Termination of cell-type specification gene programs by the miR-183 cluster determines the population sizes of low-threshold mechanosensitive neurons

Sensory neurons are heterogeneous: some neuron types have a low activation threshold and are therefore highly sensitive to stimuli. These register non-painful low-threshold mechanical stimuli and are therefore termed low-threshold mechanosensitive neurons. Myelinated (A-fiber type) low-threshold mechanoreceptors (LTMRs) terminate peripherally in the skin and participate in touch sensation. Interestingly, based on single-cell RNA sequencing, there are three types of myelinated LTMR: Aδ LTMRs, Aβ rapidly adapting (Aβ RA) LTMRs and Aβ slowly adapting (Aβ SA) LTMRs [referred to as NF1, NF2 and NF3 by Usoskin et al. (2015)]. Aδ LTMRs terminate as longitudinal lanceolate endings in hair follicles, and are involved in directional sensitivity to hair deflection and light touch. Aβ RA-LTMRs terminate as three morphologically distinct types of nerve ending: in Meissner corpuscles in glabrous skin, which detect movement across the skin; in Pacinian corpuscles in glabrous skin, which are tuned to high-frequency vibration; and in lanceolate endings in hairy skin, which detect movement and low-frequency vibration (Abraira and Ginty, 2013). Finally, Aβ SA-LTMRs terminate on Merkel cells in hairy and glabrous skin, and are involved in sensitivity to skin indentation; in addition, when terminating as circumferential endings in hair follicles, they mediate sensitivity to gentle skin stroking. These three types of myelinated LTMR can be defined by: expression of NECAB2 and high levels of TRKB (Aδ LTMRs); expression of CALB1, RET and low levels of TRKB (Aβ RA-LTMRs); and finally, expression of TRKC and RET (Aβ SA-LTMRs) (Usoskin et al., 2015).

During development, touch-sensitive neurons are specified from a common progenitor pool by a series of instructive signals and transcription factors (Lallemend and Ernfors, 2012; Liu and Ma, 2011). Early hybrid TRKB⁺/TRKC⁺ neurons diverge around E11-E12 in the mouse and eventually generate two mature TRKB⁺ (NTRK2⁺) neuronal types, TRKB^high Aδ LTMRs and TRKB^low Aβ RA LTMRs, both of which extinguish TRKC (NTRK3) expression, whereas those that maintain TRKC but extinguish TRKB expression become Aβ SA-LTMRs neurons (Bourane et al., 2009; Kramer et al., 2006; Lallemend and Ernfors, 2012). In contrast to the above touch-sensitive neuron types, TRKC⁺ proprioceptive mechanosensitive neurons express RUNX3, which promotes the proprioceptive neuronal fate (Abdo et al., 2011; Kramer et al., 2006; Levanon et al., 2002; Scott et al., 2011).

MicroRNAs (miRNAs) can play a crucial role during development. For example, one of the first miRNAs identified, Lin-4, regulates developmental timing across different tissues in the nematode C. elegans, and loss-of-function mutations reiterate early developmental programs at inappropriate late larval stages (Chalfie et al., 1981; Lee et al., 1993; Wightman et al., 1993). For development of sensory systems, the miR-183-96-182 (the miR-183 cluster) is particularly interesting because it is expressed in many sensory organs, such as retina, inner ear, dorsal root ganglion (DRG), olfactory epithelia and tongue epithelia in mouse (Bak et al., 2008; Lagos-Quintana et al., 2003; Lumayag et al., 2013; Sacheli et al., 2009). A high expression level of the miR-183 cluster in retina is necessary for the development of its outer segments and functional maintenance in the adult (Busskamp et al., 2014a,b; Lumayag et al., 2013; Zhu et al., 2011). The miR-183 cluster also regulates the development of sensory inner ear hair cells, and point mutations on miR-96 affects hair cell function or hair cell development [causing progressive hearing loss in both mouse and human (Gu et al., 2013; Lewis et al., 2009; Mencía et al., 2009; Soldà et al., 2012)]. In this work, we conditionally delete the miR-183 cluster in neural crest derivatives, including the DRG. We report that the miR-183 cluster is expressed in DRG sensory neuron progenitors where it acts as a timer to extinguish Shox2 through direct targeting of its mRNA, which aborts the gene programs for Aδ NECAB2⁺TRKB^high LTMR neurons and re-directs cell specification towards the generation of Aβ TRKC⁺RUNX3⁻ SA-LTMR neurons.

All three miR-183 cluster members (miR-183, miR-96 and miR-182) are specifically expressed in mouse DRG already at E10.5 and are continuously expressed through development into adulthood (Peng et al., 2017). To directly examine whether the miR-183 cluster plays a role in somatosensory neuron development, we crossed Wnt1-Cre mice (Danielian et al., 1998) with miR-183-96-182^flox/flox mice (Peng et al., 2017) to generate Wnt1-Cre; miR-183-96-182^flox/flox conditional knockout mice lacking the miR-183 cluster in all DRG neurons (referred to as miR^CKO mice). In situ hybridization confirmed that expression of the miR-183 cluster was depleted in DRG from E11.5 onwards in the miR^CKO mice (Fig. S1A). Immunohistochemical analysis of newly born (P0) mice revealed that the proportion of TRKB⁺ neurons was increased in the miR^CKO mice when compared with Wnt1-Cre mice (Fig. 1A-C). Importantly, the total number of neurons in L5 DRG was unchanged (Fig. 1D). In contrast, the percentage of TRKC⁺ neurons was decreased in the miR^CKO mice. This reduction was caused by a decrease in development of TRKC⁺/RUNX3⁻ sensory neurons (Aβ SA-LTMRs), whereas TRKC⁺/RUNX3⁺ proprioceptive neurons appear to be unchanged (Fig. 1E-G). Loss of the miR-183 cluster had no effect at P0 on the proportions of TRKA (nociceptors), RET (some nociceptors and some LTMRs), neurofilament heavy polypeptide (NFH, all LTMRs), RET⁺/NFH⁺ (RET⁺ LTMRs) and RET⁻/NFH⁺ (all LTMRs except RET⁺) in the miR^CKO mice when compared with Wnt1-Cre mice (Fig. S1B,C). These data suggest that loss of the miR-183 cluster leads to fate-switch specifically of TRKC⁺ to TRKB⁺ neurons in the DRG without any changes in absolute numbers of LTMRs.

To identify the mechanism by which the miR-183 cluster works to cause change of the fate of early DRG progenitor cells, we profiled RNA expression by RNA sequencing in three biological replicates of E12.5 lumbar DRG from the miR^CKO and control mice (Table S1, sheet 1). A total of 777 genes were significantly (P<0.05) upregulated in the miR^CKO DRG when analyzing the RNA sequencing data using Qlucore (Fig. 2A; Table S1, sheet 2). Among these, 101 genes were putative direct targets carrying binding site(s) for at least two members of the miR-183 cluster predicted by TargetScan algorithm (www.targetscan.org; Table S1, sheet 3), and 38 out of the 101 genes were increased with more than 1.3-fold in the absence of the miR-183 cluster (Fig. 2B; Table S1, sheet 4). Only two of these were transcription factors (Shox2 and Zbtb41); however, Shox2 was among the top 10 highly expressed genes with more than 1.3-fold upregulation (Table S1, sheet 4). This caught our attention, because Shox2 has been reported to be important for proper specification of TRKB⁺ neurons in the developing DRG (Abdo et al., 2011; Scott et al., 2011). Furthermore, two highly conserved binding sites for miR-183 and miR-96 in the Shox2 3′UTR were predicted by TargetScan algorithm among many vertebrates (Fig. S2). Real-time PCR independently confirmed that Shox2 mRNA was increased 2.3-fold in the miR^CKO compared with control mice at E12.5 (Fig. 2C). SHOX2 protein was highly expressed in most DRG neurons at E10.5, rapidly reduced at E11.5 and barely detected by E12.5 (Fig. 2D,E). TRKB⁺ and TRKC⁺ neurons segregate between E11.5 and E12.5 in the lumbar DRG (Kramer et al., 2006); therefore, SHOX2 repression precedes the diversification of these neuron types. SHOX2 extinction between E10.5 and E11.5 failed in miR^CKO mice; however, by E12.5 SHOX2 was downregulated similar to control mice (Fig. 2D,E). The finding that miR-182 was expressed in all DRG neurons throughout this developmental time window (Fig. 2F) indicates that this microRNA cluster works through determining the timing of extinction of SHOX2 rather than establishing selective expression in some but not other cells.

Our results show that the miR-183 cluster rapidly extinguishes SHOX2 expression at E11.5 in most neurons of the DRG. SHOX2 drives TRKB expression, and in the Shox2 knockout mice there is a fate switch from TRKB⁺ neurons to TRKC⁺ LTMR neurons during this stage of development (Abdo et al., 2011; Scott et al., 2011). This raises the possibility that the miR-183 cluster could determine the proportion of TRKB⁺ and TRKC⁺ neurons generated during development by repression of Shox2 expression. To obtain further insights into the development of LTMRs, we examined DRG at E10.5, prior to SHOX2 repression in most neurons. At this stage, TRKB and TRKC are largely co-expressed in DRG progenitors, which a few days later diversify into TRKB⁺ LTMRs and TRKC⁺ neurons, as previously reported (Kramer et al., 2006). Triple immunostaining for SHOX2, TRKC and TRKB showed that, at E10.5 among TRKC⁺ neurons, virtually all also contained SHOX2 and, among these, 24±3% also co-stained for TRKB. Furthermore, among TRKB⁺ neurons, all were also TRKC⁺ and SHOX2⁺ (Fig. 3A). Thus, SHOX2 is expressed promiscuously both in all neurons expressing only TRKC and in all neurons expressing TRKB, and these always co-express TRKC at E10.5. This indicates that expression of TRKB is initiated in some of the TRKC⁺/SHOX2⁺ neurons. In the E11.5 control Wnt1-Cre mouse, TRKB and TRKC remained largely co-expressed, although some neurons started to lose membrane presence of TRKC, which is indicative of its downregulation [consistent with previous results (Kramer et al., 2006)], and SHOX2 was expressed in the majority of the TRKB⁺ neurons (Fig. 3B).

We next analyzed control mice (Wnt1-Cre) and miR^CKO mice to determine how SHOX2 is expressed between E11.5 and E12.5 when TRKB⁺/TRKC⁺ neurons segregate into TRKB⁺ and TRKC⁺ neurons (Kramer et al., 2006), which defines the generation of different kinds of LTMRs (Abdo et al., 2011; Bourane et al., 2009; Scott et al., 2011). Immunohistochemical staining indicated the presence of more SHOX2⁺ neurons and increased numbers of both TRKB⁺/SHOX2⁺ and TRKC⁺/SHOX2⁺ neurons at E11.5 (Fig. 3C). Quantification revealed in control (Wnt1-Cre) mice that SHOX2 expression among TRKB⁺ neurons was reduced from 84±2% of all TRKB⁺ neurons at E11.5 to 20±3% at E12.5. In contrast to maintained SHOX2 expression in some TRKB⁺ neurons at E12.5, it was repressed in nearly all TRKC⁺ neurons (Fig. 3D,F). Thus, SHOX2 segregates in TRKB- but not TRKC-expressing neurons. In miR^CKO mice, an increased percentage of TRKB⁺ neurons contained SHOX2 when compared with control mice at E11.5, but similar to control mice, only about a quarter of the TRKB⁺ neurons contained SHOX2 at E12.5 (Fig. 3D). Because SHOX2 drives differentiation of the TRKB⁺ LTMR fate, the shifted temporal repression of SHOX2 predicts that the miR^CKO mice display an overall increased number of TRKB⁺ neurons at E11.5. Quantification of the E11.5 DRG revealed that the miR^CKO mice had more TRKB⁺/SHOX2⁺ double-positive neurons than control (Wnt1-Cre) mice, whereas there was no change in the number of TRKB⁺/SHOX2⁻ neurons (Fig. 3E). This finding is in line with the failure of Shox2 repression to maintain TRKB expression in neurons that normally should have extinguished TRKB (e.g. TRKC⁺ neurons). Consistently, the percentage of TRKC⁺ neurons containing SHOX2⁺ was increased in the miR^CKO mice when compared with control Wnt1-Cre mice, with more than half of all TRKC⁺ neurons expressing SHOX2 at E11.5 (Fig. 3F) while the total number of TRKC⁺ neurons was unchanged (Fig. 3G). This shows that when the miR-183 cluster is absent, SHOX2 is maintained in TRKC⁺ neurons that normally should have lost SHOX2 by E11.5. We therefore conclude that the miR-183 cluster is crucial for a timely extinction of SHOX2 expression at E11.5 when LTMR neurons diversify into distinct types.

Since the mature sequences of miR-183, miR-96 and miR-182 and their binding sites on Shox2 3′UTR are highly conserved among vertebrates, from lizard to human (Fig. S2 and data not shown), we next investigated whether the miR-183 cluster, TRKB, TRKC and SHOX2 are expressed in a similar way in the developing human as in rodents. Immunostaining on DRG sections from human embryos at post-conception times 6, 7, 8.5 and 11 weeks showed that TRKB, TRKC and SHOX2 are expressed in many neurons at 6 weeks with reduced number of neurons as development progresses (Fig. 4A-O). TRKB and TRKC were sometimes co-localized in 6- and 7-week-old embryos (asterisk), whereas in 8.5-week-old embryos TRKB and TRKC neurons were largely segregated. This segregation was accompanied by a decline of SHOX2 in TRKC⁺ neurons from 7-8.5 weeks (Fig. 4F-O, arrows) and at 8.5 weeks, most of TRKC neurons had downregulated SHOX2 (Fig. 4K-O). In contrast, TRKB was also retained in SHOX2⁺ neurons at later embryonic stages (Fig. 4K-O, arrowheads). In 11-week-old embryos, SHOX2 was mostly expressed in TRKB⁺ neurons (Fig. 4P); however, a few TRKB⁺ cells were SHOX2⁻ (inset in Fig. 4P). Expression of Shox2 mRNA was also confirmed by reverse transcription PCR (RT-PCR) (Fig. 4Q). Together, these data show that expression of TRKB, TRKC and SHOX2 are similarly regulated in human and in mouse. In situ hybridization for the miR-183 cluster suggested that all three miRNAs were expressed in human embryonic DRG from 6-8.5 weeks (Fig. S3). Quantitative RT-PCR confirmed expression of the miR-183 cluster, with miR-182 expressed at the highest level (Fig. 4Q). miR-96, which has a higher affinity for the human Shox2 3′UTR among the three miRNAs (see Fig. S2), showed the greatest regulation. miR-96 was first downregulated to about 50% from week 5.5 to week 6, coinciding with elevated levels of Shox2. miR-96 was thereafter substantially increased at week 7, to levels slightly higher than at week 5.5 (Fig. 4Q). Our results demonstrate the similar development of TRKB⁺ and TRKC⁺ neurons and miR-183 cluster expression in the embryonic human as in mouse DRG.

Our previous results show that miR-183 cluster participates in the diversification of TRKC⁺/RUNX3⁻ LTMR but not TRKC⁺/RUNX3⁺ proprioceptors (Fig. 1G). Excluding proprioceptors, the only LTMRs expressing TRKC in the adult are Aβ SA-LTMRs (Usoskin et al., 2015). We therefore concluded that miR-183 cluster-dependent repression of TRKB participates in generating the TRKC⁺ Aβ SA-LTMRs. However, it remained unclear whether the increase of TRKB⁺ neurons occurred in Aδ LTMRs and/or Aβ RA-LTMRs, because both types express TRKB (Usoskin et al., 2015). To examine whether loss of the miR-183 cluster results in persistent changes of the proportion of the different LTMRs and, if so, which TRKB⁺ LTMRs type(s) are involved, we collected and analyzed adult DRG from control (Wnt1-Cre) and miR^CKO mice. Quantifying the total number of neurons revealed no differences between the genotypes (Fig. 5A) and, consistent with analyses at P0, the percentage of all neurons expressing TRKC⁺ was reduced (Fig. 5B). Aβ RA-LTMRs and Aδ LTMRs are molecularly different and can be distinguished by expression of the calcium-binding proteins CALB1 (calbindin) in the former and NECAB2 in the latter. Quantification revealed no changes in the percentage of all neurons expressing TRKB⁺/CALB1⁺, whereas the percentage of TRKB⁺/NECAB2⁺ neurons was significantly increased (Fig. 5C-E). Thus, this suggests that miR-183 cluster is involved in diversification of a TRKB⁺/TRKC⁺ progenitor that segregates into the TrkB^high/NECAB2 Aδ LTMRs and TRKC⁺/RUNX3⁻ Aβ SA-LTMRs.

To determine whether Shox2 is a direct target of miR-183 cluster, we cloned the mouse Shox2 3′ untranslated region (UTR), which contains the two conserved binding sites for the miR-183 cluster, into a luciferase reporter vector (sensor) and co-transfected this vector together with a miR-183-96 overexpression vector into HEK-293 cells (Fig. S4). Overexpression of miR-183-96 repressed luciferase expression from the Shox2 3′UTR sensor vector by 58% and this repression was abolished when the conserved seed sequences of miR-183-96 binding sites in the sensor vectors were mutated (Fig. 6A,B). As there is only one nucleotide difference between the seed sequence of miR-182 and that of miR-96, we examined whether miR-182 can also target Shox2 3′UTR. Co-transfection of a miR-182 overexpression vector repressed the luciferase expression from the Shox2 3′UTR sensor vector by 43%, and this repression could be fully rescued by mutating the miR-96-binding site on the sensor vectors (Fig. 6A,B). These results show that miR-183 cluster regulates the expression levels of Shox2 by directly binding to the conserved binding sites located within the 3′UTR. To confirm that the miR-183 cluster can repress Shox2 expression in vivo, we forced expression of miR-183-96 in chicken DRG by electroporating a pCAG-miR-183-96-IRES-GFP vector into the neural tube at the HH13 stage. Embryos were allowed to continue to develop and were analyzed at E7.5, when the neural crest cells have migrated, undergone neurogenesis and coalesced to form the DRG. Quantification showed that the percentage of GFP⁺/SHOX2⁺ cells over total GFP cells was dramatically reduced in the miR-183-96-overexpressing group when compared with the control pCAG-IRES-GFP-electroporated group (Fig. 6C). These results suggest that the miR-183 cluster can target the Shox2 mRNA.

Based on the strong correlation between upregulation of SHOX2 and the changed proportions of subgroups of DRG neurons in the miR^CKO mice, we next investigated whether SHOX2 represents the key target of the miR-183 cluster responsible for the phenotype in the miR^CKO mice by using gain-of-function and loss-of-function strategies. Thus, if the miR-183 cluster works through repression of Shox2, overexpression of SHOX2 is expected to phenocopy the miR-183^CKO mice and genetic ablation of Shox2 is expected to lead to the reverse phenotype. Overexpression of Shox2 resulted in more than a twofold increase of TRKB⁺ DRG neurons in chicken embryos (Fig. 6D,E). Crossing Wnt1-Cre to Shox2^flox/flox mice (Abdo et al., 2011) to generate Wnt1-Cre;Shox2^flox/flox mice that have a loss of function of Shox2 in the DRG revealed a marked reduction of TRKB⁺ DRG neurons and increased number of TRKC⁺ DRG neurons at P0, as previously reported (Abdo et al., 2011; Scott et al., 2011) (Fig. 6F-H). Moreover, we found that the TRKC⁺/RUNX3⁻ but not TRKC⁺/RUNX3⁺ neuron types were increased in the Wnt1-Cre;Shox2^flox/flox mice when compared with the Shox2^flox/flox control mice (Fig. 6G,H). Altogether (see Fig. 6I), our results suggest that the miR-183 cluster works through regulation of Shox2 in a gene-regulatory network (Fig. 6J), which determines the population sizes of Aδ LTMRs and Aβ SA-LTMRs generated during development.

This study represents, to our knowledge, the first molecular identification and characterization of a miRNA that determines the timing of expression of fate-inducing transcription factors in primary sensory neurons. We show that the miR-183 cluster terminates genesis of Aδ LTMRs neurons in favor of Aβ SA-LTMR neurons during development through extinction of Shox2 expression.

Expression of neurotrophic factor receptors confers ligand sensitivity. Neurotrophic factor signaling plays crucial roles during development for neuron survival, axon growth, peripheral target innervation, patterns of terminations in the spinal cord as well as differentiation into specialized and modality-specific sensors (Lallemend and Ernfors, 2012; Marmigère and Ernfors, 2007). Consequently, expression of growth factor receptors is one of the first distinguished features during cell type diversification in the DRG. However, distinct transcription factor programs are involved in this process, and often interact with growth factor receptor signaling. RUNX3 is crucial for the specification of TRKC-expressing proprioceptive sensory neurons (Levanon et al., 2002) where both BRN3A and RUNX3 are important for repression of TRKB to generate RUNX3⁺/TRKC⁺ proprioceptive neurons (Dykes et al., 2010; Kramer et al., 2006). Whereas RUNX3 suppresses Shox2 expression, SHOX2 is unable to suppress Runx3 (Abdo et al., 2011). Therefore, initiation of Runx3 seems to be contributing both by suppressing alternative fates and by initiating the proprioceptive neuron fate. Because of this, neither Shox2 nor the miR-183 cluster are expected to affect proprioceptive neuron development. This agrees with our results, revealing no effect on TRKC⁺/RUNX3⁺ neurons in both the miR^CKO mice and the Shox2 CKO mice at P0. Consequently, these findings predict that miR-183 specifically affects development of LTMR neurons involved in touch.

Some neurons acquire RET expression already at E10.5 in the mouse (Molliver et al., 1997), thus coinciding with segregation of other types of LTMRs (e.g. TRKB⁺ and TRKC⁺/RUNX3⁻ neurons). These neurons depend on RET signaling for the specification of RA-LTMRs (Bourane et al., 2009; Honma et al., 2010; Lecoin et al., 2010; Luo et al., 2009). We found no change in total RET⁺ neurons nor in myelinated NFH⁺/RET⁺ neurons at P0. Therefore, our findings predict that miR-183 primarily affects LTMRs rather than early RET⁺ neurons, which differentiate into Aβ RA-LTMRs. Thus, it seems that RET⁺ RA-LTMRs and TRKC⁺/RUNX3⁺ proprioceptors rely on mechanisms that are independent of the miR-183 cluster.

Consistent with distinct paths for diversification, TRKB⁺ Aδ LTMRs and TRKC⁺ Aβ SA-LTMRs seem to emerge from hybrid TRKB⁺/TRKC⁺ neurons through extinction of TRKB (Kramer et al., 2006). SHOX2 has been shown to be necessary for proper diversification of TRKB⁺ neurons, and in its absence, TRKB⁺ neurons decrease by 60%, whereas TRKC⁺ neurons increase (Abdo et al., 2011; Scott et al., 2011). In this study, we find that this increase of TRKC⁺ neurons occurs among TRKC⁺/RUNX3⁻ neurons differentiating into LTMRs. SHOX2 therefore participates specifically in the split between neurons that differentiate into Aδ LTMRs and Aβ SA-LTMRs. In this process, the miR-183 cluster determines the timing of Shox2 extinction, rather than the selective expression in particular cells between E10.5 and E11.5.

Given the fact that all subtypes of sensory neurons in the DRG come from the same progenitor pool derived from trunk neural crest cells, the inversely proportional changes between two related neuron types indicate a shared genetic pathway fating one type and repressing the other. This conclusion is generalized, as numbers do not always add up perfectly. For example, at P0, the percentage increase of TRKB⁺ neurons is less than the increase of TRKC⁺ neurons. Thus, this suggests cell fate changes or loss of cells not recorded in this study. Nevertheless, Shox2 is a direct target for the miR-183 cluster, and the de-repression of Shox2 in miR-183^CKO mice appears to be the main mechanism causing the phenotype. Consistent with this, in the Shox2 CKO mice, the number of TRKB⁺ neurons decreases and the number of TRKC⁺ neurons increases, whereas overexpression of Shox2 increases TRKB⁺ neurons; however, in the miR-183^CKO mice, the number of TRKB⁺ neurons increases and the number of TRKC⁺ neurons decreases. Furthermore, the major phenotype of miR-183^CKO mice and Shox2 CKO mice encompasses these neuron types, which differentiate into Aδ LTMRs and Aβ LTMRs.

Pre-mi-RNAs are processed to become mature miRNAs by the ribonuclease Dicer (Bernstein et al., 2001; Zhang et al., 2004). Although system-wide deletion of Dicer leads to early embryonic lethality (Bernstein et al., 2003), conditional deletion restricted to the peripheral nervous system reveals its indispensable role for cell survival, development and plasticity in the nervous system (Fiorenza and Barco, 2016). In the DRG, loss of Dicer results in abnormal development with marked loss of sensory neurons and a failure to produce axonal projections (Zehir et al., 2010). The miR-183 cluster is contained within an −4 kb genomic sequence, is produced as a polycistronic pri-miR transcript (Dambal et al., 2015) and has related seed sequences (Karali et al., 2007; Ryan et al., 2006; Xu et al., 2007). miR-183 is downregulated in animal models for neuropathic pain and knee joint osteoarthritis (Li et al., 2013; Lin et al., 2014), and, consistently, forced miR-183 expression by intrathecal injections of lentiviruses attenuate spinal nerve ligation-induced mechanical allodynia (Lin et al., 2014). Furthermore, loss of miR-183 cluster function in the adult results in increased basal mechanical sensitivity and mechanical allodynia (Peng et al., 2017). Interestingly, the increased allodynia during nerve damage involves TRKB⁺ Aδ LTMRs (Peng et al., 2017). However, this effect of miR-183 cluster-deficiency is caused by increased sensitization/excitability due to the loss of a continuous suppression of auxiliary voltage-gated calcium channel subunit genes (Cacna2d1 and Cacna2d2) in adult mice. Thus, although the miR-183 cluster targets Shox2 during development that affects cell-type specification, Cacna2d1 and Cacna2d2 are targeted in adult, which affects mechanical sensitivity. This miRNA cluster therefore serves different functions in sensory neurons during development when compared with the adult.

Given that the miR-183 cluster and many of its target genes are highly conserved in mammals, it is believed that the miR-183 cluster could play similar roles in human as in mouse. This prediction is enforced by the discovery that a point mutation in miR-96 causes progressive hearing loss in both mouse and human (Lewis et al., 2009; Mencía et al., 2009; Soldà et al., 2012). Our data show that the miR-183 cluster is also highly expressed in the developing DRG of humans along with the direct target gene Shox2. These data indicate that the miR-183 cluster play similar roles in the developing human DRG as it does in the mouse.

miR-183-96-182^flox/+ mice (Peng et al., 2017) were crossed with Wnt1-Cre 129/SvEv mice (Danielian et al., 1998) to obtain Wnt1-Cre; miR-183-96-182^flox/+ mice. Wnt1-Cre; miR-183-96-182^flox/+ mice were mated to miR-183-96-182^flox/+ mice to obtain Wnt1-Cre; miR-183-96-182^flox/flox mice and Wnt1-Cre control mice. Wnt1-Cre; Shox2^flox/flox mice were obtained by crossing Shox2^flox⁄flox mice (Cobb et al., 2006) with Wnt1-Cre mice (Danielian et al., 1998), and the genotyping of the Shox2 floxed allele and Wnt1-Cre was performed as previously described (Abdo et al., 2011). Timed pregnant females were used for collection of embryos; noon of the day of vaginal plug detection was designated as E0.5. All animal work was conducted under ethical permission from the Swedish ethical review panel (norra djurförsöksetiska nämnden).

Human embryonic and fetal tissue was retrieved from elective routine abortions at the Karolinska University Hospital with written consent from the pregnant women, and DRGs were dissected and immersed in 4% PFA for 6 h, further immersed in 20% sucrose overnight after wash in PBS for 5 min, finally embedded in OCT and sectioned at 14 μm. Age (weeks after conception) of the aborted tissue was determined using anatomical landmarks (England, 1990; Yamada and Takakuwa, 2012). Use of human fetal tissue was approved by the Stockholm vetting board on ethics in human research (2007/1477-31, 2011/1101-32 and 2013/564-32).

Paraffin sections (8 μm) and cryosections (14 μm) from lumbar mouse and human DRG were processed and hybridizations were performed as described previously (Peng et al., 2012). Digoxigenin (DIG)-labeled LNA probes for miR-183 (Exiqon) and miR-182 (Exiqon) were used. Alkaline phosphatase-conjugated anti-digoxigenin antibody (Roche, 1:2000) was used and alkaline phosphatase staining was developed with Fast Red (Roche) and then followed by counterstaining with DAPI or immunostaining for ISL1 [clone 39.4D5; Developmental Studies Hybridoma Bank (DSHB), The University of Iowa, Iowa City, USA, 1:100]. Images were taken using an Olympus FV1000 confocal microscope.

Immunostaining on 14 μm cryosections from mouse and human lumbar DRG was performed as previously described (Abdo et al., 2011). The following primary antibodies were used: mouse antibodies against ISL1 (DSHB, 1:100), SHOX2 (Santa Cruz, 1:400), NFH (neurofilament, heavy polypeptide) (CloneN52; Sigma Aldrich, 1:500); rabbit antibodies against RUNX3 (gifts from Thomas Jessell, Columbia University Medical Center, NY, USA; 1:300), ISL1 (gifts from Thomas Jessell, 1:250), NFH (Millipore, 1:200), TRKA (Millipore, 1:500), TRKC (Cell Signaling, 1:500), NECAB2 (Proteintech Europe, 1:1000), CALB1 (Millipore, 1:500) and chicken TRKB (a kind gift from Louis F. Reichardt, University of California, USA; 1:2000); goat antibodies against TRKA (R&D systems, 1:500), TRKB (R&D systems, 1:500), TRKC (R&D systems, 1:500), Ret (R&D systems, 1:100) and GFP (Abcam, 1:500); guinea pig antibodies against TLX3 (a kind gift from Carmen Birchmeier, Max Delbruck Center for Molecular Medicine, Germany, 1:10 000). Secondary antibodies were fluorescently labeled (AlexaFluor 405/488/594/647; Molecular Probes and Invitrogen). Fluorescent images were taken with a confocal laser scanning microscope (Olympus FV1000 confocal microscope or Zeiss LSM700) and processed with Adobe Photoshop CS software.

All markers were counted on DRG sections at either lumbar level 4-6 (Wnt1-Cre control and miR^CKO mice) or brachial level (Wnt1-Cre; Shox2^flox/flox and Shox2^flox/flox mice). TRKA⁺, TRKB⁺, TRKC⁺, ISL1⁺ and SHOX2⁺ cells were counted on eight position-matched sections from each Wnt1-Cre control and miR^CKO embryo. To analyze the proportion of each marker at the P0 stage, the numbers of TRKA⁺, TRKB⁺, TRKC⁺, RUNX3⁺, RET⁺ and NFH⁺ cells were counted on at least the three biggest sections from three DRG per Wnt1-Cre control, miR^CKO, Wnt1-Cre; Shox2^flox/flox and Shox2^flox/flox mouse, then normalized to the number of total neurons (ISL1⁺ or TLX3⁺) on the same sections. To get the numbers of total neurons in L5 DRG at P0 (ISL1⁺) and adult (TLX3⁺), every ninth serial section through L5 ganglion was counted, then multiplied by eight to get the number of total neurons in L5 DRG. TRKB⁺, CALB1⁺, NECAB2⁺, TRKC⁺, TRKA⁺ and NFH⁺ cells were counted on every ninth serial section through L5 ganglion of Wnt1-Cre control and the miR^CKO adult mice (5-6 months), and the number cells positive for each marker was then normalized to the number of TLX3⁺ neurons for comparison. All values shown are mean±s.e.m. Statistical significance between groups was assessed by t-tests using GraphPad Prism 5. A value of P<0.05 was considered significant.

Mouse lumbar DRG tissue was dissected from E12.5 three controls (two Wnt1-Cre and one wild type) and three miR^CKO embryos. Total RNA was isolated using Trizol Reagent (Invitrogen) and processed with TruSeq Stranded mRNA Sample Prep Kit (Illumina/USA) according to the manual to obtain the cDNA library. The adapter-ligated libraries were sequenced on the Illumina sequencer according to the manufacturer's instructions. Read processing was performed as described previously (Islam et al., 2014), except that the molecule counting by unique molecule identifiers was omitted. Alignments against UCSC mm10 genome were made with bowtie1 version 0.12.9 (Langmead et al., 2009), allowing for up to three mismatches. RPKM values were obtained by dividing each read count by the transcript length in kb and normalizing to a total of 1 million in each sample. Profiling data from mouse DRG (Table S1, sheet 1) were then analyzed using Qlucore Omics Explore, and significantly (P<0.05) upregulated genes (Table S1, sheet 2) in the miR^CKO DRG were matched to the targets (Table S1, sheet 3) predicted by TargetScan algorithm. The 38 overlapped genes with upregulation of more than 1.3-fold are listed in Table S1, sheet 4.

The genomic sequences (shown in Table S1, sheet 5) for miR-183-96 and miR-182 were synthesized and cloned into pCAGIG vector (Addgene, 11159). Levels of miR-183, miR-96 and miR-182 expression from the OE vectors pCAG-miR-183-96-IRES-GFP and pCAG-miR-182-IRES-GFP were quantified by real-time RT-PCR after transfection into HEK-293 cells (Fig. S4). The mouse Shox2 3′UTR (Entrez Gene Accession number NM_001302357.1) was amplified from cDNA of E12.5 DRG with the primers indicated in Table S1, sheet 6, and cloned into the XbaI site downstream of the Luciferase CDS in the pGL3 promoter vector (Promega) to obtain the pGL3-Shox2-3′UTR vector. Binding site-mutated Shox2 3′UTR sequences were directly synthesized and cloned into pGL3 promoter vectors. Truncated mouse Shox2 CDS (position 210-1350, without binding sites for the miR-183 cluster, Entrez Gene Accession number NM_013665.1) was subcloned from pGEMT-Shox2 cDNA plasmid (a gift from J. Cobb, University of Calgary, Canada) into the pCAGIG vector to obtain the pCAG-Shox2-IRES-GFP OE vector.

The miRNA sensor assays were conducted in HEK293 cells co-transfected with 500 ng/well pGL3-Shox2-3′UTR sensor vector and 10 ng/well pRL-SV40 (internal transfection control) along with 1 µg/well pCAG-miR-183-96-IRES-GFP vector or 1 µg/well pCAG-miR-182-IRES-GFP vector using Lipofectamine 2000 (Invitrogen). The corresponding rescue assays were carried out in HEK293 cells co-transfected with 500 ng/well pGL3-mut-Shox2-3′UTR sensor vector with 1 µg/well pCAG-miR-183-96-IRES-GFP vector or 1 µg/well pCAG-miR-182-IRES-GFP and 10 ng/well pRL-SV40 vector. Cells were lysed in passive lysis buffer 40 h post-transfection, and Firefly and Renilla Luciferase luminescence was measured in a Victor luminometer (Wallac Sverige) using the Dual-Luciferase Reporter Assay system (Promega) according to the manufacturers' instructions. Firefly luminescence was normalized against Renilla luminescence for each well. Assays were performed in triplicate and data are derived from three independent experiments.

Total RNA was isolated from mouse DRG, human DRG and HEK-293 cells using Trizol Reagent (Invitrogen) and treated with RNase-free DNase I (Qiagen). Total RNA (0.5-1 µg) was polyadenylated and reverse transcribed using Catch-All miRNA&mRNA RT-PCR Kit (Pengekiphen) according to the manufacturer's instructions. For detection of miRNA expression levels, qRT-PCR assays were conducted using the Catch-All miRNA&mRNA universal PCR primer as a reverse primer and the specific miRNA forward primers listed in Table S1, sheet 6. The amplification conditions were an initial step at 95°C for 10 min, followed by 40 cycles of 15 s at 95°C and 1 min at 60°C. All assays were performed in triplicate and included negative controls. The Ct value was recorded for each reaction, and the expression level of miRNA was calculated relative to U6B, a ubiquitously expressed snRNA, and Shox2 was normalized to either Gapdh or 18S RNA. Data are presented as target gene expression=2^−Δct.

pCAG-Shox2-IRES-GFP, pCAG-miR-183-96-IRES-GFP or backbone vector pCAGIG were injected into the neural tube of stage HH13 chicken embryos. Electroporation by five pulses of 40 V/cm was performed using a square wave electroporator (BTX). Embryos were harvested at E7.5 and fixed in 4% PFA/PBS for 6 h at 4°C and sectioned at 14 µm.

We thank Mr Thomas Tingström for great help with taking care of all the mice. We also thank Sten Linnarsson's group for providing assistance in RNA sequencing.