Mice were maintained in a conventional facility and fed in standard conditions (mice maintenance and mice breeding diets, Carfil Quality, Belgium) on a 14 h light/10 h dark cycle. Experimental procedures on animals were performed in accordance with the Belgian national guidelines and in agreement with the European directive 2010/63/UE. The protocol was approved by the animal ethic committee of the Université catholique de Louvain (approval 122803). For all gas euthanasia, CO₂ was administered by progressive delivery in the cage volume in accordance with guidelines. The transgenic CMV-CreER^T2 mice and the characterization of Cre activity with reporter mice have been described elsewhere (Santagati et al., 2005; Lizen et al., 2015). We used the Hoxa5^tm1.1Ljea mouse line carrying a Hoxa5 floxed allele as described (Tabaries et al., 2007). The characterization of Cre efficiency on the Hoxa5 floxed allele after tamoxifen induction has already been described (Lizen et al., 2015). Genotyping of animals was accomplished by polymerase chain reaction (PCR) using the primers and the PCR program described in Supplementary Table S1.

CMV-CreER^T2 transgenic mice were bred with mice carrying the conditional Hoxa5 floxed allele to generate double transgenic animals homozygous for the Hoxa5 floxed allele (Hoxa5^flox/flox) and heterozygous for the CMV-CreER^T2 transgene. Compound transgenic Hoxa5^flox/flox;CMV-CreER^T2+/- males were bred with Hoxa5^flox/flox females. These crosses had the advantage of producing in the same litter pups that were either CreER^T2 positive or CreER^T2 negative, providing intra-litter controls. The Hoxa5^flox/flox;CMV-CreER^T2+/- mice were used in a mixed background composed principally of 129/Sv, with a minor contribution of C57BL/6 and DBA/2.

For induction treatments, tamoxifen (Sigma–Aldrich, T5648) was dissolved in corn oil (Sigma–Aldrich, C8267). Intragastric injections of 50 μl of three different stock solutions were performed: we injected the first solution of 1 mg/ml at postnatal day (P)1, the second of 1.5 mg/ml was injected at P2 and P3, and the last one of 2 mg/ml was injected at P4 (Lizen et al., 2015). Pups stayed with their mother during the whole treatment until they were euthanized at P21. For euthanasia, CO₂ was administered by progressive delivery in the cage volume in accordance with guidelines.

Brains were collected after euthanasia as described above. For RNA-Seq experiments, whole brainstems were collected while for reverse transcription quantitative PCR (RT-qPCR) assays, the pons and caudal part of the medulla oblongata were dissected as illustrated in Supplementary Figure S1. Tissues were then snap frozen in liquid nitrogen, and stored at -80°C until RNA was extracted. RNA was isolated using TRI Reagent (Sigma–Aldrich, T9424) and quantified using a NanoDrop ND-1000 fluorospectrometer (Thermo Fisher Scientific, Waltham, MA, United States).

RNA sequencing analysis was carried out on total RNA extracted from tamoxifen-treated individuals of two different genotypes: Hoxa5^flox/flox;CMV-CreERT^2+/- (Hoxa5 cKO) and Hoxa5^flox/flox;CMV-CreERT^2-/- (Hoxa5 control). To minimize sample variability caused by individual differences between animals linked to tamoxifen treatment, RNA from three brainstems was pooled, and three pools were used per genotype (for a total of nine samples per genotype) (Supplementary Figure S1). Library preparation and sequencing were performed at the Genomics Facility of the Interdisciplinary Cluster for Applied Genoproteomics (GIGA Genomics, ULg-Liège, Belgium). RNA integrity was verified on the Bioanalyzer 2100 with RNA 6000 Nano chips, RNA integrity number scores were more than 8 for all samples. Illumina truSeq stranded total RNA library prep kit with ribo-zero was used to prepare libraries from 1 μg of total RNA following Illumina’s protocol. Ribosomal RNA was depleted from total RNA. The remaining RNA was purified, fragmented, and then primed with random hexamers and reverse transcribed into first strand cDNA followed by second strand synthesis. A single A nucleotide was added to the 3′end of the fragments before ligation to multiple indexing adapters. DNA fragments with adapter molecules on both ends were enriched by PCR and purified with Ampure XP magnetic beads. Libraries were validated on Bioanalyzer DNA 1000 chip and quantified by qPCR with the KAPA library quantification kit. Sequencing was performed on Illumina NextSeq 500 with NextSeq 500 High Output V2 kit (2 × 75 cycles). For data analysis, Fastq files were trimmed for adaptor sequences. The reads were aligned with Tophat 2.0.9¹ to the mouse genome. Cufflinks 2.2.0 suite² was used to generate fragments per kilobase of transcript per million fragments mapped values. CuffDiff was also used to identify significantly differentially expressed genes between the two genotypes. Statistical significance was defined by the Benjamini–Hochberg method. The q-value is a corrected p-value. The q-values below 0.05 were considered as a significant difference of gene expression between the two conditions. Gene ontology (GO) term analysis was performed using DAVID 6.7 Bioinformatics resource tools (Huang da et al., 2009). High-throughput sequencing data used in this publication have been deposited in NCBI’s Gene Expression Omnibus (Edgar et al., 2002) and are accessible through GEO Series accession number GSE103390³.

Brain subregions were homogenized in 1 ml of TRI Reagent (Sigma–Aldrich, T9424) during 30 s using high-throughput tissue homogenizer (Precellys^®24). RNA was isolated from the dissected brain subregions following manufacturer’s instructions. The precipitated RNA pellet was dissolved in 100 μl of nuclease free water and reverse-transcription was performed using a reverse transcription kit (Qiagen, 205311) according to the manufacturer’s instructions. Hoxa5 and target genes expression was assessed by qPCR on a StepOne+ apparatus (Applied Biosystems) using SYBR Green (Qiagen, 204143) as the detection method, and the geometric mean of two reference genes (H2A and Gapdh) as internal normalization reference. The relative transcripts abundance (fold change) was calculated using the 2^-ΔΔC_T calculation, where ΔCt = Ct_genetarget - Ct_{referencegenes} and ΔΔCt = ΔCt_mutant - ΔCt_control. Supplementary Table S2 provides primers sequences and concentrations, as well as the efficiency of the amplification reaction.

Eight sets of 14 μm thick serial coronal cryosections per brain were cut on a Leica CM 3050S cryostat. Gene expression was detected using digoxigenin-labeled RNA probes, as previously described, and as optimized for Hox genes analysis in the adult brain (Chotteau-Lelievre et al., 2006; Hutlet et al., 2016).

Plasmids allowing in vitro transcription of digoxigenin-labeled riboprobes were obtained from collaborators (En2: Dr A. Joyner; NeuroD2: M. Studer; Wnt7a: N. Bobbola). For other genes, probes were transcribed directly from PCR amplification products, following an adaptation of the Allen Brain Atlas procedures. For probes synthesis, cDNA of adult mouse brain was used to target genes of interest by a first PCR with designed gene-specific primers (NCBI primer blast). For practical purposes, the T7 universal primer sequence (5′-GCGTAATACGACTCACTATAGGG-3′) was directly attached to the forward and the Sp6 one (5′-ATTTAGGTGACACTATAG-3′) to the reverse primer. PCR products were then amplified by a second PCR reaction, purified using a PCR purification kit (Qiagen, 28104), and eluted with 30 μl of elution buffer (10 mM Tris, pH 8.5). In vitro transcription with SP6 and T7 polymerases was performed on the resulting eluate to generate both RNA sense and anti-sense probes. Probes size and quality were evaluated following in vitro transcription by migration on agarose gel. Prior use for in situ hybridization (ISH), all probes were further diluted to a concentration of 100 ng/μl of hybridization mix (Life Technologies, B8807G), and stored at -20°C. A detailed list of data regarding specific probe primers and sequences is available upon request.

Double ISH was performed on cryostat sections using a combination of digoxigenin- and fluorescein-labeled probes as described (Mayer et al., 2010). Anti-Fluorescein (Roche, 11426338910, 1:3000) immunodetection was used first to reveal candidate target genes, in combination with NBT-BCIP for the first staining reaction. The reaction was stopped with PBT/EDTA. Sections were washed 2 × 10 min in maleic acid buffer with 0.1% Tween 20 (MABT) (Sigma–Aldrich, 9005-64-5). Sections were then incubated in a blocking solution containing MABT with 2% of blocking reagent (Roche, 11 096 176 001) and 20% heat-inactivated goat serum (Sigma–Aldrich, G6767). To target the Hoxa5 probe, the sections were then incubated with anti-digoxigenin antibody (Roche, 11093274910, 1/1500) and INT-BCIP (Roche, 11 681 460 001) were used for the second staining reaction.

Hybridized sections were observed on a Leica DM2500 microscope, and pictures were captured with a Leica DFC420C camera. Annotation of expression thresholds versus background were established for each ISH taking into account the appropriate control and the duration of the staining reaction. A frequently used control for specific hybridization to mRNA is a sense strand probe which should not specifically anneal to the target RNA. In this study, we used as a negative control an antisense probe for a Hox gene not expressed in the brain (e.g., Hoxa10) and an antisense probe for Gad67 as a positive control.

To explore the HOXA5 transcriptional landscape in the postnatal brainstem, we carried out transcriptomic analyses by RNA-Seq using a model of postnatal Hoxa5 loss-of-function. We induced Hoxa5 inactivation after birth (P1–P4) using the tamoxifen-inducible CMV-CreER^T2 mice and conditional Hoxa5 floxed allele mice (Hoxa5^flox) (Santagati et al., 2005; Tabaries et al., 2007; Lizen et al., 2015). This treatment results in a severe reduction of Hoxa5 transcripts in brainstem subregions (8–12%, see below, RT-qPCR analysis) and of HOXA5-positive cells in the brain (7–26% depending on nuclei) (Lizen et al., 2015). RNA was extracted from the brainstem of P21 tamoxifen-treated Hoxa5^flox/flox;CMV-CreER^T2+/- (herein called Hoxa5 cKO) pups and from tamoxifen-treated Hoxa5^flox/flox;CMV-CreER^T2-/- (herein called Hoxa5 control) littermates. To identify differentially expressed genes, we performed three independent replicate experiments, in which each replicate contains RNA pooled from three brainstems (see section “Materials and Methods”; Supplementary Figure S1). Transcriptome analysis revealed 116 transcripts whose expression levels were significantly modified in brainstem from P21 mutants (q-value <0.05) and 318 transcripts with a p-value lower than 0.01 (Figure 1A and Supplementary Table S3). Among the 116 transcripts, 98 (84.5%) were downregulated and 18 (15.5%) were upregulated in mutants (Figure 1 and Supplementary Table S3). The relative abundance levels of transcripts that are significantly modified are illustrated for each of the three pooled samples in the heat map (Figure 1B).

To gain insights into the molecular programs regulated by HOXA5, we applied a GO analysis on the 116 differentially expressed genes. The top 3 most significant enriched terms were respectively glutamatergic synapse (p-value = 2.6 × 10^-7), calcium (p-value = 5.6 × 10^-7), and synapse (p-value = 5.7 × 10^-6), and included only downregulated genes (Figure 2A and Supplementary Table S4). A striking feature of this analysis is the importance of enriched terms associated with synapse such as glutamatergic synapse, calcium, synapse, calcium signaling pathway, long-term memory, cell junction, GABAergic synapse, long-term potentiation, postsynaptic membrane (Figure 2A). Further analyses through NCBI⁴ allowed to identify a total of 39 genes with evidence of synaptic function among the 98 downregulated genes (∼40%). In contrast, we did not identify any genes with evidence of function at the synapse among the upregulated transcripts in mutant brains.

Detailed analysis of candidate target genes involved in glutamatergic synapse revealed downregulation of several receptors and transporters, such as the metabotropic glutamate receptor Grm4, the subunit of the N-methyl-D-aspartate (NMDA) receptor Grin2c, the L-glutamate transmembrane transporters Slc1a6 (Eaat4) and Slc1a3 (Eaat1, Glast) and the vesicular glutamate transporter Slc17a7 (Vglut1) (Figure 2B). Downstream effectors, such as Prkcg, the member of the PKC family that regulates NMDA receptors trafficking (Lan et al., 2001), and the synaptic plasticity modulator Adcy1 were also downregulated (Wang and Zhang, 2012). Other candidate targets include members of the inositol 1,4,5-trisphosphate (IP3) pathway: Itpr1 (inositol 1,4,5-trisphosphate receptor), which mediates calcium release from the endoplasmic reticulum, and its interactor Homer3 (Figure 2B) (Kammermeier, 2006; Mizutani et al., 2008; Goto and Mikoshiba, 2011). Several of these targets are also influenced by calcium or involved in calcium signaling: Itpr1, Adcy1, Prkcg, and Grin2c (Figure 2C) (Oancea and Meyer, 1998; Wong et al., 1999; Goto and Mikoshiba, 2011). Additional downregulated targets involved in calcium signaling pathway are: Camk4, the calcium/calmodulin-dependent protein kinase IV involved in neural calcium signaling; Atp2a3, the ATPase endoplasmic reticulum Ca²⁺ transporting 3; Itpka, the inositol-trisphosphate 3-kinase; and Car8, the carbonic anhydrase 8 which regulates calcium homeostasis through ITPR1 inhibition (Zhuang et al., 2015). Finally, downregulated transcripts involved in GABAergic synapse included the two GABA receptor subunits Gabra6 and Gabrd, which were highly downregulated in mutants (fold change >3) and the System N Glutamine Transporter Slc38a5 (SNAT5) (Figure 2D). As indicated by the two enriched GO terms postsynaptic membrane and postsynaptic cell membrane (Figure 2A and Supplementary Table S4), most of the downregulated transcripts involved in glutamatergic synapse and GABAergic signal are mainly postsynaptic.

In summary, deletion of Hoxa5 gene function early after birth causes downregulation of numerous genes involved in synaptic function in the brain few weeks after. Notably, transcripts of multiple genes coding major players of glutamate and GABA neurotransmission, as well as key actors of calcium signaling, were downregulated. Due to the absence of highly enriched terms associated to upregulated genes (Supplementary Table S5), further analysis was carried out only for downregulated genes.

A subset of candidate HOXA5 target genes identified by RNA-Seq analysis was analyzed by real-time RT-qPCR. Selection of genes was based on GO functional annotations, expression fold changes, and literature. Genes that were downregulated at least 1.5-fold in Hoxa5 cKO brainstem (Supplementary Table S4) and known to be functionally related to synapse and calcium signaling (Figures 2B–D) were favored. Zic genes encoding transcription factors were also selected based on their known expression and critical function in mossy fibers precerebellar neurons (Dipietrantonio and Dymecki, 2009). For quantitative analysis through RT-qPCR, we improved the resolution of the analysis by measuring RNA levels of the selected genes in dissected pons and caudal part of the medulla oblongata since HOXA5 expression was shown to be enriched in these restricted areas (Supplementary Figure S1) (Lizen et al., 2017).

RT-qPCR confirmed downregulation of Cadps2, Wnt7a, Zic2, Neurod2, Zic1, Prkcg, Grm4, Adcy1, and Homer3 in the dissected medulla oblongata of Hoxa5 cKO specimens (p-value <0.05; Table 1). For Gabra6, Cbln3, En2, and Slc1a6, transcript levels were reduced in Hoxa5 cKO samples compared to Hoxa5 control but did not reach the significance (p-value between 0.1 and 0.3) (Figure 3A and Table 1). In the pons of Hoxa5 cKO, Calb1 was the only candidate for which the downregulation was close to the level of significance (p-value 0.0519), but the level of Zic1 was also reduced in Hoxa5 cKO samples compared to Hoxa5 control, although it did not reach the significance (p-value 0.169) (Figure 3B and Table 1). Indeed, although we observed reduced transcription levels for several other candidate genes, the variability between the samples, possibly explained by the absence of precise landmarks for macro-dissection, prevented us from concluding.

Thus, RT-qPCR data confirmed the RNA-Seq results, thereby validating the robustness of the analysis. Moreover, this analysis suggested region-specific regulation of HOXA5 downstream genes. This prompted us to evaluate the neuroanatomical expression profiles of HOXA5 candidate target genes.

To gain insights into the expression pattern of candidate target genes in brainstem, ISH was performed on the brain of wild-type mice at P21. Probes were designed for the subset of target genes analyzed by RT-qPCR (see above for criteria). Reliable labeling was detected for Calb1, Cadps2, Camk4, Cbln1, Grin2c, Itpr1, Neurod2, Vglut1, Wnt7a, Zic1, and Zic2 in Hoxa5-expressing nuclei (Table 2).

Nuclei that showed the highest enrichment in candidate target genes are the PG, the tegmental reticular nucleus (TRN), the lateral reticular nucleus (LRN) and the external cuneate nucleus (Parys et al., 2012), which are all major precerebellar nuclei providing inputs almost exclusively to the cerebellum. By forming mossy fibers that synapse onto the cerebellar granule cells, those nuclei are essential for coordinated motor activity, motor learning, and procedural memory (Rodriguez and Dymecki, 2000; Fu et al., 2013). Expression of Camk4, Neurod2, Vglut1, Zic1, and Zic2 was observed in the PG and TRN, in a pattern overlapping with Hoxa5 expression (Figure 4 and Table 2). Cbln1, Neurod2, Vglut1, and Zic1 transcripts were also detected in the interfascicular trigeminal nucleus (IF5), another Hoxa5-expressing major precerebellar pontine nucleus (Figure 4 and Table 2). Co-labeling by double ISH revealed that all Hoxa5-positive cells were also positive for Camk4, Zic1, and Zic2 in these nuclei (Figures 4C,F,I,L). In the medulla oblongata, Camk4, Cbln1, NeuroD2, Vglut1, Zic1, and Zic2 expression was detected in the LRN as Hoxa5 (Figure 5 and Table 2). As observed in the pons, all the Hoxa5-positive cells in the caudal part of LRN were also Camk4, NeuroD2, Zic1, and Zic2-positive, while only a portion of Hoxa5-positive cells were Cbln1-positive (Figures 5C,F,I and data not shown). In the external cuneate nucleus, signal for Cbln1, Grin2c, Vglut1, Zic1, and Zic2 was detected (Table 2). For Zic1 and Zic2, the signal seemed to completely co-localize with Hoxa5 expression, while only a portion of Hoxa5-positive cells were Cbln1-positive (data not shown). In addition to major precerebellar nuclei, transcripts of target genes were also detected in minor precerebellar nuclei, which have a less prominent projection to cerebellum, as illustrated in the caudal part of the spinal nucleus of the trigeminal (SPVC) for Adcy1 and Cbln1 (Figures 5K,N and Table 2). In this nucleus, co-localization of Adcy1 and Cbln1 with Hoxa5 was partial (Figures 5L,O).

Altogether, the expression data revealed regionalized expression of several HOXA5 candidate targets in the brainstem of P21 wild-type mice, in patterns that highly overlapped with the Hoxa5-expression pattern. Double ISH showed cellular co-localization of Hoxa5 and candidate targets transcripts in many precerebellar nuclei, although in PG, TRN, IF5, and LRN candidate targets displayed broader expression profiles than Hoxa5 (data not shown). This supports a regulation of these candidate target genes by HOXA5 in the precerebellar system, regulation that could be either direct or indirect.

Strikingly, very few candidate target genes are components or essential regulators of the Wnt/β-catenin, FGF, and TGF-β/Activin/BMP signaling pathways that have been identified as downstream pathways of several HOX proteins in embryonic development, neural differentiation and tumorigenesis (Chen et al., 2005; Schiedlmeier et al., 2007; Bami et al., 2011; Makki and Capecchi, 2011; Donaldson et al., 2012; Carroll and Capecchi, 2015; Duan et al., 2015; Hrycaj et al., 2015; Roux et al., 2015; Karmakar et al., 2017; Zhang et al., 2017). Moreover, except from Pcp2 (see Supplementary Table S3, line 17), none of the already identified direct or indirect targets of HOXA5 was significantly downregulated in our analysis (e.g., Ptn, Fslt1) (Sanlioglu et al., 1998; Chen et al., 2005; Boucherat et al., 2013). Our data support a model in which HOX transcription factors regulate the expression of different target genes depending on the ontogenetic context. In the postnatal brainstem, HOXA5 seems to specifically participate in the regulation of synaptic function.

Indeed, GO analysis and literature mining revealed that about 40% of the candidate genes are associated to the synapse, notably the glutamatergic synapse. A tentative synthesis of the functional links between proteins encoded by the HOXA5 candidate target genes associated to the glutamatergic synapse and calcium/calcium signaling is illustrated in Figure 6A, highlighting the importance of partnership between these molecules. Although the GABAergic synapse was not among the most significant enriched terms, the associated genes were among those whose fold change was the highest, such as Gabra6 and Gabrd (Figure 6B).

Most of the downregulated transcripts in Hoxa5 mutants involved in glutamatergic synapse are mainly postsynaptic while only a few are presynaptic (Figure 6A). Among the postsynaptic actors, we identified the two glutamate receptors: GRM4, a G protein-coupled receptor involved in the inhibition of cyclic AMP cascade, and GRIN2C, a NMDA receptor subunit that forms a channel permeable to calcium, potassium, or sodium depending on the other subunits. In the downstream pathway, we detected the calcium release mediator ITPR1, together with three of its interacting proteins, namely HOMER3, CAR8 and CALB1 (Tu et al., 1998; Hirota et al., 2003). Many proteins whose activity is influenced by calcium were identified, such as CAMK4, ADCY1, PRKCG, and ITPKA. We also detected the glutamatergic synapse organizer CBLN1, which interacts with NRXN and GRID2 to form a trimolecular trans-synaptic organizer (Hirai et al., 2005; Matsuda et al., 2010; Uemura et al., 2010; Ryu et al., 2012). Interestingly, GRID2 was also found downregulated in Hoxa5 mutant specimens with a p-value <0.01. Through the regulation of these postsynaptic targets, HOXA5 could actively influence neuronal excitability and synaptic plasticity. Among the presynaptic actors of the glutamatergic synapse, we detected Slc17a7 (VGLUT1), which has an essential role during postnatal development by loading presynaptic vesicle with glutamate, thereby regulating the efficacy of neurotransmission (Wojcik et al., 2004). Another presynaptic target of HOXA5 is the Ca²⁺-dependent activator protein for secretion 2 (CADPS2), which enhances release of the neurotrophic factor BDNF involved in the functional maturation of the glutamatergic synapses and essential for normal postnatal cerebellar development (Gottmann et al., 2009; Sadakata and Furuichi, 2009). By regulating Vglut1 and Cadps2, HOXA5 could thus control postnatal development of the cerebellum through the maturation of mossy fiber connections with the granular cells. This hypothesis is also supported by the downregulation of Zic1 transcripts in Hoxa5 cKO brainstem, which encode a Zinc-finger transcription factor involved in mossy fiber development (Dipietrantonio and Dymecki, 2009).

Although few members of GABAergic synapses are downregulated in Hoxa5 mutants, they are specific actors of cerebellar function. Indeed Gabra6 has been reported to have a very restricted expression in granule cells within the cerebellum and in a subset of precerebellar nuclei (Funfschilling and Reichardt, 2002; Maejima et al., 2013). This GABAa receptor subunit was shown to associate to the δ subunit encoded by Gabrd in the cerebellar granule cells, where they could impact on processes underlying motor coordination (Hanchar et al., 2005).

Together, these results support a new role for HOXA5 as a critical regulator of synaptic function, such as glutamatergic synapse integrity, excitability and/or plasticity, during the first three postnatal weeks.

A key feature of Hoxa5 expression in the postnatal and adult hindbrain is its enrichment in precerebellar nuclei in the pons and medulla oblongata (Lizen et al., 2017). Neurons of the precerebellar system, which relay both peripheral sensation and cortical input to the cerebellum, derive from the caudal part of the rhombic lip (r6–r8), using tangential migration to reach their final position. It has been suggested that a rhombomere-specific HOX code controls appropriate precerebellar neuron position during tangential migration as well as appropriate cortical input on pontine nuclei. Notably, PG5 HOX proteins were shown to be functionally implicated in the topographic organization of migrating precerebellar pontine neurons (Di Meglio et al., 2013).

Strikingly, a high proportion of significantly downregulated genes identified in the brainstem of Hoxa5 cKO mice at P21 are associated to the cerebellum. The ISH analysis revealed that these candidate target genes are also expressed in the precerebellar nuclei, where several co-localize at the cellular level with Hoxa5. These results suggest a postnatal requirement of HOXA5 in these nuclei for processes beyond the neuronal migration, such as activity-dependent refinement of neuronal pre-established circuits. As cerebellum grows and matures essentially during the postnatal period, cerebellar and precerebellar circuits are particular vulnerable to environmental signals (Wang and Zoghbi, 2001), as shown for pontine neurons that carry on their maturation during the first weeks after birth. Indeed, around P1, pyramidal tract fibers from the cortex grow toward the PG where they enter and establish synapses. Refinement of cortico-pontine projections is then observed by the decreasing of collateral branches around P16 (Inoue et al., 1991; Leergaard et al., 1995). During this postnatal refinement, both presynaptic (cortico-pontine axons) and postsynaptic (pontine neurons) elements play an active role. Our data support a role for HOXA5 in cortico-pontine projections refinement through regulation of transcripts involved in postsynaptic ending on pontine neurons. Of note, calcium and CAMK4 are part of signaling mechanisms involved in the regulation of dendrite growth, branching, and elongation (Nagendran and Hardy, 2011). A role for ADCY1 in the regulation of synapse stabilization/elimination during the late stage of fine topography of sensory circuits has also been demonstrated (Nicol and Gaspar, 2014). Finally, through its regulation of Wnt7a, HOXA5 could influence synapse development, remodeling, and maintenance at both presynaptic and postsynaptic levels (Dickins and Salinas, 2013). This hypothesis is further supported by the involvement of HOX factors in refinement of connectivity in brainstem cochlear neurons (Karmakar et al., 2017), where PG2 HOX proteins present in the postsynaptic neurons, the glutamatergic bushy cells, orchestrate the tonotopic precision of pre-synaptic input.

HOXA5 could also influence the postnatal maturation of pontine output in the granular layer of the cerebellum through regulating presynaptic actors such as VGLUT1 and CADPS2 mentioned above. Involvement of HOXA5 in the postnatal maturation of cortico-ponto-cerebellar connections could affect the different functions in which these circuits are involved such as procedural memory, motor learning, and motor coordination. In this context, it is worth mentioning that our GO analysis revealed enrichment in associated terms (e.g., long-term memory and long-term potentiation).

As PG2 to PG5 Hox genes are expressed in pontine nuclei from fetal to postnatal stages (Hutlet et al., 2016), they may share different functions in postnatal maturation of cortico-ponto-cerebellar circuits. Expression of candidate target genes in anterior part of pontine nuclei, where Hoxa5 is not expressed, suggests co-regulation by HOX proteins expressed more anteriorly in pontine nuclei. These proteins could regionally specify cortico-ponto-cerebellar projections during postnatal maturation through regulation of specific transcriptional codes in each pontine nuclei sub-population. As such, Hox genes could be involved in postnatal activity-dependent refinement of topographic precision of connectivity. In support of this hypothesis, a few of our candidate targets genes, such as Cadps2, Camk4, NeuroD1, Zic1, Cbln1, were identified downstream of other HOX proteins (Bami et al., 2011; Makki and Capecchi, 2011; Anderson et al., 2013; Karmakar et al., 2017).

Glutamatergic synapse dysfunction has profound effects both in neurological diseases and psychiatric disorders such as Parkinson and Alzheimer diseases, depressive disorders, schizophrenia, and autism spectrum disorder (Ravasi et al., 2010). Dysfunction of these synapses in the first week postnatally could impact on ASD. In this context, several genes downregulated in the brainstem of Hoxa5 cKO mouse at P21, such as Grm4, Cbln1, En2, Camk4, and Cadps2, have been associated to autism traits either in humans or in mouse models (Sadakata and Furuichi, 2009; Becker et al., 2014; Waltes et al., 2014; Genestine et al., 2015; Krishnan et al., 2017). Moreover, several HOXA5 candidate target genes identified such as Cbln1, Cbln3, Homer3, Grm4, En2, Atp2a3, Fat2, and Car8, are also downregulated in a recent mouse model of ASD, generated by increasing dosage of E3 ubiquitin ligase UBE3A, which recapitulates a highly penetrant type of autism in humans (Smith et al., 2011; Krishnan et al., 2017). Although HOXA5 protein has not been detected in the forebrain-derived territories, which have been mostly investigated in deciphering ASD underlying pathways, recent analysis highlights a role for the cerebellum in the perinatal risks for autism (Wang et al., 2014). Therefore, these data point to HOXA5 as a potential regulator in pathways that are affected in ASD.