Temporal-specific roles of fragile X mental retardation protein in the development of the hindbrain auditory circuit

The fragile X mental retardation protein (FMRP; encoded by FMR1) is an RNA-binding protein that regulates many aspects of gene expression and protein function (Bagni and Greenough, 2005; Bassell and Warren, 2008; Davis and Broadie, 2017). Functional loss of FMRP during development leads to fragile X syndrome (FXS), an intellectual disability. Many FXS symptoms appear early in life, including increasing autism features and emerging sensory hyperarousal, anxiety and hyperactivity (Hagerman et al., 2017). These clinical observations, along with FMRP expression throughout gestation (Abitbol et al., 1993; Hinds et al., 1993), implicate a role of FMRP in embryonic and early postnatal brains. Although FMRP regulation of neurotransmission and synaptic plasticity plays important roles in relatively mature brains (Bagni and Zukin, 2019; Bear et al., 2004; Deng et al., 2013; Ferron et al., 2014), how FMRP regulates brain development during embryonic stages is largely unknown, except its involvement in cortical neurogenesis (Castrén, 2016).

Axon growth is a multi-event process of embryonic brain development, including axonogenesis, pathfinding, arborization, and establishment of terminals on appropriate postsynaptic structures (reviewed by Chédotal and Richards, 2010; Comer et al., 2019; Stoeckli, 2018). Multiple lines of evidence support an involvement of FMRP in axonal development. In the Drosophila mushroom body, FMRP limits axonal growth and controls axonal pruning (Bodaleo and Gonzalez-Billault, 2016; Pan et al., 2004; Tessier and Broadie, 2008). In vertebrates, FMRP knockout results in excessive axonal branches in zebrafish motor neurons (Shamay-Ramot et al., 2015) and abnormal projection patterns in the mouse forebrain (Bureau et al., 2008; Scharkowski et al., 2018). FMRP also associates with RNAs that encode proteins involved in axonogenesis and synaptogenesis, including the microtubule-associated protein MAP1b (Bodaleo and Gonzalez-Billault, 2016), cell adhesion molecule Dscam (Jain and Welshhans, 2016) and the axon guidance cue netrin (Kang et al., 2019). However, the exact in vivo functions of FMRP in distinct axonal events are unclear.

Here, we investigated the roles of FMRP in axonal development of the auditory brainstem using the chick embryo as a model system. The avian nucleus magnocellularis (NM) and nucleus laminaris (NL) are structurally and functionally similar to the mammalian anteroventral cochlear nucleus (AVCN) and medial superior olive (MSO), respectively. NM/AVCN neurons receive temporally precise excitation from the auditory nerve and, in turn, send bilaterally segregated signals to the NL/MSO. Bipolar neurons in the NL and MSO are specialized to compute interaural time differences (ITDs), time disparities in the arrival of signals between the two ears; these binaural cues are crucial for sound localization and segregation (Nothwang, 2016; Overholt et al., 1992; Vonderschen and Wagner, 2014). Clinical studies have revealed a tight association between FMRP level and temporal performance and have found impaired temporal processing of visual and auditory information in FXS (Farzin et al., 2011; Hall et al., 2009; Kéri and Benedek, 2011; Kogan et al., 2004; Rais et al., 2018). Cellular studies have further identified structural and physiological abnormalities in the AVCN and its target cell groups in FMRP knockout rodents (Brown et al., 2010; El-Hassar et al., 2019; Garcia-Pino et al., 2017; Lu, 2019; McCullagh et al., 2017; Rotschafer et al., 2015; Ruby et al., 2015; Strumbos et al., 2010; Wang et al., 2015a). Finally, the nucleotide and amino acid sequences of chicken FMRP are similar to human FMRP (Price et al., 1996; Wang et al., 2014). Thus, studying FMRP regulation of NM and NL neurons is functionally relevant for understanding FXS. Additionally, the stereotyped pattern of axonal projection from the NM to the NL (Fig. 1A) provides a suitable model for mechanistic studies of axonal circuitry development (Allen-Sharpley and Cramer, 2012; Cramer et al., 2004; Seidl et al., 2014).

To track specific cell types and neural circuits in complex vertebrate brains, we developed several genetic tools to selectively label and manipulate NM precursors and neurons in developing chicken embryos. We have identified an early-onset FMRP localization in axons of NM precursors and neurons, and discovered that FMRP is required for the orderly and timely development of multiple axon events. These findings provide insights into the potential contribution of compromised embryonic brain development to FXS pathogenesis.

NM neurons project to the NL bilaterally (Fig. 1A). NL neurons are bipolar, with dendrites extending dorsally and ventrally from the soma to form two segregated dendritic domains. Cell bodies of NL neurons align into a single sheet, resulting in separate dorsal and ventral dendritic neuropil laminas. Individual NM axons bifurcate and project to the dorsal neuropil of the ipsilateral NL and the ventral neuropil of the contralateral NL. This segregated innervation pattern forms the anatomical substrate for ITD computation.

To label NM precursors and neurons selectively, we combined genetic markers with spatially controlled plasmid expression (Fig. 1B). The progenitor dA1 cells located along the dorsal-most region of the caudal rhombic lip express the basic helix-loop-helix transcription factor atonal homolog 1 (Atoh1), which gives rise to excitatory neurons in the auditory brainstem and precerebellar nuclei (Farago et al., 2006; Fujiyama et al., 2009; Helms et al., 2000; Machold and Fishell, 2005; Maricich et al., 2009). To enhance the specific labeling of the auditory neurons, we introduced a plasmid expressing the Atoh1-enhancer element upstream of Cre recombinase along with a Cre-dependent myristoylated-GFP (mGFP) reporter plasmid into rhombomeres 5-6 (r5-6), which contain NM and NL precursors, via in ovo electroporation (Avraham et al., 2009; Cramer et al., 2000; Helms et al., 2000; Kohl et al., 2012, 2013; Lipovsek and Wingate, 2018; Fig. S1). The electroporated Cre-conditional mGFP sequence was integrated into the chick genome by applying the PiggyBac transposition method (Wang et al., 2009), allowing prolonged expression of the reporter in the auditory neurons (Hadas et al., 2014; Lu et al., 2009). For more restricted NM labeling, we performed the electroporation at embryonic day (E) 2-2.5, before NL cells are born (Rubel et al., 1976). Following electroporation, mGFP⁺ cell bodies exhibited a restricted distribution in anatomically defined NM on the transfected side when examined at later stages (Fig. 1C). Axons of mGFP⁺ cells originated from the NM and projected to the NL bilaterally, exhibiting the characteristic pattern of NM-NL projection (Fig. 1C,D). The transfection rate, calculated as the percentage of mGFP⁺ neurons among all neurons in the NM, was 15.3%±10.3% (mean±s.d.; n=8 embryos) ranging from 3.4% to 34.4% (Fig. 1E). No mGFP⁺ cells were detected in the contralateral NM, NL, or surrounding brainstem regions. Thus, our genetic targeting of Atoh1-mGFP cells was predominantly the NM precursors, termed ‘Atoh1 precursors of NM’ henceforth, that establish the NM-NL circuit.

Next, we examined the development of the NM circuit stage by stage. We previously demonstrated that Atoh1/dA1 cells across r2-7 give rise to two contralateral axon projections (Kohl et al., 2012, 2015). One projection originated from the caudal hindbrain and elongated in a dorsal funiculus (DF), whereas the other arose from the more anterior hindbrain and formed a lateral funiculus (LF; Fig. 2A). The Atoh1 precursors of NM located at r5-6 extended their axons within the DF bundle (Fig. 2B). On transverse sections at E4.5, mGFP⁺ axons had crossed the midline and arrived at the location where the NM and NL will form (Fig. 2C, yellow arrow), as indicated by a midline crossing rate of 1.060 (n=7 embryos; Fig. 2D,E). On the ipsilateral side, mGFP⁺ axons formed a well-defined dorsal-to-ventral fascicule (Fig. 2C, white arrow), confirmed quantitatively by a small axonal bundle width ratio (0.259, n=7 embryos; Fig. 2D,F). At E7, the NL was separating from the NM with rostral-to-caudal progress (Fig. S2), consistent with a previous report (Hendricks et al., 2006). mGFP⁺ axons arrived at the emerging NL on the contralateral side. In contrast, the ipsilateral projection was not visible, which is consistent with the results of individual axonal reconstructions that showed no ipsilateral projection until E8 (Young and Rubel, 1986). At E9 and later, the NM and NL were recognizable as individual nuclei. The ipsilateral projection of mGFP⁺ cells to the dorsal neuropil of NL had formed, revealing the characteristic bilateral NM-NL projection (Fig. S2). To confirm this connectivity at the synaptic level, E2 embryos were electroporated with SV2-GFP along with Atoh1 enhancers and the PiggyBac transposase (Fig. 2G), enabling the expression of GFP in presynaptic vesicles (Hadas et al., 2014; Kohl et al., 2012). SV2-GFP was detected in the dorsal NL ipsilaterally and ventral NL contralaterally at E9 (Fig. 2H-I′), confirming the segregated synaptic projection from Atoh1-NM neurons to the NL.

Closer examination of NM axons within NL revealed a stage-dependent terminal maturation (Fig. 3). At E11-13, the incoming NM axons ended with a typical growth cone morphology with one to five filopodia (Fig. 3, white arrows). These filopodia gradually disappeared and turned into bouton endings at E15 (Fig. 3, yellow arrows). By post-hatch day (P) 6, NM axons exhibited a mature terminal morphology (Figs S3 and S4). Immunostaining demonstrated a distribution of vesicular glutamate transporters (vGluT2) along the axon course of Atoh1 precursors of NM at E4.5 (Fig. S5). At E15, NM axonal terminals contain a presynaptic SNARE component, SNAP25 (Fig. S6), indicating functional synapses. The time frame of the terminal morphological change was similar between the ipsilateral and contralateral projections of NM neurons, which indicates that the maturation of presynaptic terminals from the two NM inputs to NL neurons is temporally synchronized, although the two inputs differ in their time of arrival at the target area.

FMRP is strongly expressed in hindbrain (Fig. 4A). It is not known whether FMRP is localized in NM axons, and if so, when this localization emerges during development. Here, we addressed this question by immunostaining endogenous FMRP and localizing exogenous FMRP. Embryos were electroporated with Atoh1-mGFP at E2. At E4-5 (n=5 embryos), mGFP⁺ cells consistently showed somatic FMRP immunoreactivity (Fig. 4B-C″). Contralaterally, mGFP⁺ axons terminated in a cell-free region where FMRP staining was generally low (Fig. 4B-B″, yellow arrows). Closer observation demonstrated distinct FMRP puncta in this region (Fig. 4D′). These puncta were 0.2-0.7 µm in diameter, with an average density of 4.3 puncta per 100 µm² (28 sections from 5 embryos). A subset of FMRP puncta overlapped with mGFP⁺ axon processes (Fig. 4D-E), confirming FMRP localization in distal axons of NM precursors.

We next determined whether FMRP is localized in NM axons at late embryonic stages when they have formed synaptic connectivity with NL neurons. During this time window (E9 to E19), the neuropil regions of NL contain a mixture of NM axons, NL dendrites, and astrocyte processes. We developed a transposon-based vector system expressing chick FMRP (chFMRP) fused with mCherry (Fig. 5A) for constitutive expression (Schecterson et al., 2012). At E4, mCherry⁺ puncta were identified in the fibrous area where contralateral axons of NM precursors terminate (Fig. S7), consistent with the localization of endogenous FMRP puncta shown in Fig. 4. We co-electroporated E2 embryos with chFMRP-mCherry and Atoh1-mGFP (Fig. 5B) and harvested brainstem sections between E9 and E19 (n=13 embryos). A substantial number of NM cells expressed chFMRP-mCherry on the transfection side (Fig. 5C, left column). In addition, mCherry⁺ NL neurons were seen on the same side in some cases. To avoid this confounding factor, further analyses were performed in the contralateral NL in which mCherry labeling was exclusively derived from transfected NM axons. Across all cases, mCherry⁺ puncta were identified in the fiber region between the NL and the ventral brainstem, which contains incoming NM axons, as well as within the ventral neuropil domain of the NL (Fig. 5C, right column). This localization pattern indicates that the introduced chicken FMRP is localized in the distal portions of NM axons. This was further confirmed by the presence of mCherry⁺ puncta in Atoh1-mGFP expressing axons (Fig. 5D). Next, we replaced chFMRP-mCherry with human FMRP (hFMRP)-EGFP in the plasmid (Fig. 5A) and identified a similar pattern of FMRP distribution (Fig. 5E). This result indicates that the sequence of FMRP coding for its axon localization in NM axons is conserved between birds and humans.

In vitro studies implicate FMRP regulation in neurite outgrowth (Doers et al., 2014), axon elongation (Wang et al., 2015b), and branching (Zimmer et al., 2017). Together with our finding that Atoh1 precursors of NM contain FMRP in distal axons (Fig. 4), these studies raise the possibility that FMRP regulates axonal growth and pathfinding of NM precursors in vivo. We examined this possibility by determining the effects of downregulating FMRP on axon development of Atoh1 precursors of NM.

We first downregulated FMRP in Atoh1⁺ neurons using the CRISPR/Cas9 system (Cong et al., 2013; Hille and Charpentier, 2016). Two guide RNAs (gRNA₃ and gRNA₄) were designed to target exon 8 of the FMRP coding sequence to cause a deletion of ∼260 bp (Fig. 6A; Table S1). To verify this deletion, gRNA₃₊₄ plasmids, which contain Cas9 and GFP on the same pCAG-construct, were co-electroporated into the dorsal-most region of E2.5 embryos. Control embryos were electroporated with a control-gRNA construct (gRNA_control; Table S1). Both gRNA_control and gRNA₃₊₄ electroporated embryos demonstrated a 459 bp fragment of the size of the intact Fmr1 sequence, but gRNA₃₊₄ embryos also presented a lower-size band of 260 bp (Fig. 6B, red arrow), which reflects the deletion of ∼200 bp in electroporated cells. Next, we confirmed that this deletion prevents FMRP synthesis. At E6.5 (n=7 embryos), the majority of GFP⁺ cells (80%) were FMRP immunoreactive in embryos electroporated with gRNA_control (Fig. 6C-C″, arrows). In contrast, only 10% of GFP⁺ cells expressed FMRP following gRNA₃₊₄ expression (Fig. 6D-D″, arrowheads; Fig. 6E). Finally, we confirmed that expression of gRNA_control and gRNA₃₊₄ plasmids was confined to dA1 neurons, showing the overlapping expression of GFP with Lhx2/9 (Fig. 6F-H), a specific marker for dA1/Atoh1⁺ interneurons (Bermingham et al., 2001; Gray, 2013; Kohl et al., 2012).

To examine whether FMRP knockout affects dA1 axonal projections, embryos were electroporated with RNA_control or gRNA₃₊₄ CAG plasmids at E2.5 and harvested at E4.5 (n=7-10 embryos for each plasmid) and E6.5 (n=6-9 embryos for each plasmid). These time points encompass the period during which dA1 interneurons extend their axons along a well-defined dorsal-to-ventral fascicule, cross the midline, and project in a parallel ventral-to-dorsal trajectory until reaching the contralateral auditory nuclei anlage (Fig. 2C; Kohl et al., 2012). As expected, flat-mount views of E4.5 control embryos exhibited this typical trajectory of dA1 axons that cross the midline (Fig. 7A,A′, arrows), indicating that gRNA_control expression did not affect axonal growth. Observations from transverse sections further demonstrated that these axons projected in a fasciculated lateral bundle in the ipsilateral route and projected to the contralateral side (Fig. 7C,C′, arrows). Strikingly, many gRNA₃₊₄-expressing axons did not extend toward the floor plate and showed disorganized ipsilateral routes (Fig. 7B,B′, dashed arrows). Observations from transverse sections confirmed that axons projected ventrally in a broad mediolateral pattern rather than in a directional ventrolateral route, as well as extending medially toward the ventricle (Fig. 7D-E′, arrowheads). Quantitative analyses (as illustrated in Fig. 2D) revealed that the width of the GFP⁺ axonal bundle, measured in the circumferential axis, was significantly greater in gRNA₃₊₄-electroporated embryos than the control embryos (Fig. 7H; non-parametric P<0.001; Mann–Whitney U-test for this and all following comparisons). In addition, the angle of individual axons in relation to the mantle zone angle of the neural tube (Fig. 2D) was significantly increased following FMRP knockout (P<0.0001; Fig. S8A). This randomized axonal growth phenotype persisted in E6.5 embryos (Fig. 7G,G′, arrowheads) as opposed to control embryos (Fig. 7F-F′, arrows, 7I; P<0.001), but at a significantly reduced degree compared with E4.5 (Fig. S8B; P<0.05). To further validate the effect of FMRP knockout using the CRISPR/Cas approach, we designed an additional set of guide RNAs (gRNA₁ and gRNA₂) to target exon 4 of FMRP (Fig. S9A). Electroporation of gRNA₁₊₂ plasmids demonstrated significant disorganized growth of NM-GFP⁺ axons (P<0.05; Fig. S9B-D) as well as loss of FMRP immunoreactivity in the electroporated cells (Fig. S9E-F). Together, these results indicate that FMRP is required for the directed growth of NM precursor axons in a tight dorsal-to-ventral fascicule.

In addition to the disoriented pattern of axonal growth, possibly due to axon defasciculation, fewer axons crossed and progressed to the contralateral side following FMRP knockout on flat-mount views of E4.5 embryos (Fig. 7B,B′). Observations from transverse sections confirmed that fewer axons reached the level of the floor plate (Fig. 7E, arrows). We evaluated the rate of midline crossing by calculating the ipsilateral/contralateral ratio of GFP⁺ axons of the same transverse section, as described in Fig. 2D. At E4.5, the majority of GFP⁺ axons crossed the midline in control embryos, whereas less than half extended contralaterally following FMRP knockout (Fig. 7J; P<0.01). Yet, 2 days later at E6.5, the majority of GFP⁺ axons had crossed the midline in gRNA₃₊₄-electroporated embryos (Fig. 7G, arrows), similar to control embryos (Fig. 7F, arrows; Fig. 7K; P=0.645). This observation demonstrates that FMRP knockout induces a delay in axons reaching the floor plate but axons maintain the ability to cross the midline.

We next examined whether a partial reduction in FMRP expression affects the axonal growth pattern using a shRNA method. Fmr1 and control (scrambled) shRNAs were cloned into a transposon-based vector system with a Tol2 vector containing doxycycline (Dox) regulatory components and an EGFP reporter (Wang et al., 2018), enabling Dox-dependent temporal control of gene expression. We electroporated Fmr1 and scrambled shRNA plasmids into E2.5 hindbrains, triggered shRNA expression with Dox treatment immediately following the electroporation, and fixed embryos at E4.5 and E6.5 (n=6-8 embryos for each plasmid at each stage). As expected, the scrambled-shRNA group exhibited the typical dA1 projecting pattern (Fig. 8A,A′,C,C′,F,F′, arrows). Embryos expressing Fmr1-shRNA, however, showed profoundly aberrant axons (Fig. 8B,B′, dashed arrows), similar to the effect of FMRP knockout. Transverse section views confirmed that many Fmr1-shRNA-EGFP⁺ axons projected randomly toward the ventricular zone or toward the midline in a disorganized manner (Fig. 8D,D′,E,E′,G,G′, arrowheads), in stark contrast to the organized and directional pattern in control embryos (Fig. 8C,C′,F,F′). The width of Fmr1-shRNA-GFP⁺ axons was significantly larger than that of control axons at both E4.5 (Fig. 8H; P<0.01) and E6.5 (Fig. 8I; P<0.05). Nevertheless, similar to the effect of gRNA₃₊₄ expression, the axonal bundle width at E6.5 was reduced compared with E4.5 (Fig. S8C; P<0.05). Two-way ANOVA analyses did not reveal a significant effect of either the type of FMRP manipulation [F(1,29)=4.127; P=0.052] or the developmental stage [F(1,29)=1.176; P=0.287] on the degree of FMRP deficiency-induced changes in the width of the axon bundle. In contrast to the FMRP knockout, the majority of axons following shRNA-induced FMRP knockdown appeared to cross the midline normally at E4.5 (Fig. 8D,E). The rate of midline crossing was not significantly different between the groups at either developmental stage (Fig. 8J,K; E4.5: P=0.2403; E6.5: P=0.7209). Altogether, using two loss-of-function strategies we confirmed that FMRP expression in dA1 axons is required for directional axonal growth in a defined fascicule while navigating through developing brains.

To further determine whether loss of FMRP impairs the organized axonal growth of NM precursor axons, we analyzed its effect in vitro. Following electroporation of gRNA_control or gRNA₃₊₄ plasmids at E2.5 (n=12 embryos for each plasmid), hindbrains were isolated at E3.5, suspended into single cells, and incubated for 5 days. The cultures contained GFP⁺ cells along with non-transfected hindbrain cells (Fig. 9). To monitor the dynamics of neurite outgrowth, cultures were traced by live imaging every 6 h. Cells expressing gRNA_control plasmid demonstrated a gradual extension and elongation of neurites (Fig. 9A,C,E,G; Movie 1). Strikingly, cells expressing gRNA₃₊₄ plasmid demonstrated neurite overgrowth accompanied by aberrant turning of axons and enhanced branching along the neurites and in their terminals (Fig. 9B,D,F,H,I-L; Movie 2). Quantification of the results (n=6 wells for each plasmid) confirmed a gradual increase in neurite branch point (P<0.01) and length (P<0.001) over time in both treatments (Fig. 9M,N). However, the values differ greatly between the groups, as indicated for instance by the ∼3.5-fold increase in neurite branch points and length in cells expressing gRNA₃₊₄ plasmid compared with control cells at day 4. These in vitro results demonstrate that axons tend to spread and branch more extensively in the absence of FMRP, further verifying that FMRP is required to control the axonal growth behavior of NM precursors.

We next determined whether FMRP is required for presynaptic targeting by assessing the effects of FMRP downregulation on the pattern of synaptic connectivity of NM axons within NL. We electroporated E2 embryos with Fmr1-shRNA or control (scrambled) shRNA into NM precursors and triggered shRNA expression with Dox treatment at E8 (Fig. 10A). This late-onset expression preserved earlier developmental events of NM axons before NL neurons reach their final destination. During this time window, FMRP immunoreactivity was reduced by 40-60% in NM cell bodies as we measured previously (Wang et al., 2018).

We first examined embryos at E15 (n=8 embryos for scrambled-shRNA and 9 for Fmr1-shRNA). A typical projection pattern of NM axons was seen in both groups: EGFP⁺ axons arising from the NM extended to both the ipsilateral and contralateral NL. In embryos expressing scrambled-shRNA, NM axons were restricted to the dorsal NL ipsilaterally and ventral NL contralaterally (Fig. 10B,D). In contrast, embryos expressing Fmr1-shRNA demonstrated EGFP⁺ axons that projected beyond their assigned neuropil domain, extended through the cell body layer, and terminated within the other domain (Fig. 10C,E). We measured the area containing EGFP⁺ axons in each neuropil domain of the contralateral NL and calculated the dorsal/ventral ratio of this measure. This ratio was low in embryos expressing scrambled-shRNA, indicating a strong preference for ventral localization, and was significantly enhanced following Fmr1-shRNA transfection (P=0.0079; Fig. 10G), demonstrating abnormal axonal overshoot. This phenotype was not observed at E19 (n=5 embryos; Fig. 10F,G), indicating that the effect of FMRP deficiency on axon targeting is stage dependent.

We next wanted to examine whether the aberrant NM axons form synapses. By dye-filling individual NL neurons, we found that EGFP⁺ axons were located immediately opposite the dorsal dendrites of NL neurons (Fig. 11A-A″). These EGFP⁺ axons were immunoreactive to synaptotagmin 2 (Syt2; Fig. 11B-B″), a presynaptic vesicle calcium sensor for neurotransmitter release. Together, these observations demonstrate that the aberrant NM axons form synapses.

Finally, we examined whether FMRP knockdown altered the morphological maturation of NM axonal terminals. In embryos expressing Fmr1-shRNA, the number of filopodia per EGFP⁺ terminal was zero, one or two at E15, similar to the control group as measured from Atoh1-mGFP labeled terminals (Fig. 12; P=0.5695).

Using high-specificity genetic tools in chicken embryos, we uncovered an early onset of FMRP localization in developing axons of auditory neurons and demonstrated that cell-autonomous FMRP expression is required for orderly and timely axonal navigation and synaptic targeting in vivo during discrete episodes of axon and circuit development.

NM cells are born at E2-2.5 (Rubel et al., 1976). FMRP localization can be detected as early as E4 in developing axons of NM precursors, demonstrating that FMRP starts localizing in distal axons of NM precursors shortly after Fmr1 gene expression and axon genesis. This finding is consistent with FMRP localization in newly formed neurites of PC-12 cells (De Diego Otero et al., 2002) and axon growth cones of cultured mammalian neurons (Antar et al., 2006; Hengst et al., 2006; Jain and Welshhans, 2016). FMRP has also been identified in relatively mature axons as a component of fragile X granules (FXGs) in postnatal mammalian brains (Christie et al., 2009; Chyung et al., 2018; Korsak et al., 2017; Shepard et al., 2020). FMRP puncta found in developing NM axons resemble these FXGs in size and density (Christie et al., 2009). However, the majority of FXGs in postnatal mouse brainstems contain the fragile X-related proteins FXR1P and FXR2P (FXR1 and FXR2, respectively) but not FMRP (Chyung et al., 2018). Whether this difference reflects interspecies variation or developmental stage dependency is yet to be determined.

Consistent with axon localization of FMRP during early development, FMRP deficiency in Atoh1/NM precursors results in widened axonal bundles due to randomized axonal growth instead of directional growing in a defined fascicule. It is known that axon fasciculation can be controlled at the level of axonal growth cones (Honig et al., 1998) and/or regulated by axon tension through shaft-shaft interactions (Šmít et al., 2017). Our in vitro results support a likely involvement of growth cone behaviors as the absence of FMRP in NM precursor axons leads to excessively branched growth cones together with axonal overgrowth. Indeed, previous studies showed that FMRP loss enhances growth cone filopodia and attenuates growth cone collapse in vitro (Antar et al., 2006; Doers et al., 2014; Li et al., 2009), and these actions may involve FMRP regulation of cell adhesion and axon guidance cues. For example, FMRP colocalizes with Dscam mRNAs in cortical axons (Jain and Welshhans, 2016) and Dscam promotes axon fasciculation in the developing optic fiber (Bruce et al., 2017). Netrin mRNAs are associated with FMRP in HEK293 cells and was linked to axon extension phenotype in Fmr1 knockout Drosophila (Kang et al., 2019). Notably, netrin has a profound role in navigating commissural axons in the hindbrain and spinal cord in a tight bundle toward the midline (Moreno-Bravo et al., 2019; Serafini et al., 1996; Varadarajan et al., 2017; Yung et al., 2018). Notably, the degree of the aberrant projections decreases as development proceeds. The partial recovery in the axonal directionality may suggest that FMRP-deficient axons are capable of correcting their growth pattern with time, as shown for instance in an ascending projection connecting specific cortical layers in Fmr1 knockout mice (Bureau et al., 2008). Yet, to fully decipher the fate of FMRP-deficient axons, advanced in vivo live-imaging techniques will be needed to trace the behavior of individual axons.

The second phenotype we identified was a delay in axonal midline crossing. In control embryos, axons of Atoh1/NM precursors crossed the midline at E4.5. Following FMRP knockout, the axon crossing was not complete until 2 days later at E6.5. This phenotype may be caused by a general slowing down of axon growth in vivo. For example, FXS neurons derived from human pluripotent stem cells show reduced neurite outgrowth (Doers et al., 2014). FMRP knockdown significantly reduces axonal growth of cultured mouse neurons in response to nerve growth factor (Wang et al., 2015b). This slowed growth may be partially associated with FMRP regulation of microtubule signaling and dynamics (Bodaleo and Gonzalez-Billault, 2016; Wang et al., 2015b). Alternatively, a delay in midline crossing could be secondary to axon defasciculation. In the zebrafish forebrain, axon-axon interaction (likely axon fasciculation) shapes the midline kinetics of commissural axons (Bak and Fraser, 2003). Moreover, overgrowth and overbranching of axons in brains of Drosophila FMRP mutants have been reported (Pan et al., 2004), consistent with our in vitro data in which, rather than attenuation in axonal growth, we observed extensive neurite growth and enhanced branching points upon FMRP knockout. Reduced axon fasciculation thus may negatively affect midline crossing in auditory neurons. However, a delay in midline crossing was not detected following FMRP knockdown, although FMRP knockdown resulted in similar degrees of axon defasciculation as did knockout of FMRP. This, then, suggests that FMRP regulates multiple factors in controlling the speed of axon crossing. Additional mechanisms may include suppressed expression of axon guidance genes and compromised neuronal response to guidance cues following FMRP loss (Halevy et al., 2015; Li et al., 2009).

In addition to controlling axon pathfinding, FMRP is also involved in determining the pattern of local axon projection in their target area. Following acute FMRP deficiency, NM axons terminate, and likely form functional synapses, on both the dorsal and ventral dendrites of the same NL neurons. This projection pattern is expected to negatively affect the accuracy of coincidence detection of NL neurons. This change can be interpreted as a compromised ability of developmental axon pruning, as seen in Drosophila FMRP mutants (Pan et al., 2004; Tessier and Broadie, 2008). Defective synaptic elimination and dendritic pruning have also been observed in brains of FXS individuals and FMRP knockout mice (Comery et al., 1997; Ivanco and Greenough, 2002; Jawaid et al., 2018) as well as in FMRP-reduced NM neurons (Wang et al., 2018). However, there is no evidence that under normal circumstances NM axons project to both dendritic domains of the same NL neurons and subsequently retract from one domain (Young and Rubel, 1986; Rubel and Fritzsch, 2002). It is therefore likely that the aberrant axon projection following FMRP knockdown reflects errors in axon targeting. NM axons with less FMRP may become less sensitive to guiding cues from NL neurons or local astrocytes that control the pattern of synaptic distribution (Allen-Sharpley and Cramer, 2012; Korn et al., 2012; Rotschafer et al., 2016). This possibility is consistent with the localization of FMRP puncta in the distal axonal processes (Fig. 5). Although their exact relationship with synapses is yet to be determined, it is notable that many FMRP puncta are not in the region where synapses are located. Thus, FMRP is likely to exert the axonal functions that have been identified in our study without being associated with synapses.

Additional lines of evidence in support of FMRP regulation of axonal targeting via growth cone dynamics include the presence of abnormal protein patterns only during the period when NM axons exhibit dynamic growth cones with filopodia and the normal maturation of axonal endings from growth cones to bouton-like terminals independent of FMRP expression.

It is worth noting that axon-glia interactions may also contribute to FMRP regulation of axon events, given their well-established roles in axon guidance, fasciculation and targeting (Rigby et al., 2020). Interestingly, some of the molecules that participate in a direct axon-glia contact, such as NCAM and Semaphorins-Plexins (Franceschini and Barnett, 1996; Goldberg et al., 2004; Keilhauer et al., 1985; Miragall et al., 1989; Moreau-Fauvarque et al., 2003; Neugebauer et al., 1988; Shim et al., 2012), are known as FMRP targets in neurons (Li et al., 2009; Liao et al., 2008; Menon and Mihailescu, 2007). Hence, it is possible that lack of FMRP in NM axons prevents their interaction with glial cells via these proteins, which, in turn, leads to aberrant axonal growth. Additionally, FMRP may control axonal targeting by regulating the formation of axon myelination (Doll et al., 2020; Pacey et al., 2013) which influences functional development of axon terminals (Berret et al., 2016; Xu et al., 2017).

It remains unknown whether the tonotopic organization of NM axonal projection was affected by FMRP deficiency. Our manipulations affected only ∼15% NM neurons, which were often scattered throughout the cell group, thus it was not possible to determine the effect on the tonotopic organization. Studies of Fmr1 knockout mice demonstrated a normal tonotopic frequency representation in the auditory cortex (Kim et al., 2013). However, FMRP loss diminishes the developmental plasticity of this representation (Kim et al., 2013), flattens the tonotopic organization of potassium channel Kv3.1b (Strumbos et al., 2010), and results in frequency-specific decreases in inhibitory presynaptic structures (McCullagh et al., 2017), suggesting a potential link of FMRP with specific features of tonotopic regulations.

Our results enhance the current understanding of FXS pathogenesis in three aspects. First, we strengthen the concept that FXS neuropathology involves sensory systems. FMRP is strongly expressed in the auditory system (Zorio et al., 2017) and FMRP loss alters cellular properties of auditory neurons and auditory processing (reviewed by McCullagh et al., 2020). Our current and previous studies (Wang et al., 2018) further demonstrate a role of FMRP in the proper development of auditory connectivity. Second, we reveal a cell-autonomous regulation of FMRP in axon navigation. Early-onset axon localization of FMRP suggests that this regulation occurs locally in axons, supporting axonal mechanisms of FXS pathology. For example, diffusion tensor imaging in FXS females revealed morphological changes in white matter tracts that may reflect alterations in axon density or coherence (Barnea-Goraly et al., 2003). Thus, FMRP loss-induced axon defasciculation may be a mechanism that underlies this clinical phenotype. Lastly, our results add to the existing literature that FMRP loss leads to substantial alterations in developing brains that may be undetectable later in life. FMRP knockout mouse cortex shows alterations in connection probability, axon shape and dendritic spine length at early, but not late, postnatal ages (Bureau et al., 2008; Galvez and Greenough, 2005; Nimchinsky et al., 2001). Our current and previous studies further show developmentally restricted dendritic and axonal alterations in auditory neurons (Wang et al., 2018). The significance of these early-onset and transient changes was recently highlighted in Drosophila, in which the requirement of FMRP for normal brain function and behaviors is tightly restricted to an early developmental period (Doll and Broadie, 2015; Sears and Broadie, 2018). If this holds true in vertebrates, it would suggest that early axon deficits following FMRP loss may be responsible for life-long behavioral deficits in FXS. Although challenging, identifying FMRP regulation of early developmental events and determining how this regulation influences later circuit properties may be the beginning of a deeper understanding of FXS neuropathology. The auditory brainstem circuits characterized and the novel genetic tools developed in this study provide a strategy that contributes to this effort.

Fertilized White Leghorn and Loman Broiler chicken eggs (Gallus gallus domesticus) were obtained from Charles River Laboratories (Wilmington, MA, USA) and Gil-Guy Farm (Orot, Israel), respectively. Eggs were incubated for 2-2.5 days at 38°C until Hamburger-Hamilton stage 12-15. In ovo electroporation was performed as described previously (Kohl et al., 2012; Wang et al., 2018). Briefly, DNA constructs (4-5 μg/μl, diluted in PBS) were injected into the lumen of neural tubes at the rhombomere 5-6 level. Electroporation was performed with a platinum bipolar electrode or bent L-shaped gold electrodes that were placed on the two sides of the hindbrain to gain unilateral transfection. Embryos underwent four electrical pulses of 20-25 V 30-45 ms in duration using a BTX 3000 (Harvard Apparatus) or a Grass SD9 electroporator (Grass instruments). Following electroporation, the eggs were re-incubated until dissection at the desired developmental stages. Embryos electroporated with drug-inducible constructs (see below) were treated by adding 50 µl of doxycycline (1 mg/ml in sterile PBS; MilliporeSigma) onto the chorioallantoic membrane to trigger transgene transcription. Following the first Dox administration, embryos were treated again every other day to maintain gene expression before tissue dissection.

Hindbrains from electroporated embryos were dissected at E3.5 and incubated for 10 min at 37°C with TrypLE Express (Gibco, Thermo Fisher Scientific) to dissociate the tissue into single cells, as previously described (Peretz et al., 2016, 2018). TrypLE was neutralized with embryonic stem cell media containing DMEM/F-12 1:1, 20% KnockOut serum replacement, 2 mM GlutaMax L-alanyl-L-glutamine, 0.1 mM nonessential amino acids and 1:50 penicillin-streptomycin (all from Gibco, Thermo Fisher Scientific), together with 0.1 mM β-mercaptoethanol and amphotericin B (1:400) (both from MilliporeSigma). Cells were passed through a 100 μm mesh strainer, centrifuged at 600 g for 10 min, seeded in 48-well plates (∼2×10⁵ cells/well) (Nunclon Delta Surface, Thermo Fisher Scientific), and incubated at 37°C in 5% CO₂. For live imaging, well plates were imaged every 6 h in the IncuCyte S3 Zoom HD/2CLR time-lapse microscopy system equipped with ×20 Plan Fluorobjective (Sartorius). Time-lapse movies were generated by capturing phase and green fluorescence images of cells in wells for up to 5 days. Stacks of images were exported in TIF format using the IncuCyte graph/export menu. Videos were assembled by exporting into MP4 format.

For genetic targeting of Atoh1-expressing neurons, an Atoh1 enhancer element (Helms et al., 2000; Pennacchio et al., 2006) was cloned upstream of a Cre-recombinase sequence (Atoh1-Cre) and electroporated along with a conditional reporter plasmid containing a floxed STOP cassette in between the CAGG enhancer/promoter module and nuclear (n) or membranal (m) GFP gene (pCAGG-LoxP-STOP-LoxP-n/mGFP), as previously reported (Avraham et al., 2009; Kohl et al., 2012; Lumpkin et al., 2003; Reeber et al., 2008). For plasmid integration into the genome, the conditional reporter cassette was cloned between two PiggyBac (PB) arms (PB-CAGG-LoxP-STOPLoxPSTOP-GFP-PB) and electroporated along with the Atoh1-Cre and Pbase transposase plasmids (Hadas et al., 2014; Kohl et al., 2012; Lu et al., 2009; Wang et al., 2010). For tracing pre-synaptic connections, a reporter plasmid containing the synaptic tracer SV2-GFP (PB-CAG-LoxP-STOP-LoxP-SV2-GFP-PB) (Hadas et al., 2014; Kohl et al., 2012) was electroporated along with the Atoh1 enhancer and the Pbase transposase.

For constitutive expression of chicken or human Fmr1, mCherry-Fmr1 fused coding sequence was chemically synthesized (GenScript) and sub-cloned into the pT2K-CAGGS vector. For electroporation, the two plasmids (pT2K-CAGGS-mCherry-chFMRP and pCAGGS-T2TP) were concentrated at 4-5 µg/µl and mixed in equal amounts.

For shRNA targeting of FMRP, five shRNAs directed against specific sequences of chicken Fmr1 were designed using siRNA Wizard v3.1 (InvivoGen) and the siDESIGN Center (Thermo Fisher Scientific). Plasmids were chemically synthesized (GENEWIZ) and EndoFree DNA Maxi Preps were performed (Qiagen). The most effective shRNA (gaggatcaagatgcagtgaaata; nucleotides 951-973 of chicken Fmr1) was determined based on its knockdown effect in the developing brainstem (Wang et al., 2018) and used for subsequent experiments. A scrambled shRNA (attagaataagtgcgagagaata) was designed using the Genscript algorithm and confirmed by blasting this shRNA sequence against the chicken genome. Fmr1 and scrambled shRNAs were cloned into a transposon-based vector system with a Tol2 vector containing doxycycline regulatory components (Schecterson et al., 2012; Wang et al., 2018). Tol2 transposable element sequences enable stable integration of the transposon into the chick genome, whereas doxycycline regulatory elements allow temporal control of gene expression.

For CRISPR/Cas9 targeting of FMRP, we used the Genome Engineering Toolbox that was designed by the Zhang lab (Cong et al., 2013). The pX330 plasmid (#42230, Addgene) (Sakuma et al., 2014) was modified by adding a T2A-EGFP cassette at the carboxyl terminus of Cas9. gRNAs for Fmr1 were designed utilizing the chopchop design tool (https://chopchop.cbu.uib.no/). gRNAs targeting exon 8 were cloned into the modified pX330 plasmid (Table S1). For testing the efficiency of the gRNA, the targeting plasmids were electroporated into the hindbrain at E2.5. Hindbrains were dissected 48 h following electroporation, and a 2 mm piece of hindbrain tissue was processed for DNA extraction, using a previously published ‘tail digestion and DNA extraction’ protocol (Wang and Storm, 2006). Genomic DNA was analyzed by polymerase chain reaction (PCR) using primers specific to sequences up- and downstream of the FMRP-gRNA₃₊₄ target sites. Nested PCR was used to amplify the targeted region. For exon 8 targeting, Test-F3 and Test-R1 were used for the first round of PCR, followed by Test-F2 and Test-R2 for the second round.

Brainstem was dissected at various stages and immersed in 4% paraformaldehyde in 0.1 M phosphate buffer (PB) overnight at 4°C. For whole-mount preparation, hindbrains were cut open along the roof plate, after which the tissue was spread open on slides to produce flat-mount preparations (Kayam et al., 2013; Weisinger et al., 2012). For transverse sections, brainstems were transferred to 30% sucrose in PB until settling, followed by their sectioning in the coronal plane at 30 μm. Alternate sections were immunohistochemically stained by incubation with primary antibody solutions diluted in PBS with 0.3% Triton X-100 overnight at 4°C, followed by Alexa Fluor secondary antibodies (goat anti-rabbit or anti-mouse; Thermo Fisher, A11019 and A11036, or A11012) at 1:1000 overnight at 4°C. Some sections were counterstained with DAPI and/or NeuroTrace (Life Technologies), a fluorescent Nissl stain, at a concentration of 1:1000 and incubated together with secondary antibodies. Sections were mounted on gelatin-coated slides and coverslipped with Fluoromount-G mounting medium (Southern Biotech) for imaging.

Primary antibodies used include the custom-made polyclonal rabbit anti-FMRP (1:1000, RRID: AB_2861242; Wang et al., 2018; Yu et al., 2020), anti-synaptotagmin 2 (1:1000, DSHB znp-1, RRID: AB_2315626), anti-SNAP25 (1:1000, Abcam 5666, RRID: AB_305033), anti-microtubule associated protein 2 (MAP2; Millipore MAB 3418; RRID: AB_94856), custom-made polyclonal rabbit anti-Lhx2/9 (1:100, I. Sibony and T. Schultheiss, unpublished data; kind gift from T. Schultheiss, Technion-Israel Institute of Technology, Haifa, Israel) and polyclonal rabbit anti VGluT2 (1:150, Synaptic Systems 135402).

Following electroporation with Fmr1-shRNA plasmids and doxycycline treatment, E15 embryos were used to prepare acute brainstem slices as previously described (Wang et al., 2017). NL cells were individually dye-filled with Alexa Fluor 568 dextran (Invitrogen) following our published protocol (Wang and Rubel, 2012; Wang et al., 2017).

Images for illustration were captured with the Olympus FV1200 at Florida State University and with the E400 microscope (Nikon) with DP70 CCD camera (Olympus) at the Hebrew University of Jerusalem. Image brightness, gamma and contrast adjustments were performed in Adobe Photoshop. All adjustments were applied equally to all images of the same set of staining from the same animal. In some cases, multiple images were taken and stitched by the automatic Image Composite Editor (ICE) software unless otherwise stated.

Atoh1-Cre-transfected NM neurons and total NM neurons counterstained by NeuroTrace on each section were counted using Cell Counter of Image J. Sections from the same animal were grouped, and the transfection ratio was calculated as: transfection ratio=number of Atoh1-mGFP⁺ NM neurons/total NM neurons (n=8 embryos).

The axon terminal morphology was characterized by numbers of filopodia. Image stacks containing identifiable intact axon terminals were reconstructed using ImageJ, and the numbers of filopodia on each terminal were counted on both ipsilateral and contralateral sides. Number of filopodium per terminal was then calculated and compared between the immature stages (E11-E13) and E15.

FMRP granule localization in Atoh1-mGFP-labeled axons were analyzed using ImageJ. Briefly, a straight-line region of interest was drawn across an FMRP granule with Atoh1-mGFP transfection and applied to both channels. The fluorescent intensity profile was then analyzed and plotted using GraphPad Prism 7 software.

Quantification of expression of gRNA_control and gRNA₃₊₄ plasmids in dA1 cells was demonstrated by box plot analysis. For each group, two transverse sections obtained from three different embryos at E4.5 were taken. Each data point represents one section. The ratio of cells co-expressing gRNA-GFP⁺ and the dA1-specific marker Lhx2/9 out of the total gRNA-GFP⁺-expressing cells is presented.

Quantification of the extent of FMRP expression in gRNA_control- and gRNA₃₊₄-expressing cells is demonstrated by box plot analysis. For each group, electroporated sagittal sections obtained from seven different embryos at E6 were taken. Each data point represents one section for which the ratio of (FMRP⁺+GFP⁺)/GFP⁺ cells was measured.

Quantification of expression of gRNA_control and gRNA₃₊₄ plasmids in dA1 cells is demonstrated by box plot analysis. For each group_, two transverse sections obtained from three different embryos at E4.5 were taken. Each data point represents one section. The ratio of cells co-expressing gRNA-GFP⁺ and the dA1-specific marker Lhx2/9 out of the total gRNA-GFP⁺-expressing cells is presented.

Axonal width measurement was performed for two different experiments (Fmr1-shRNA and FMRP-CRISPR) at E4.5 and E6.5. Each stage included two groups: (1) gRNA_control- and gRNA₃₊₄-expressing cells and (2) sc-shRNA-GFP- and Fmr1-shRNA-GFP-expressing cells. Box plots are demonstrated for each group, from which cross-sections from seven different embryos (E4.5) or four embryos (E6.5) were taken. Each data point represents one section for which the ratio of the axonal length relative to the mantle-ventricular width was measured using ImageJ software.

Box plot quantification of axonal crossing was performed for two different experiments at E4.5 and E6.5. Each stage contained two groups: (1) gRNA_control- and gRNA₃₊₄-expressing cells and (2) sc-shRNA-GFP- and FMR1-shRNA-GFP-expressing cells. For each group, cross-sections from seven different embryos (E4.5) or four embryos (E6.5) were taken. Each data point represents one section for which the ratio of the signal intensity between commissural axons and non-commissural axons was measured using ImageJ software.

Neurite length (mm/mm²) and branch point (per mm²) were calculated in gRNA_control- and gRNA₃₊₄-expressing neurons in each well (n=6 wells for each treatment) using the IncuCyte Zoom NeuroTrack software module (Sartorius), as described by Wurster et al. (2019). Microplate graphs were generated using the time plot feature in the graph/export menu of the IncuCyte Zoom software.

The percentage of Atoh1-Cre::nGFP-expressing cells was calculated by counting the number of GFP⁺ nuclei co-expressing the dA1-specific marker Lhx2/9 out of the total number of GFP-expressing nuclei (n=7 embryos).

Axonal projection was measured from Fmr1-shRNA-transfected embryos at E15 and E19, as well as from scrambled-shRNA-transfected embryos at E15 using ImageJ. For each embryo, transverse sections containing the middle and rostral NL, where NL cell bodies are aligned into a single layer, were used for the analysis. For each section, the dorsal and ventral neuropil regions of the NL on the side contralateral to the transfection were outlined based on NeuroTrace staining. The neuropil area covered by EGFP⁺ axons was then measured for each neuropil region. The specificity of axon projection was evaluated by calculating the ratio of the dorsal EGFP⁺ area to the ventral EGFP⁺ area. The ratios from all sections (usually two or three) of the same embryo were averaged as individual data points and compared between Fmr1-shRNA- and control-shRNA-transfected animals (n=5-9 animals for each group).

Statistics were performed by Mann–Whitney non-parametric U-test using the GraphPad Prism 7 software package (GraphPad Software). P<0.05 was considered statistically significant. Data are displayed as mean±s.d. as indicated in the Results. Each individual data point represents one animal. Two-way ANOVA was used for Tukey multiple comparisons.

We thank Edwin W Rubel (University of Washington) for facilitating the collaboration between Y.W. and D.S.-D. We are grateful to David Fradkin for assistance with the IncuCyte system, Itzhak Sibony and Tom Schultheiss for the Lhx2/9 antibody, and Daniel Lazar for help with the tissue culture. We thank Jeniffer Darnell (Rockefeller University) for the hFMRP-EGFP plasmid.

The peer review history is available online at https://dev.biologists.org/lookup/doi/10.1242/dev.188797.reviewer-comments.pdf