1−/− mice, multiple aspects of LMC development are affected and mice die at mid embryonic stages due to heart failure (Wang et al., 2004). Conditional inactivation of Foxp1 in MNs reveals additional defects in these mice that manifest at later stages. In the formation of motor-sensory circuits, Ia proprioceptive afferents centrally project to MNs innervating the same muscle. In the absence of Hox activity from MNs, this sensory-motor connectivity is perturbed, resulting in severe motor discoordination (Surmeli et al., 2011). In the hindbrain, perturbing Hox5 paralog expression by eliminating the histone methyltransferase Ezh2 leads to abnormal migration and ectopic formation of precerebellar pontine nuclei (Di Meglio et al., 2013). An additional contribution of Hox genes to movement control can be seen by their contribution to the vestibular system. Hoxb1 activity is required for specifying the lateral vestibular nucleus (LVN), originating in r4, which modulates postural adjustments to movements (Chen et al., 2012).

Nociception

An unexpected role for a Hox gene, unrelated to the specification of rostrocaudal identity, emerged from the analysis of genes induced by nerve growth factor (NGF) in nociceptors of dorsal root ganglia (DRG) (Guo et al., 2011). Hoxd1 is the only Hox1 paralog that is not expressed in the hindbrain and shows an atypical expression pattern in DRGs that is continuous along the rostrocaudal axis. Hoxd1 is robustly induced by NGF in mouse, but not chick, and controls mammalian-specific features of nociceptor development such as peripheral target innervation and central axonal projections (Figure 8D). Hoxd1−/− mice exhibit behavioral defects consistent with decreased pain sensitivity and central projection patterns of nociceptors revert to the corresponding avian pattern. Remarkably, ectopic Hoxd1 expression in chick nociceptors is sufficient for the acquirement of murine nociceptor features, indicating a critical role for Hoxd1 in the evolution of nociceptive circuits.

Conclusions

The establishment of neural circuits in the hindbrain and spinal cord relies on the coordination of many events that play out over the course of development, including neuronal differentiation, migration, axonal guidance, and synaptogenesis. Each of these steps depends on programs that can be genetically encoded and/or driven by neuronal activity. Hox genes appear to predominate in hardwired aspects of circuit connectivity as best evident by their multifaceted roles in MNs. However, it is important to note that MN differentiation and connectivity also depend on a number of apparently Hox-independent pathways, including progenitor subtype specification (Sabharwal et al., 2011) and activity-dependent pathfinding and gene regulatory programs (Hanson and Landmesser, 2004, 2006). Although the mechanisms through which Hox genes contribute to synaptic specificity within neural circuits are not fully resolved, some general principles have emerged from analysis of their roles in different contexts.

The coordinate expression of the same Hox gene in groups of neurons and their targets provides an attractive model for driving synaptic specificity during the assembly of neuronal circuits. This idea is supported by studies in the hindbrain demonstrating a requirement for Hoxb1 both in facial MNs and tissues of the periphery for proper connectivity. Facial MNs generated in r4 innervate 2^nd branchial arch targets derived from neural crest cells which also originate from r4. Tissue-specific deletion of Hoxb1 from neural crest cells results in axonal guidance defects and death of facial MNs, confirming a role for Hoxb1 in target tissues (Arenkiel et al., 2004). A Hox-based matching system may resolve the question of synaptic specificity in circuits confined to specific rostrocaudal segments. For example, spinal interneurons display similar Hox expression profiles as MNs at the same rostrocaudal level (Dasen et al., 2005), and it is plausible that coordinated Hox activity contributes to their connectivity. On the other hand, long range projections such as those of spinocerebellar tract neurons, if Hox-dependent, would appear to be governed by a different set of principles, as Hox expression is distinct between these neurons and their supraspinal targets.

An additional mechanism that could contribute to circuit formation is the control of neuronal migration and settling as a function of Hox activity. This idea is exemplified by sensory-motor connectivity in the spinal cord, where a scrambling of dorsoventral MN cell body position resulting from the inactivation of Hox-dependent programs leads to defects in synaptic specificity (Surmeli et al., 2011). This phenotype appears to arise primarily from alterations in cell body position, as MNs receive proprioceptive inputs appropriate for their dorsoventral coordinates. However, in some cases motor or sensory nuclei appear to receive appropriate input despite changes in their position. In Pea3−/− mice for example, MNs innervating the triceps muscle show an altered position in the spinal cord but receive appropriate sensory input (Vrieseling and Arber, 2006). In Ezh2−/− mice, ectopic pontine nuclei also receive appropriate input from the cortex despite their aberrant position (Di Meglio et al., 2013). It is therefore likely that a combination of both positional and molecular cues defined by Hox genes orchestrate the specificity of connections during neural circuit formation.

Studies of Hox genes in CNS development have provided basic insights into the strategies through which a highly related group of transcription factors determines neuronal subtype identity. While there is compelling evidence that Hox genes play key roles in defining the identity, organization, and peripheral connectivity of motor neuron subtypes, and their target effectors are beginning to be defined, the contribution of Hox genes to synaptic specificity in neural circuits within the CNS remains to be resolved. As methodologies for tracing synaptic connectivity are improved, in conjunction with the ability to selectively deplete Hox genes from specific neuronal classes, the system is poised for a new set of discoveries that will undoubtedly reveal new and exciting roles for Hox genes in CNS development and function.