Dicer maintains the identity and function of proprioceptive sensory neurons

We have demonstrated that selectively impairing Dicer in parvalbumin-positive neurons, which include the proprioceptors, triggers behavioral changes, a lack of muscle connectivity, and a loss of transcriptional identity as observed through RNA sequencing. These results suggest that Dicer and, most likely by extension, microRNAs are crucially important for maintaining proprioception. Additionally, this study hints at the larger question of how neurons maintain their functional and molecular specificity.

Abstract

Neuronal cell identity is established during development and must be maintained throughout an animal’s life (Fishell G, Heintz N. Neuron 80: 602–612, 2013). Transcription factors critical for establishing neuronal identity can be required for maintaining it (Deneris ES, Hobert O. Nat Neurosci 17: 899–907, 2014). Posttranscriptional regulation also plays an important role in neuronal differentiation (Bian S, Sun T. Mol Neurobiol 44: 359–373, 2011), but its role in maintaining cell identity is less established. To better understand how posttranscriptional regulation might contribute to cell identity, we examined the proprioceptive neurons in the dorsal root ganglion (DRG), a highly specialized sensory neuron class, with well-established properties that distinguish them from other neurons in the ganglion. By conditionally ablating Dicer in mice, using parvalbumin (Pvalb)-driven Cre recombinase, we impaired posttranscriptional regulation in the proprioceptive sensory neuron population. Knockout (KO) animals display a progressive form of ataxia at the beginning of the fourth postnatal week that is accompanied by a cell death within the DRG. Before cell loss, expression profiling shows a reduction of proprioceptor specific genes and an increased expression of nonproprioceptive genes normally enriched in other ganglion neurons. Furthermore, although central connections of these neurons are intact, the peripheral connections to the muscle are functionally impaired. Posttranscriptional regulation is therefore necessary to retain the transcriptional identity and support functional specialization of the proprioceptive sensory neurons.

NEW & NOTEWORTHY We have demonstrated that selectively impairing Dicer in parvalbumin-positive neurons, which include the proprioceptors, triggers behavioral changes, a lack of muscle connectivity, and a loss of transcriptional identity as observed through RNA sequencing. These results suggest that Dicer and, most likely by extension, microRNAs are crucially important for maintaining proprioception. Additionally, this study hints at the larger question of how neurons maintain their functional and molecular specificity.

the neuronal diversity characteristic of the mature central and peripheral nervous systems arises through progressive stages of proliferation, migration, and differentiation, tightly regulated by transcriptional and posttranscriptional mechanisms. Less understood is how neuronal cell types retain their identity after they have differentiated and matured. The identities of specific classes of cells are at least partially determined by transcription factor codes and increasing evidence indicates that these same transcription factors are necessary to maintain identity later in life (Deneris and Hobert 2014). MicroRNAs are posttranscriptional regulators that selectively inhibit the translation, stability, or polyadenylation of specific transcripts through complimentary basepair recognition encoded within seed regions of the mature microRNAs (Bartel 2009; Krol et al. 2010). Like transcription factors, they have been shown to be crucial for several early developmental transitions that establish cell-type specificity (Huang et al. 2010; Makeyev et al. 2007; Zhao et al. 2009). Less is known, however, about whether this class of molecules, which is highly abundant within the nervous system (Kosik 2006), may also work to reinforce and maintain neuronal identity after it has been established.

To study the mechanisms that maintain neuronal identity, we focused on a highly characterized cell type. Proprioceptive sensory neurons, located in the dorsal root ganglion, encode changes in muscle length and tension through innervation of muscle spindles and Golgi tendon organs, specialized compartments in skeletal muscle (Windhorst 2007). Three distinct subtypes of proprioceptive neurons differ in their central connections with spinal neurons and associated peripheral end organs. The intrafusal fibers in muscle spindles are supplied by primary (group Ia) and secondary (group II) afferents, while Golgi tendon organs are innervated by group Ib afferents. These endings in muscle as well as synaptic terminals in the spinal cord can be visualized by expression of the vesicular glutamate transport 1 (VGluT1; Wu et al. 2004). In particular, Ia afferents make strong synaptic connections with motor neurons in the spinal cord innervating homonymous muscle groups (Windhorst 2007). Proprioceptive sensory neurons can be genetically targeted through their common expression of the calcium binding protein Pvalb (Arber et al. 2000). Disruption of proprioceptor central or peripheral connections triggers easily observable behavioral phenotypes, demonstrated by impairments of neurotrophic or transcription factors such as TrkC, Nt3, Etv1, or Runx3 (Arber et al. 2000; Chen et al. 2002; Ernfors et al. 1994; Klein et al. 1994; Levanon et al. 2002). Additionally, a more comprehensive transcriptional profile of proprioceptors is emerging (Li et al. 2016; Usoskin et al. 2015). Given this functional and genetic knowledge base, proprioceptive sensory neurons lend themselves to the study of how neuronal identity is maintained.

MicroRNAs are crucial for numerous aspects of sensory neuron development, such as determining left/right functional asymmetery in Caenorhabditis elegans taste receptors (Johnston and Hobert 2003), controlling the number of sensory neuron precursors in drosophila (Li et al. 2006), and ensuring proper inner ear development in mice (Soukup et al. 2009). MicroRNAs, averaging 22 bp in size, are generated from longer transcripts derived from intronic or intergenic regions. They are further processed by the Drosha/DGCR8 complex or by the splicesome, transported out of the nucleus, and then terminally processed by the microRNA biogenesis enzyme Dicer (Krol et al. 2010), with one notable exception being mir-451, which uses Argonaute-2 for its terminal processing step (Cheloufi et al. 2010; Cifuentes et al. 2010; Yang et al. 2010). To better tackle how microRNAs contribute to cell identity in proprioceptive sensory neurons, we conditionally ablated Dicer (Harfe et al. 2005) through a Pvalb Cre driver (Hippenmeyer et al. 2005). By conditionally ablating Dicer from Pvalb-positive neurons, we can ask how proprioceptive sensory neuron identity is affected by global impairment of posttranscriptional regulation. Dicer knockout (KO) animals displayed a clear behavioral decline, reminiscent of other proprioceptive mutants. This decline in behavior was accompanied by cell death within the ganglion. Before the onset of cell death, there was a loss of transcriptional identity as well as reduced transmission at peripheral but not central connections.

MATERIALS AND METHODS

Mice and Genotyping

All mice were acquired from Jackson Laboratories. All procedures for animal experiments were approved by the Brandeis University and Wright State University Animal Care and Use Committees. The tdTomato reporter line Ai9 (Madisen et al. 2010; stock no. 007909) was mated to floxed Dicer animals (Harfe et al. 2005; stock no. 006001) along with parvalbumin-ires-Cre mice (Hippenmeyer et al. 2005; stock no. 008069). After several generations, Pvalb^cre/cre, Ai9^+/+, and Dicer^flx/wt parents were bred to generate experimental animals, of either sex, that were either homozygous or wild type (WT) for the floxed allele. All genotyping was done using the extract and amp PCR kit (Sigma-Aldrich) with the primers and PCR programs recommended by Jackson Laboratories for each line.

Behavioral Analysis

Animals were observed daily using two behavioral tests to examine motor function from postnatal days (p) 20–30. First, each animal was subjected to a cylinder test (Brooks and Dunnett 2009; Fleming et al. 2013). Briefly, mice were placed into a small transparent cylinder (diameter: 12.7 cm; height: 15.5 cm) and were scored for the number of rears during a 3-min observation. Rears were scored when the mouse stood on its hindlimbs and pressed both forepaws against the side of the glass. Subsequently, each mouse was subjected to a modified form of footprint analysis (Brooks and Dunnett 2009; Carter et al. 1999) using a custom-built, elevated linear track (length: 26 cm; width: 5 cm). The bottom of the track was transparent, enabling us to film the animals with an iPhone video camera using a minimum frame rate of 120 fps. Gait patterns were analyzed using FIJI. The central position of each paw placement was used to measure the average distance between left and right limbs for each static position of the animal (width) and the average forward movement for each step (stride; Fig. 1C). Statistical analyses were performed in R using a two-way repeated-measures ANOVA. Additionally, post hoc tests were performed using a Student’s t-test with a Šidák correction.

Fig. 1.

Conditional ablation of Dicer in parvalbumin-positive (Pvalb⁺) neurons elicits progressive ataxia. WT, wild type; KO, knockout. A: photos of Pvalb^cre/cre, Ai9^+/+, Dicer^wt/wt, WT mice (top) and Pv^cre/cre, Ai9^+/+,Dicer^flx/flx, KO (bottom). B: number of rears per 3-min session at postnatal day (p) 30 for both WT (n = 8, black) and KO (n = 6, red). Error bars are SD. *Ages at which t-test of P < 0.05 after a Šidák correction. C: example images of mice walking on a linear track while filmed from below (left: KO; right: WT). Paw patterns were measured for hind- and forelimbs, and separated into horizontal (stance width) and vertical (stride length) components. D: stance width (mm), defined as the average difference in the x-axis between either the fore or hind paws, increased progressively between p20 and p30, but the increase was much greater for KO than WT. E: WT stride length, calculated as the average distance (in the y-axis) between movements within a limb, increased over the same developmental period, but KO stride length began to decrease after p25.

DRG Cell Count and Area Analysis

Tissue collection.

At p20, p25, and p30, animals were deeply anesthetized with isofluorane, followed by a mixture of ketamine and xylazine (equal to or greater than 80 and 10 mg/kg, ip) and transcardially perfused with ice-cold phosphate buffered saline (PBS) followed by 4% paraformaldehyde in PBS. The spinal column with some muscle and portions of the rib cage still attached was pinned down onto a Sylgard-coated glass dish containing ice-cold PBS and a laminectomy was performed under a dissection microscope. The spinal cord was carefully removed while keeping the dorsal root ganglion attached to the remaining portions of the vertebrae. The lumbar 4 dorsal root ganglion (L4 DRG) was identified relative to lumbar 1, the first ganglion past the rib cage. The DRG was placed in PBS and the roots were partially trimmed. It was then placed in ice-cold 4% PFA for 1 h at 4°C, subjected to gentle rotation, and protected from light. The DRG was washed (3×) in 30% sucrose and suspended in PBS overnight at 4°C. Ganglia were equilibrated in Tissue-TEK OCT compound (VWR), for at least 20 min, and were then flash frozen in 2-methylbutane with CO₂ pellets and sectioned at 14 μM on a Leica CM3050 cryostat. Sections were adhered to Fisherbrand Colorfrost Plus slides.

Staining and quantification.

Nissl-stained slides (NeuroTrace 640/660 Deep-Red Fluorescent Nissl Stain; Invitrogen) were counterstained with DAPI (Vector Laboratories), mounted (Vectashield HardSet Mounting Medium). and imaged (Leica SP5 confocal microscope) within several days of staining to prevent decreased signal and increased background. Total numbers of tdTomato⁺ and Nissl⁺ neurons were counted in FIJI in at least five sections. The average ratio of tdTomato⁺ neurons to Nissl⁺ cells was compared across ages using a two-way ANOVA followed by a post hoc Tukey test in R. Cell areas were measured in thresholded tdTomato images using the FIJI analyze particles plugin (circularity = 0.3, area > 100 μm²). Overlapping cells were excluded. Area measurements were excluded from one animal due to high background. Comparisons of the mean and overall distributions for KO and WT were performed in R using a Welch’s t-test and a Kolmogorov-Smirnov test, respectively.

RNA Sequencing and Expression Analysis

RNA collection and library construction.

Both L4 DRGs were dissected from animals perfused with PBS (without PFA) and were placed into a 1.5-ml low retention microtube (Phenix Research Products), containing 1 ml 1× PBS with 1 mg type XIV protease and 1 mg of collagenase (Sigma-Aldrich) and then placed in a 37°C water bath for 10 min. Subsequently, L4 DRGs were subjected to manual cell sorting as previously described (Hempel et al. 2007; Sugino et al. 2006). DRGs were washed in ice-cold artificial cerebral spinal fluid (ACSF; containing in mM: 126 NaCl, 3 KCl, 1.25 NaH₂PO₄, 20 dextrose, 20 NaHCO₃, 2 MgSO₄, 2 CaCl₂, 0.05 APV, 0.02 DNQX, and 0.0001 TTX pH 7.35; osmolarity = 320 mosM; 1% fetal bovine serum). The ganglia were triturated using fire-polished Pasteur pipettes with successfully smaller inner diameters (~600, 300, and 150 μm) and then diluted in 20 ml of ACSF. Individual tdTomato⁺ neurons were sorted manually using a micropipette with an inner diameter of 30–50 μm. Fluorescent cells were then passed through three subsequent petri dishes to increase sample purity and to dilute any RNA released during dissociation. For each sample, ~150 to 200 neurons were sorted, transferred into 50 μl of extraction buffer, processed using the picopure RNA isolation kit (Life Technologies), and subjected to on-column DNAse digestion. RNA samples were amplified using the Ovation RNA-seq system (NuGEN). The cDNA was sonicated into ~200-bp fragments using a Covaris S 220 Shearing Device and then converted into sequencing libraries with the Ovation Rapid DR Multiplex System (Nugen). RNA library concentration was quantified with a qPCR-based Library Quantification Kit (KAPA biosystems) and then sequenced on an Illumina Nextseq.

Expression analysis.

Sequencing data was trimmed and quality filtered with cutAdapt (Martin 2011). Subsequently ribosomal, mitochondrial, and low complexity reads were removed and the data were mapped using RSEM (Li and Dewey 2011). Expression of previously described proprioceptor- and other DRG sensory neuron-enriched genes (Usoskin et al. 2015) was filtered to remove those expressed at very low levels (sum of KO and WT expression <2 transcripts per million) or those that were common to both sets. Expression changes were calculated as the difference between KO and WT samples divided by the sum. Monte Carlo simulations were performed using in-house software written in python and R to assess the likelihood of observing changes in the expression of enriched genes by chance. For both lists, the summed expression changes were compared with summed expression changes of randomly selected genes, matching the approximate WT expression level of the enriched genes. Distributions of expected expression changes for expression matched gene lists were generated from 1,000,000 points. The accession number for the RNAseq data is GSE86019.

Spinal Cord and Muscle Physiology

Spinal cord preparation.

p21 control (Dicer WT and heterozygous) and mutant animals were used for both ventral root recording and muscle/nerve preparations. Animals were anesthetized and transcardially perfused with 5 ml of ice-cold oxygenated (95% O₂–5% CO₂)ACSF (containing in mM: 127 NaCl, 1.9 KCl, 1.2 KH₂PO₄, 1 MgSO₄·7H₂O, 26 NaHCO₃, 16.9 D(+)-glucose monohydrate, and 2 CaCl₂). The spinal column and attached lower limbs were dissected free and immersed in a recirculating bath of cold (16–18°C), oxygenated ACSF. To improve preparation viability for ventral root recordings, dissection was performed in ACSF in which NaCl was replaced by equal osmolar sucrose (254 mM). The spinal cord was exposed via one laminectomy and hemisected to increase its oxygenation (Jiang et al. 1999). The preparation was transferred to a recording chamber and allowed to recover for 1 h in recirculating, oxygenated standard ACSF while the temperature was slowly brought to the recording temperature of 25°C.

Ventral root recordings.

The ventral root of the fifth lumbar (L5) segment was drawn into a glass capillary and sealed against the ventral surface of the spinal cord by applied suction (A-M Systems, Sequim, WA). Stimulation of the dorsal root of the same segment (L5) was accomplished via a constant current stimulator (0.1-ms pulse duration, A365 stimulus isolator; World Precision Instruments, Sarasota, FL). Threshold intensity required to evoke ventral root potentials ranged from 8 to 15 µA. A standardized stimulation intensity of two times the threshold was used in these experiments. Differential recordings of ventral root potentials were recorded with an EX4–400 quad channel amplifier (1,000× gain, 2-Hz low cut, 500-Hz high cut; Dagan, Minneapolis, MN) and digitized at 20 kHz with WinLTP software (WinLTP, Bristol, UK). Peak amplitude, peak latency, and maximum slope of the rising phase of the response were determined offline using analysis tools in WinLTP (Anderson and Collingridge 2007). Student’s t-test was used to compare genotype responses. Mean responses ± SD are presented.

Muscle and nerve preparation.

Initial dissection procedures were similar to those for ventral root recordings preparations, but sucrose-replaced ACSF was not required. The nerve supply to the rectus femoris muscle of the quadriceps group was maintained up to the level of the third and fourth lumbar roots (L3 and L4). All other muscles were dennervated. Tendons anchoring the rectus femoris on the ischial bone were preserved intact, while the tibia bone was cut free at the region of the insertion of the patellar tendon. Numerous minutiae pins were placed to stabilize the femur and hip bone to minimize compliance of the joints during muscle stretch or contraction. A 6–0 silk suture was tied at the distal end of the patellar tendon and attached to the lever arm of a force-transducing micro-stepping motor (300C dual-mode muscle lever; Aurora Scientific, Ontario, Canada). Movements were controlled and digitized with Spike2 software (CED, Cambridge, UK).

The spinal cord was cut away in these preparations, but the peripheral ends of both dorsal and ventral roots were placed in suction electrodes. The force-length relationship was determined for each preparation to find the level of baseline tension (a function of muscle length) that generated the largest force during contraction. Muscle contraction was initiated by suprathreshold stimulation of motor axons innervating the rectus femoris. In our experience, the majority of motor axons supplying quadriceps muscles traveled through the L3 ventral root, and we found the greatest muscle force (monitored by the force-transducing motor) was generated when stimulating this root. One-second vibration sequences were repeated three times with 3- to 5-s intervening rest periods. A variety of amplitude (20, 40, 80, and 120 µm) and frequency (10, 20, 50, and 100 Hz) combinations were used with background stretches of ~110% of the baseline tension. Recordings of rectus femoris sensory responses were made via a suction electrode placed on L4 dorsal root (DRL4). We observed sensory axons supplying the quadriceps were found in both DRL3 and DRL4, but the majority of axons pass through DRL4. Recordings were made at room temperature (21–23°C). Analysis was performed offline using custom scripts in Spike2 and MATLAB (The MathWorks, Natick, MA). Main effects and interactions between genotype, vibration frequency, and vibration amplitude were analyzed with a three-way ANOVA in SPSS (version 23, IBM). Data are presented as means ± SD.

Muscle Immunohistochemistry

Tissue Collection.

P30 mice were anesthetized with isofluorane and ketamine/xylazene and the gastrocnemius and tibialis anterior were dissected. For VGluT1 staining of proprioceptive sensory endings, animals were transcardially perfused with ice-cold PFA, followed by PBS. Muscles were fixed in PFA overnight at 4°C and then washed (3 times) and equilibrated overnight in 30% sucrose, and then embedded in Tissue-TEK O.C.T. compound (VWR).

Immunhistochemisty.

Muscles stained for S46 (intrafusal fibers) were flash frozen and not fixed. All sections were cut longitudinally at 60 μm (S46) or 35 μm (VGluT1) and were mounted on Fisherbrand Colorfrost Plus slides. Intrafusal fibers were visualized as previously described (Taylor et al. 2005) with some modifications. Sections were blocked with 0.5% porcine gelatin (Sigma-Aldrich) and 1.5% goat serum (Invitrogen) in 1% Triton-X-100 (Fisher) in Superblock buffer (Thermo Scientific) for at least 2 h. Muscle tissue was stained overnight at 4°C with a mouse monoclonal S46 antibody (Developmental Studies Hybridoma Bank), along with a chicken anti neurofilament-h polyclonal antibody (Aves Laboratories), at 1:50 and 1:1,000, respectively. Sections were washed in PBS and stained using goat anti-mouse Alexafluor 546 and goat anti-chicken Alexafluor 488 (Invitogen) secondary antibodies at a 1:500 dilution, followed by three washes in PBS. Slides were mounted using Vectashield HardSet Mounting Medium with DAPI (VWR). The staining procedure for vGluT1 was similar except sections were incubated in the primary antibody and guinea pig anti-vGluT1 (Millipore) at 1:1,000 for at least 2 days at 4°C. Secondary incubations were performed at room temperature for 3 h with rhodamine red goat anti-guinea pig (Jackson ImmunoResearch). Subsequently, sections were washed four times for 30 min. Statistics were done in R using a Welch’s t-test.

RESULTS

Parvalbumin-Driven Conditional Ablation of Dicer Leads to Progressive Ataxia in the Fourth Postnatal Week

To better understand the role of Dicer in proprioception, Pvalb-Ires-Cre mice (Hippenmeyer et al. 2005) were mated to floxed Dicer animals (Harfe et al. 2005), and, in most cases, to a conditional reporter line, Ai9 (Madisen et al. 2010), for visualization and isolation of the Pvalb positive population. Most experimental animals were generated from matings between Pvalb^cre/cre, Ai9^+/+, Dicer^flx/wt parents, to generate animals that were either WT or homozygous for the floxed allele. These animals displayed a number of noticeable phenotypes, as observed previously (Valdez et al. 2014), Dicer KOs displayed clear ataxia, easily noticeable at p30 (Fig. 1A). Additionally, as early as p20 KO animals were hyperactive and occasionally displayed circling behavior. Also, KO animals were typically smaller and more emaciated than their WT littermates. Almost all KO animals appeared to be scoliotic and some animals experienced epileptic bouts that would occasionally lead to death. However, these animals were also highly uncoordinated in a manner similar to, but perhaps more severe than, that seen when proprioceptive sensory neurons conditionally express diptheria toxin (Akay et al. 2014). These animals were also reminiscent of the Etv1 and Neurotrophin3 mutants (Arber et al. 2000; Ernfors et al. 1994) as well as mice lacking Piezo2 in proprioceptive sensory neurons (Woo et al. 2015). Animals were born in normal Mendelian ratios becoming progressively more symptomatic and displaying reduced weight that worsened during the fourth postnatal week (data not shown); additionally, animals rarely survived after age p70 without the aid of food pellets and hydrogel on the cage floor. To better characterize the onset of ataxia, two behavioral assays were performed. First, rearing behavior was assayed in a transparent cylinder. WT animals reared at a roughly equal rate across ages, while KOs after some initial hyperactivity exhibited no rearing beginning at age p27 (Fig. 1B). A repeated-measures ANOVA suggested that the effect of age was not significant (P = 0.058), while genotype as well as an interaction between genotype and age were significant (P = 1.4 × 10⁻⁴, 2.1 × 10⁻³) suggesting that loss of Dicer impairs the use of hindlimbs for rearing and that the deficit increases with age. Furthermore, a post hoc t-test with a Sidák correction suggested significant differences at p26–28 (P = 3.3 × 10⁻³) and p29–30 (P = 6.7 × 10⁻³). The gait of WT and KO animals was characterized by filming the movements of animals on a clear linear track (Fig. 1C) and measuring the position of their fore- and hindlimbs. This enabled the quantification of both the average stance width (Fig. 1D), the separation between forelimbs or hindlimbs in the x-axis (example in Fig. 1C), and stride length (Fig. 1E), the forward distance traveled by a single limb in the y-axis. A repeated measures ANOVA for forelimb stance width showed significant effects for genotype and age as well as an interaction between these variables (P = 0.03, 1.23 × 10⁻⁵ , 8.86 × 10⁻⁴). Genotype and age effects were also significant for the hindlimbs but showed no interaction (P = 0.046, 0.011,0.067). A post hoc test suggested that only the p27 hindlimb and p30 forelimb results were significant (P = 1.9 × 10⁻³, 1.1 × 10⁻⁴). Changes in hindlimb stride length were not significant for genotype but were significant for age as well as an interaction between age and genotype (P = 0.09, 2.47 × 10⁻³, 9.56 × 10⁻⁵). Lastly forelimb stride length was significant for genotype, age and an interaction between these two (P = 0.039, 2.22 × 10⁻³, 9.66 × 10⁻⁵). Post hoc tests suggested that hindlimb and forelimb effects were significant at both p27 (P = 5.5 × 10⁻³, 3.5 × 10⁻³) and p30 (P = 2.2 × 10⁻⁴, 1.7 × 10⁻⁵). This data suggest that loss of Dicer leads to a progressive ataxia that largely incapacitates the animals by the end of the fourth postnatal week.

Progressive Ataxia is Accompanied by a Loss of Proprioceptive Sensory Neurons in the Dorsal Root Ganglion

Since deletion of Dicer in other cell types can lead to cell death (Davis et al. 2008; Schaefer et al. 2007; Zehir et al. 2010) and since the behavioral phenotype was reminiscent of mutations affecting proproprioceptive sensory neurons within the DRG (Arber et al. 2000; Ernfors et al. 1994; Tourtellotte and Milbrandt 1998); we decided to look at the number of tdTomato-positive neurons labeled by Cre activity within the lumbar 4 DRG, a segmental level carrying sensory input from many hindlimb muscles, at p20, p25, and p30. DRGs were isolated, cryosectioned, and then Nissl stained, and the numbers of tdTomato-positive neurons per section was counted. Due to variability between sections and animals we normalized the number of tdTomato-positive neurons to the number of Nissl-positive cells within each section to quantify the fractional change in the labeled population (examples shown in Fig. 2A). The results reveal that the fraction of tdTomato-positive neurons decline with age (Fig. 2B). A two way ANOVA demonstrated a significant effect for genotype (P = 2.62 × 10⁻⁴) and a post hoc Tukey test showed that the effect of genotype was insignificant at ages p20 and p25 (P = 0.95, 0.34) but was significant at p30 (P = 2.9 × 10⁻³). It should also be mentioned that as expected there were no significant differences across ages for the Nissl-positive tdTomato-negative population (WT 178.4 ± 35.3, n = 16; KO 162.7 ± 53.0, n = 16, t-test: P = 0.33). Since Pvalb⁺ neurons vary in size, the distribution of cross-sectional areas was measured at each age. Although, the area distributions are slightly different (Fig. 2C), as shown with a Kolmogov-Smirnov test (P = 0.041), the mean values for each animal were not significantly different (WT, n = 16, mean = 564.7 ± 25.8; KO, n = 16, mean = 557.5 ± 18.9; P = 0.82). These data suggest that cell loss occurs roughly equally for all subgroups within the Pvalb⁺ population. Therefore, over time, conditional ablation of Dicer within Pvalb⁺ cells leads to a loss of the proprioceptive sensory neurons.

Fig. 2.

Dicer KO animals progressively lose Pvalb⁺ neurons within the dorsal root ganglion (DRG). A: example images of L4 DRG cryosections from WT and KO animals at p20, p25, and p30. Cre-activated tandem dimer tomato (tdTomato) labeled neurons are shown in green, Nissl counterstain in magenta. Percentage of tdTomato⁺ neurons (of total in section) indicated (bottom left); scale bar (bottom right) = 100 µm. B: average %tdTomato⁺ neurons for WT (black; n = 4,5,7) and KO (red; n = 5,5,7) for p20, p25, and p30; error bars are SD and individual data points are represented by red dots. A two way ANOVA determined that the effect of genotype was significant (P = 2.6 × 10⁻⁴). Post hoc Tukey test demonstrated a significant reduction at p30 (P = 2.9 × 10⁻³) but not at p20 or p25 (P = 0.951, 0.167). C: cumulative probability distributions of cell areas for WT (16 animals, 2,840 cells) and KO (16 animals, 2,081 cells).

A Loss of Transcriptional Specificity Precedes Proprioceptor Cell Death

To gain insight into the etiology of ataxia in KO animals, proprioceptive sensory neurons were isolated from lumbar 4, before the onset of cell death, at p21, and then subjected to RNA extraction and deep sequencing. It has been reported in retinal pigment epithelial cells, that loss of Dicer can cause global transcriptional misregulation and cell death as a result of runaway accumulation of transcribed Alu elements (Kaneko et al. 2011). To assess whether or not loss of Dicer in proprioceptive neurons causes accumulation of Alu elements or other transcribed retrotransposons, we aligned the sequencing data to the repeat masker track from the University of California, Santa Cruz (UCSC) genome browser and normalized the read count for each motif to the total number of mapped reads. This revealed no significant differences in the abundance of SINE elements or of other categories of retrotransposons in KO vs. WT mice.

Having ruled out retrotransposons, the mRNA profiles of the KO cells were examined. Reads were mapped with RSEM (Li and Dewey 2011) and differential expression was assessed using edgeR with a false discovery rate of 5% and a minimum baseline expression value of 20 transcripts per million (TPM). Overall, 2146 genes were downregulated while 1728 were upregulated in the KO. The 15 up- and downregulated genes displaying the largest fold changes are shown in Fig. 3A. More than half (8 of 15) of the highly downregulated genes (Pln, Tuba8, Relt, Heatr5a, Wls, Vstm2b, Kcnc1, and Nxph1) are highly enriched in proprioceptive sensory neurons relative to other neurons of the DRG (Usoskin et al. 2015). This prompted us to examine other genes strongly linked to proprioceptor survival or function. Etv1, TrkC, Runx3, and Sad-B were all downregulated (fold changes: −1.83, −3.39, −1.39, and −1.43). Piezo2, Tmem150C and Asic3 were either upregulated or unchanged (fold changes: 1.17 1.07, and 2.17) and Pea3 (Etv4) was expressed at very low levels (<1 TPM) in both WT and KO. Together, these data suggest a widespread loss of proprioceptor-enriched genes.

Fig. 3.

Pvalb⁺ sensory neurons display a loss of propriocepor-specific gene (PSG) expression identity. A: heat map of the top 15 up- and downregulated genes whose expression values were greater than 20 transcripts per million (TPM); the expression value for each biological replicate (n = 3) is normalized to the maximum value within the column and calculated as the log base 2 of that value plus 1; the color map is shown at the bottom. B: an additional heat map of expression values for selected PSGs (Usoskin et al. 2015), n = 3 for both genotypes. C: Monte Carlo simulation showing that downregulation of PSGs, (−0.38, arrow) is greater than that for randomly selected genes matched for expression level (mean = −0.035; SD = 0.019, n = 10⁶). D: as in A except for genes selectively enriched in DRG sensory neurons excluding proprioceptors. E: as in C but for nonproprioceptor genes and random selections matched to these expression levels. Modest upregulation of these genes (0.078, arrow) is greater than expected by chance for randomly selected genes (mean = −0.022; SD = 0.012, n = 10⁶).

To more quantitatively characterize the effect of Dicer KO on transcriptional identity, we used data from a previously published study (Usoskin et al. 2015), in which DRG sensory neurons were clustered into 11 groups, including 2 proprioceptive groups. We aggregated the top 50 genes that best defined each cluster in Usoskin et al. (2015) (after removing genes expressed at very low levels in both WT and KO) into two nonredundant lists of genes, enriched in proprioceptors (n = 69) and nonproprioceptors (n = 251), respectively. Proprioceptor enriched genes were much more likely to be downregulated in the KO neurons (Fig. 3B). To assess the likelihood that this downregulation could occur by chance, we performed a Monte Carlo simulation by repeatedly calculating expression changes for other randomly selected genes expressed at similar levels in proprioceptors but not identified as proprioceptor enriched. The observed expression change (EC) of the proprioceptor group (Fig. 3C, EC = −0.38) greatly exceeded that observed by chance (mean = −0.035; SD = 0.019, n = 10⁶). Conversely, genes enriched in other sensory neurons were largely upregulated (Fig. 3D) and this value (Fig. 3E, EC = 0.078) was more than eight SD from the mean of the randomized distribution (mean = −0.022; SD = 0.012, n = 10⁶). These data suggest a loss of cell-type specific identity for the proprioceptive sensory neurons in response to Dicer KO.

Group Ia Sensory Neurons Maintain Their Connection with Motor Neurons Despite Dicer KO

To better understand the functional consequences of the KO, synaptic connections between Ia proprioceptive afferents and motor neurons in the spinal cord were examined through measurements of ventral root potentials (Fig. 4A) following dorsal root stimulation using an ex vivo preparation (Mears and Frank 1997). Ventral root responses increased with stimulus intensity from threshold (T), the current required to evoke a measurable response in the ventral root recordings, to 2×T and then plateaued (data not shown). Consequently, comparisons were made with data from 2×T stimulation. Unexpectedly, motor neuron responses were not reduced in KOs (Fig. 4B). While differences in peak amplitude (control 67.2 ± 21.5 μV, n = 5; KO 107.9 ± 54.9 μV, n = 8, t-test: P = 0.09) and maximum slope of the rising phase of the response (control 21 ± 10 μV/ms, n = 5; KO 51 ± 40 μV/ms, n = 8, t-test: P = 0.08) were not significant, the data suggest responses in KOs may, if anything, be larger than in control animals. The peak latencies were comparable in control (5.8 ± 0.9 ms, n = 5) and KO (5.4 ± 1.5 ms, n = 8; t-test: P = 0.577) animals, suggesting central action potential conduction and synaptic delays are not affected in KOs. These data suggest that any functional impairment before cell loss is not present at central proprioceptor synapses and suggest the phenotype of the KO animals may be largely driven by a deficit in the periphery.

Fig. 4.

L5 motor neuron responses recorded in the L5 dorsal root following stimulation of sensory afferents. A: representative average traces (15 individual sweeps at 0.2 Hz) from control and mutant preparations illustrating compound motor neuron responses. Arrow highlights depolarization immediately preceding synaptic potential caused by synchronized sensory axon action potentials arriving in ventral horn of spinal cord. This potential is unaffected by high-frequency sensory stimulation, while the synaptic potential is depressed (data not shown). B and C: peak amplitude and maximum slope measurements for control (triangles, n = 5) and KO (circles, n = 8) animals (bars indicate means ± SD). Stimulation intensity for all data shown was 2 times threshold intensity required to elicit a measurable ventral root response.

By postnatal day 30 VGluT1+ sensory endings are largely absent from the muscle

Considering that the gene expression data demonstrated a loss of cell type-specific expression (Fig. 3), relative to other DRG sensory neurons, and that proprioceptive sensory neurons are distinguished from other sensory neurons by their ability to innervate and transduce information about the muscles, sensory neuron connectivity to intrafusal muscle fibers was assessed using immunohistochemistry. The experiments were performed at p30 when all KO animals exhibited a robust phenotype. Since Pvalb is expressed postnatally in extrafusal muscle fibers (Celio and Heizmann 1982), it was necessary to use another marker to determine if the intrafusal fibers, the targets of the Ia afferents, were still present, as a lack of these structures might explain the behavioral deficits we observed in the KO animals (Fig. 1). These intrafusal fibers can be specifically labeled with the S46 antibody (Miller et al. 1985), which targets a specialized form of myosin (Kucera and Walro 1995). Annulospiral endings were also visualized using the neurofilament-h antibody in conjunction with S46 (Fig. 5A). S46⁺ fibers were readily observed in both WT and KO animals (Fig. 5, B and C). S46 counts were performed on both the gastrocnemius as well as the tibialis anterior to rule out the possibility that conditional Dicer ablation might have differential effects on antagonistic muscle groups as seen in the ETV1 mutant (de Nooij et al. 2013). The observations suggested that intrafusal fibers were largely unaffected (Fig. 5G) in the tibialis (P = 0.904) as well as the gastrocnemius (P = 0.255). Although large differences in the weight of both muscles were observed (Fig. 5I, P = 2.28 × 10⁻⁵, 0.01), these differences paralleled differences in the weight of KO animals. To ask if the loss of muscle mass might be related to malnourishment and/or some global problem with growth we normalized muscle weights to body weights (Fig. 5J) and found that if body mass is taken into account there were no significant differences in both muscle groups (P = 0.11, 0.56), suggesting that the loss of muscle mass was driven globally, most likely by malnourishment, rather than by some process intrinsic to the muscles themselves. These data suggest that the intrafusal fibers remain intact upon conditional ablation of Dicer.

Fig. 5.

Hindlimb muscles lose proprioceptive connectivity despite conservation of intrafusal fibers at p30. A: example image of an annulospiral ending from a WT p30 gastrocnemius muscle, intrafusal fiber labeled by S46 antibody (magenta) sensory axon labeled with neurofilament-h antibody (green); scale bars = 100 μm. B and C: example WT and KO S46-positive fibers at p30. D: close-up image of Vglut1 image from a p30 animal showing adjacent annulospiral endings. E: lower magnification of 2 spindle associated sensory endings (SSEs) in WT. F: an example section from a KO muscle displaying no SSEs. G: S46 fibers per section for the gastrocnemius and tibialis anterior muscle groups for WT (black; n = 6) KO (white; n = 5) are not significantly different (P = 0.90, 0.25); error bars are SD, while red dots show individual data points. H: the average number of spindle-associated sensory endings (SSEs) per muscle section was determined for the WT (black, n = 3) and KO (white, n = 3), in the tibialis anterior, and gastrocnemius WT (n = 6), KO (n = 7). Both differences were significant (P = 5 × 10⁻⁵, 0.025). Muscle weights (mg) for both WT (n = 6) and KO (n = 7) were significantly different for the tibialis and gastrocnemius (P = 2.3 × 10⁻⁴, 0.01) (I) but were not significant when normalized to body weight (muscle weight/body weight)and expressed as a percentage of the average normalized WT value (J; P = 0.11, 0.5455).

To further evaluate the effect of conditional ablation of Dicer on proprioceptive sensory neuron connectivity to the muscle, we examined VGluT1 expression in the tibialis and gastrocnemius. VGluT1 specifically labels proprioceptive sensory neurons at the site of their innervation of intrafusal fibers (Wu et al. 2004). Although VGluT1 is thought to not be directly responsible for the mechanotransductive ability of these sensory neurons, it may serve a secondary signal-amplification role (Bewick et al. 2005). Disruption of staining for this marker can indicate a problem with proprioceptive connectivity in the muscle (de Nooij et al. 2013). Numerous VGluT1 positive sensory endings were observed in the WT (Fig. 5D) but staining was largely absent from the KO animals (Fig. 5, E and F). The number of VGluT1⁺ sensory endings observed per muscle section was steeply reduced (Fig. 5H) both in the tibialis (P = 5.41 × 10⁻⁵) and in the gastrocnemius (P = 0.025). These data support the conclusion that a Pvalb-driven KO of Dicer specifically affects sensory endings of proprioceptive sensory neurons while leaving their target intrafusal fibers intact.

Dicer Is Necessary for Maintaining Functional Ia Afferent Connectivity with Muscle Spindles

Having established that Ia afferents can make functional connections with motor neurons and that VGluT1 sensory afferents are absent in KO mice after the behavioral deficit is fully established, the peripheral sensitivity of these afferents was investigated at p21 before the onset of cell loss. Passive, small amplitude tendon vibrations selectively activate Ia afferents but do not drive secondary spindle or Golgi tendon organ afferents (Brown et al. 1967). We assayed the sensitivity of these afferents in the rectus femoris (a knee extensor) muscle using a modified ex vivo preparation (Wilkinson et al. 2012; Franco et al. 2014). Small amplitude stretches of the patellar tendon were delivered at various frequencies using a force-transducing motor and the compound responses of Ia afferents in the rectus femoris were measured by extracellular recordings at the L4 dorsal root.

A variety of amplitude (20, 40, 80, and 120 μm) and frequency (10, 20, 50, 100 and Hz) combinations were used to probe Ia afferent sensitivity to vibration in control and KO animals. Representative examples of afferent responses to 100-Hz vibration at 80-μm amplitude are shown in Fig. 6A. Compound action potentials were reliably observed with each vibration cycle in controls with little variability in amplitude throughout the 1-s vibration sequences. Deficits in KO animals were observed in two key parameters. Compound action potential amplitude was dramatically reduced in KOs animals at all vibration frequencies and amplitudes [Fig. 6, A and B; F(1,174) = 555.3, P < 0.0001]. Thus fewer Ia afferents responded to this natural stimulus in KOs. The reduced cohort of responding afferents also failed to encode every vibration cycle, with unresponsive cycles observed with even the largest amplitude vibrations [see Fig. 6A for example and Fig. 6B for quantification at all frequency and amplitude combinations; genotype effect F(1,174) = 129.9, P < 0.0001]. Encoding failures occurred less frequently at lower frequency vibrations (see responses at 20 and 10 Hz, for examples).

Fig. 6.

Group responses of proprioceptive neurons to small-amplitude tendon vibration. A: representative traces from control and mutant preparations illustrating responses to 6 cycles of 100-Hz vibration with an amplitude of 80 µm. Top trace: changes in muscle length during vibration. Middle and bottom traces: extracellular recordings of proprioceptive afferent responses measured at the L4 dorsal root in control and KO preparations, respectively. Note lack of response during some vibration cycles and overall diminished response amplitude in the KO trace. B, left: average responses for control (n = 6) and KO (n = 5) animals to 100-, 50-, 20-, and 10-Hz vibrations at varying amplitudes (±SD). Average compound action potential amplitudes increase with increasing vibration amplitude in control animals, but mutant responses are significantly reduced compared with controls at all vibration amplitudes and frequencies. B, right: average consistency of responses to vibration. Greater deficits in response fidelity are observed at lower vibration amplitudes (20 and 40 µm) in KO animals at all frequencies tested.

DISCUSSION

We have demonstrated that conditional ablation of Dicer in Pvalb⁺ neurons leads to a progressive ataxia beginning during the fourth postnatal week. Development of the ataxia is accompanied by a loss of tdTomato-labeled Pvalb⁺ neurons in the DRG. Before cell loss, however, these neurons lose aspects of their mature identity, showing a downregulation of proprioceptor-specific genes and an upregulation of genes that are specific to other sensory neurons within the ganglion. Although afferents remain connected to spinal neurons in the KOs, deficits appear in the peripheral connections of these cells, culminating in a near complete loss of the proprioceptive sensory ending marker VGluT1 during the fifth postnatal week. These data suggest Dicer is necessary to maintain the identity and function of proprioceptive sensory neurons.

Although we have focused on the role of proprioceptors, parvalbumin is also expressed in other parts of the motor system including Purkinje cells in the cerebellum, motorneurons, and a subset of interneurons within the spinal cord and the muscles themselves. There are several lines of evidence that lead us to believe that the observed ataxia results primarily from loss of function of proprioceptors rather than deficits in these other cell types. First, conditionally ablating Dicer solely in the parvalbumin-positive Purkinje cells with a Pcp2-Cre driver causes degeneration of the Purkinje neurons. However, cell loss and the resulting mild ataxia occur at several months of age rather than within the first several weeks (Schaefer et al. 2007). Second, a prior study found that Pvalb driven Dicer KO does not alter the initial formation and maintenance of neuromuscular junctions (Valdez et al. 2014), and we observed intact ventral root responses at p21 (Fig. 4) when compound responses of IA afferents are already highly abnormal (Fig. 6). Both of these findings argue against a primary defect in the motoneurons. Furthermore, some of the intrafusal fibers innervated by proprioceptors also express parvalbumin and therefore also potentially contribute to the behavior. Arguing against this is our observation that the S46 marker for intrafusal fibers remains stable. Finally, although the loss of Pvalb⁺ neurons and the downregulation of Etv1 expression (Fig. 3) are reminiscent of those observed following loss of muscle NT3 (Patel et al. 2003), this is not likely to occur following Pvalb-driven loss of Dicer. Of the intrafusal fibers, only bag fibers express NT3 (Copray and Brouwer 1994), but bag fibers do not express Pvalb (Celio and Heizmann 1982); therefore, we can rule out a cell autonomous involvement of Dicer in spindle NT3 secretion. Note that although it is unlikely that an initial malfunction in spindles causes loss of proprioceptor connectivity, this does not rule out the possibility that loss of connectivity may cause loss of trophic support of the proprioceptors. If a Dicer KO were to destabilize these connections, causing them to withdraw, this might then lead to a drop in TrkC signaling, due to a lack of contact with the NT3 secreting bag fibers, and a resulting increase in apoptosis and a decline in Etv1 expression.

Most directly and importantly, while this paper was being revised, Imai et al. (2016) reported similar ataxia and changes in proprioceptor survival and connectivity both following Dicer deletion using Pvalb-Cre and using Advillin-Cre (which targets trigeminal and DRG sensory neurons). This strongly argues against a major contribution from Dicer expressed in other structures including both extra- and intrafusal muscle fibers and argues that the motor symptoms reported here arise from a loss of Dicer in Pvalb-positive sensory neurons including proprioceptors.

Although this report focused on the ataxia phenotype, these animals display this was not the only deficit we observed. For example, the weight loss, circling behavior, hyperactivity, and bouts of epilepsy potentially point toward involvement of other circuits. Parvalbumin-positive neurons in structures such as the hypothalamus, inner ear, striatum, and neocortex, as well as others, could potentially contribute to aspects of the observed aberrant behavior.

The timing of the behavioral phenotype during the fourth postnatal week contrasts with mutants reported in the literature. Etv1 mutants display noticeable ataxia days after birth (Arber et al. 2000) and Runx3 mutants are also ataxic after birth. Although the precise onset is not reported (Levanon et al. 2002), these mutants do display a loss of proprioceptive afferents within the spinal cord at p0. Animals lacking TrkC, which is critical to proprioceptive sensory neuron survival, show abnormal movements as early as p4 (Klein et al. 1994). Additionally, knockout of Egr3, a gene essential for the induction of intrafusal fibers, produces a noticeable phenotype by p5 (Tourtellotte and Milbrandt 1998). The onset of these effects is much earlier than that observed in the Dicer KO described here, although we cannot exclude the possibility that more sensitive tests may have revealed problems at an earlier age. Although Pvalb is expressed relatively late in some brain structures, it occurs in the DRG as early as E16.5 (Hippenmeyer et al. 2005) and parvalbumin-Cre conditional ablation of Piezo2, the principal mechanotransduction channel of the proprioceptors, also produces abnormal motor behavior as early as p7 (Woo et al. 2015). It is also possible that the effects of impaired microRNA processing may be delayed, due to the stability of the microRNAs themselves (Guo et al. 2015). However, studies in other systems demonstrate that such delays often do not occur. For example, loss of Dicer can produce strong and rapid deficits in the development of dopaminergic neurons (Huang et al. 2010) and can increase neuronal apoptosis days after Dicer deletion through Emx1-Cre (De Pietri Tonelli et al. 2008). The results presented here suggest the possibility that the late phenotype of the Dicer KO animals reflects a role in late functional maturation or in maintenance of the adult identity, rather than a role in the initial differentiation and early survival attributed to genes giving rise to earlier phenotypes. What is also interesting is that the timing of the stance width effect appears different between the forelimbs and hindlimbs. This difference might reflect differences in developmental periods between these two sets of appendages. Finally, it is tempting to speculate that a functional role for Dicer in late development might involve balancing the expression, posttranscriptionally, of critical components involved in mechanotransduction at the periphery of the proprioceptive axons to maintain and reliably transduce useful information about the muscle groups they innervate.

One surprising feature of the late behavioral deficit is the degree to which the connections onto the muscle are impaired before a measureable difference in the gait. We observed a large difference in the group response of the proprioceptive sensory neurons during small amplitude vibrations of the rectus femorus at p21 but no significant changes in the behavior at this age. This could reflect central compensation for a peripheral deficit. Such compensation has been demonstrated in studies showing that chronically blocking transmission from sensory afferents elevates the ventral excitatory postsynaptic potential in response to dorsal root stimulation (Webb and Cope 1992). However, we did not observe a significant difference in the ventral excitatory postsynaptic potential. It is possible that other aspects of central function not assessed here, such as motoneuron excitability, contributed to behavioral compensation or that the behavior has a large functional reserve making it robust in the face of early losses of peripheral input.

Two other conditional knockouts also produce late effects on proprioceptors, and interestingly, both affect their peripheral, but not central connections, as observed following Dicer knockout. Targeted deletion of Erbb2 in the muscle produces an observable effect on behavior by around p15, as does selective elimination of NT3 from intrafusal fibers (Shneider et al. 2009). Both mutants affect the differentiation and/or trophic signaling from the intrafusal fibers and, like Dicer KO, produce modest or no effects on the central connections of proprioceptors.

How then does Dicer KO in the proprioceptors lead to deficits in muscle connectivity and transcriptional identity? In mammals, Dicer has several reported roles, such as processing microRNA precursors into mature transcripts, the generation of endogenous siRNAs, and in some instances preventing the accumulation of transcribed retro-transposons (Kurzynska-Kokorniak et al. 2015). We were able to eliminate the latter function, as our RNA sequencing data did not demonstrate an increase in SINE elements or other categories of retrotransposons. This suggests that a microRNA, siRNA, or group of either is responsible for the decline in identity and function. Unfortunately, isolating and sequencing the small RNA population from the proprioceptor population proved difficult and effects at the level of mRNA were widespread, precluding identification of a specific set of microRNAs or siRNAs responsible.

Loss of Dicer produced highly significant, but distributed, changes in the proprioceptor transcriptome. Genes previously identified as critically important to proprioceptor development such as Etv1, Runx3, and TrkC were modestly downregulated (1.4 to 3.4-fold) and on average proprioceptive-enriched genes declined by 40%. This more distributed deficit is consistent with the effects of microRNAs in other systems (Bartel 2009). Although we could not test the effects of each of these genes, one of us has found previously that a parvalbumin-driven knockout of Etv1 produced no observable behavioral deficit (Ladle DR, personal communication).

The decline in cell type-specific expression within proprioceptors is reminiscent of the effects of loss of microRNA function in retinal cone cells (Busskamp et al. 2014), which leads to a decline in cone-enriched genes and a loss of their primary cellular specializations, the outer segment. A similar posttranscriptional regulation of identity and function has also been demonstrated in motor neurons (Amin et al. 2015; Thiebes et al. 2015) in which developmental expression of mir-218 is required to suppress a network of genes expressed in surrounding interneuron cell types. It seems feasible that proprioceptive sensory neurons require active posttranscriptional suppression of transcripts normally enriched in other DRG cell types as this would help explain the upregulation of this gene group (Fig. 3, D and E). Finally, in the work reported here, it is intriguing that the loss of cell identity at the transcriptional level is accompanied by a loss of functional muscle connectivity but an at least initial preservation of the central connections. An appealing explanation is that because all sensory neurons within the ganglion ultimately innervate the spinal cord, this property is more hardwired than some of the sensory specializations these groups acquire. While there is ample evidence for genetic and epigenetic specification of neuronal cell types, the present results highlight the possibility that posttranscriptional regulation may serve to fine- une and further specialize specific properties of neuronal cell types.

GRANTS

This work was supported by National Institute of Neurological Disorders and Stroke Grants P01-NS-079149 (to S. B. Nelson) and NS-072454 (to D. R. Ladle).

DISCLOSURES

No conflicts of interest, financial or otherwise, are declared by the author(s).

AUTHOR CONTRIBUTIONS

S.M.O., D.R.L., and S.B.N. designed experiments; S.M.O., M.M.F., J.M., H.Z., Y.S., and D.R.L. performed experiments; S.M.O., Y.S., and D.R.L. analyzed data; S.M.O., D.R.L., and S.B.N. interpreted results of experiments; S.M.O. and D.R.L. prepared figures; S.M.O., D.R.L., and S.B.N. drafted manuscript; S.M.O., D.R.L., and S.B.N. edited and revised manuscript; S.B.N. approved final version of manuscript.