The miRNA biogenesis pathway prevents inappropriate expression of injury response genes in developing and adult Schwann cells

Proper function of the nervous system depends on myelination. In peripheral nerves, Schwann cells (SCs) myelinate axons and the miRNA biogenesis pathway is required for developmental myelination and myelin maintenance. However, regulatory roles of this pathway at different stages of myelination are only partially understood. We addressed the requirement of the core miRNA biogenesis pathway components Dgcr8, Drosha, and Dicer in developing and adult SCs using mouse mutants with a comparative genetics and transcriptomics approach. We found that the microprocessor components Dgcr8 and Drosha are crucial for axonal radial sorting and to establish correct SC numbers upon myelination. Transcriptome analyses revealed a requirement of the microprocessor to prevent aberrantly increased expression of injury‐response genes. Those genes are predicted targets of abundant miRNAs in sciatic nerves (SNs) during developmental myelination. In agreement, Dgcr8 and Dicer are required for proper maintenance of the myelinated SC state, where abundant miRNAs in adult SNs are predicted to target injury‐response genes. We conclude that the miRNA biogenesis pathway in SCs is crucial for preventing inappropriate activity of injury‐response genes in developing and adult SCs.

Proper function of the nervous system depends on myelination. In peripheral nerves, Schwann cells (SCs) myelinate axons and the miRNA biogenesis pathway is required for developmental myelination and myelin maintenance. However, regulatory roles of this pathway at different stages of myelination are only partially understood. We addressed the requirement of the core miRNA biogenesis pathway components Dgcr8, Drosha, and Dicer in developing and adult SCs using mouse mutants with a comparative genetics and transcriptomics approach. We found that the microprocessor components Dgcr8 and Drosha are crucial for axonal radial sorting and to establish correct SC numbers upon myelination. Transcriptome analyses revealed a requirement of the microprocessor to prevent aberrantly increased expression of injury-response genes.
Those genes are predicted targets of abundant miRNAs in sciatic nerves (SNs) during developmental myelination. In agreement, Dgcr8 and Dicer are required for proper maintenance of the myelinated SC state, where abundant miRNAs in adult SNs are predicted to target injuryresponse genes. We conclude that the miRNA biogenesis pathway in SCs is crucial for preventing inappropriate activity of injury-response genes in developing and adult SCs.

| INTRODUCTION
A system is defined as robust when it maintains its function in the presence of internal or external perturbations (Kitano, 2004). Posttranscriptional regulation by microRNAs (miRNAs) is an important mean to confer such robustness to developmental transitions and cellular identities (Ebert & Sharp, 2012). miRNAs are a prominent class of small noncoding RNAs that mainly serve to silence their target mRNAs. Biogenesis of most miRNAs requires the sequential processing of the primary miRNA transcript by RNAse III-like enzymes Drosha, which together with Dgcr8 constitutes the microprocessor, and Dicer to yield a~22 nucleotide duplex. One strand of the mature miRNA duplex is loaded into Argonaute (Ago) proteins and serves as a guide by imperfect base pairing with target mRNAs, allowing Ago-dependent recruitment of additional factors and resulting in translational repression, mRNA deadenylation, and mRNA decay. Besides this canonical miRNA biogenesis, some miRNAs (referred to as noncanonical miRNAs) bypass the requirement of the microprocessor or Dicer by being substrates of alternative processing factors such as the spliceosome or the cleaving activity of Ago2 (Ha & Kim, 2014).
Myelination is a crucial feature of the nervous system, fostering refined propagation of electrical signals and providing support to axons. In the peripheral nervous system (PNS), myelination is accomplished by Schwann cells (SCs). These specialized glial cells are derived from the neural crest as the product of differentiating SC precursors (Jessen & Mirsky, 2005;Pereira, Lebrun-Julien, & Suter, 2012). Initially, immature SCs surround large bundles of axons in groups, while gradually expanding their population as they engage and sort out single axonal segments from the bundles, a process termed radial sorting (Feltri, Poitelon, & Previtali, 2016). After radial sorting is accomplished, myelination is started by SCs that exit the cell cycle and activate a gene program associated with the myelinated SC fate. The resulting myelinating SC state needs to be actively maintained to counteract demyelination-inducing gene programs and to balance myelin turnover (Decker et al., 2006). In response to peripheral nerve injury, however, myelinating SCs have the remarkable capability to be reprogrammed to repair SCs (also known as Bungner cells) that promote regeneration of injured nerves (Jessen, Mirsky, & Lloyd, 2015).
In this study, we aimed to complement the existing knowledge and to gain further insights into the mechanistic functions of the miRNA biogenesis pathway in developmental myelination and myelin maintenance. To achieve this goal, we used a comparative genetic approach to quantitatively dissect the relative requirements and contributions of the SC-expressed microprocessor-components Dgcr8 and Drosha, as well as Dicer in regulating the myelinating SC fate. We found that the microprocessor is required for the early onset of radial sorting by promoting the establishment of correct SC numbers. Furthermore, comparative RNA sequencing analysis revealed a regulatory requirement of the microprocessor to prevent inappropriately elevated or de novo expression of injury-induced genes in developing nerves. Additionally, we found that both Dicer and Dgcr8 are required for proper maintenance of myelinating SCs.

| Reagents
For a detailed list of reagents used in this study, see Supporting Information, Table S1.
2.3 | miRNA target prediction miRNA target prediction was based on TargetScan for mouse. We used the R/Bioconductor package targetscan.Mm.eg.db (version 0.6.1) to extract predicted targets from TargetScan (Supporting Information, Files S1 and S2).

| Crush injury surgery
Two-month-old C57BL/6J male mice were injected with buprenorphinum (Temgesic, Reckitt Benckiser) at a dosage of 0.1 mg/kg of bodyweight. The first injection was prior to the surgery, and treatment was maintained up to 2 days postsurgery. The surgery was performed under isoflurane anesthesia. An incision was performed at the height of the hip, and the sciatic nerve exposed on one side. The nerve was crushed using Dumont S&T JF-5 forceps (FST tools) by compressing the same site (at the sciatic notch) 3 times in a row for 10 s each. The wound was sealed, and the animals euthanized 3 days after the surgery. A 7 mm piece of the injured sciatic nerve was harvested from 1 mm up to 8 mm distal to the injury site. The contralateral noninjured sciatic nerve was also harvested and processed in parallel.

| RNA sequencing
Small RNA sequencing data from sciatic nerves at P1 and P60 were retrieved from the dataset available at the NCBI GEO database (GSE64562) (Gökbuget et al., 2015).
To generate Venn plots, we used the R package VennDiagram (v1.6.17). The long RNA sequencing data have been deposited at the NCBI GEO database with the primary accession code GSE109075.
The long RNA sequencing data for conditional Dicer mutants and controls was retrieved from the dataset available at the NCBI GEO database (GSE64562) (Gökbuget et al., 2015).
Alexa-dye conjugated secondary antibodies were used at a dilution of 1:500. Nuclei were stained by incubation with 0.1 ng/ml DAPI (Sigma) in PBS for 10 min. Microscopy was performed with the Axio Imager.
M2 (Zeiss). Images were quantified using Adobe Photoshop CS5 or CS6 on at least two sections, or more than 450 cells per animal.

| Reverse transcription and real-time PCR
Frozen SNs were freed of peri-and epineurial layers and homogenized in 20 μl Qiazol (Qiagen) using chilled pestles. After addition of another 480 μl of Qiazol, total RNA was extracted according to the Qiazol manufacturer's instructions (Qiagen). RNA precipitation was enhanced using GlycoBlue Coprecipitant (Thermo Scientific). Taqman miRNA assays were performed according to manufacturer's instructions using 2-5 ng total RNA input (Applied Biosystems). Reverse transcription of total RNA was performed using 50-100 ng input with Maxima First Strand cDNA Synthesis Kit (Thermo). Subsequent quantitative realtime PCR was performed with 2× FastStart Essential DNA Green Master Mix (Roche) on a Lightcycler 480II (Roche) using 2 μl 1:10 diluted total cDNA. PCR reactions were carried out with a minimum of three technical replicates per sciatic nerve sample. Primer sequences are supplied in Supporting Information, Table S2.

| Statistical analysis
Statistics were performed using GraphPad Prism 6. Statistical significance of two-group comparisons was assessed with the two-sided two-sample Student's t test and of multiple group comparisons with the two-way ANOVA with Tukey's multiple comparisons test was used, unless stated otherwise. Data distribution was assumed normal and variances were assumed equal, although these were not formally tested due to low sample sizes. Detailed information about error bars and sample sizes are included in all figure legends with further specifications in the Experimental Procedures section (see above). Sample size was not predetermined by statistical methods. Randomization was not applied unless stated otherwise. No blinding was applied with exception to all morphometric quantifications of electron micrographs, which were performed blinded. No samples or data were excluded from the analysis.  Figure S1). Furthermore, a confounding role of putative differences in genetic background is unlikely based on examinations of crossbred Dgcr8 and Dicer lines for multiple generations. Homozygous Dgcr8 mutants (heterozygous for Dicer) generated from these matings reproduced the developmental impairment displayed by Dgcr8 cKOs. Littermate homozygous Dicer mutants (heterozygous for DGCR8) closely resembled Dicer cKO (Supporting Information, Figure S1). Interestingly, no mature Remak bundles were found in any of the mutants mentioned above. Taken together, ablation of Dgcr8 and Drosha in SCs revealed a critical requirement of the microprocessor for early onset radial sorting, a necessity which is less evident when SCs lack Dicer.
3.2 | The microprocessor ensures accurate numbers of Schwann cells during radial sorting Next, we addressed the cellular mechanism connected to the pronounced early onset radial sorting defect observed upon microprocessor deletion. Successful radial sorting depends on, among other factors, correct expansion of the SC population to match the number of available axonal segments to be myelinated (Feltri et al., 2016). Failure to do so results in early onset radial sorting defects as observed in Dgcr8 and Drosha cKO. To address whether there are sufficient SCs in these mutants, we quantified the number of SCs in SNs of Dicer cKO, Dgcr8 cKO, Drosha cKO, and controls. We found fewer SCs in

| The microprocessor precludes incorrect upregulation of injury-response genes in developing Schwann cells
To gain insights into the molecular changes linked to the early onset radial sorting defects observed upon microprocessor ablation, we per-  (Figure 3b). To determine which biological processes are associated with these differences, we performed a gene ontology (GO) term enrichment analysis using the Enrichr tool (Chen et al., 2013). Given that the microprocessor mainly acts by giving rise to miRNAs that suppress target mRNAs, or in certain settings by directly binding and cleaving target mRNAs, we focused our analy- and Lif levels are not due to increased macrophage numbers (Supporting Information, Figure S2c), or a developmental delay (Supporting Information, Figure S2d,e), arguing toward de novo upregulation of these cytokines. Interestingly, miRNA target site prediction revealed conserved seed matches for SN-enriched miRNAs at P1 (Gökbuget et al., 2015) within the 3 0 -UTRs of Lif, Ngfr and Gdnf (Supporting Information, Figure S2f ). Of note, the 533 transcripts upregulated in Dgcr8 and Drosha cKO show also a trend toward upregulation in Dicer cKO (Supporting Information, Figure S2g). These findings concur with the interpretation of miRNAs as major regulatory effectors in our mutants. Furthermore, the data raise the distinct possibility that differences in miRNA depletion efficiencies may underlie some of the more robust upregulations found in Dgcr8 and Drosha cKO.
GO analyses of downregulated genes in Dgcr8 and Drosha cKO yielded mainly terms associated with myelination, likely reflecting the observed developmental delay compared to Dicer cKO and controls (Supporting Information, Figure S2h).
In summary, our molecular analysis revealed major differences in gene expression of Dgcr8 and Drosha cKO compared to Dicer cKO and controls, with a particular enrichment for injury-related gene sets.
3.4 | Microprocessor-dependent injury-response genes are expected targets of abundant miRNAs and differentially expressed transcription factors Given the de novo appearance or accumulation of repair SC markers and injury-related transcripts in response to microprocessor ablation in developing SCs, we sought to address to what extent this observed injury response matches the physiological response of SNs after injury. Thus, we performed whole-transcriptome analysis by RNA sequencing of wild-type mouse SNs 3 days post crush injury (dpc) in comparison to uncrushed and contralateral control nerves. We found 2802 significantly upregulated (≥twofold) and 1550 significantly downregulated genes (≤0.5-fold) in 3 dpc SNs compared to uncrushed and contralateral control nerves (Supporting Information, Figure S3a). GO term gene set enrichment analysis of these differentially upregulated genes unveiled large clusters associated with immune system functions such as immune regulation, leukocyte migration, response to injury, lymphocyte activation, blood coagulation, and neutrophil migration (Supporting Information, Figure S3b). Other large clusters were associated with protein localization, cell cycle, nucleosome assembly, telomere maintenance, GTPase activity, and negative regulation of phosphorylation. The significantly downregulated genes were mainly associated with fatty acid biosynthesis and cholesterol biosynthesis (Supporting Information, Figure S3b).
To examine which genes are both upregulated in expression upon SN injury and after developmental microprocessor ablation, we compared the differentially upregulated genes in 3 dpc SNs with those upregulated in both, Dgcr8 and Drosha cKO. We found that 196 out of 533 (37%) genes induced in developing Dgcr8 and Drosha cKO were also upregulated upon SN injury designated "jointly induced injury-response genes" (Figure 4a). These jointly induced injuryresponse genes were highly enriched for immune regulatory functions such as the GO terms "regulation of response to wounding," "leukocyte activation," "response to other organism," "lymphocyte activation," and "regulation of wound healing" (Figure 4b We next sought to apply an unbiased bioinformatics approach to explore putative regulatory networks underlying the differential expression of injury-response genes. Given that miRNAs often target central signaling hubs such as transcription factors to shape developmental gene expression (Ebert & Sharp, 2012), we predicted conserved transcription factor-binding sites within the promoter regions of all jointly differentially expressed injury-response genes (i.e., both upregulated and downregulated) upon SN injury and in Dgcr8 and Drosha cKO using the TRANSFAC/JASPAR tool embedded within Enrichr. For the genes differentially expressed at 3 dpc, we identified 233 significantly enriched transcription factor motifs. For the differentially expressed genes in Dgcr8 and Drosha cKO, we found 147 significantly enriched transcription factor (TF) motifs. Out of both sets, 145 were common. When we required additionally that motifassociated TFs had also to be upregulated upon injury and microprocessor ablation, we came up with eight candidates (Tcf4, c-Jun, Ikzf1, Tfap2a, Stat3, Spi1, Arid3a, and Irf8) that are predicted to regulate jointly differentially expressed injury-response genes (Figure 4d). The identified TFs included c-Jun, an already known key regulator of the injury response (Arthur-Farraj et al., 2012;Fontana et al., 2012) and interestingly other TFs that have not been previously implicated in the nerve injury response, such as Tfap2a, Tcf4, and Irf8 ( Figure 4d).

| Proper maintenance of myelinating Schwann cells depends on Dgcr8 and Dicer function
Given the de novo appearance of repair cell markers and the accumulation of injury-related gene sets, we hypothesized that the miRNA biogenesis pathway might be even more generally necessary for the suppression of such gene programs in SCs. To address whether the miRNA pathway is required in this way also in adulthood, we employed a tamoxifen (TAM)-inducible Cre recombinase system  (Figure 5e-j). In addition, we observed axons with thin myelin sheaths in Dicer and Dgcr8 iKO indicative of on-going remyelination (Figure 5f,h). However, not all DGCR8 and Dicer iKO animals  Microprocessor-dependent injury-response genes are predicted targets of abundant miRNAs and differentially expressed transcription factors. (a) Venn diagram of injury-induced genes (3 dpc) and genes induced in Dgcr8 and Drosha cKO. 196 genes are induced in both conditions (termed jointly induced injury-response genes). (b) Network graph produced by Cytoscape 3 showing significantly enriched GO biological process gene sets among the common injury-response genes imported into Enrichment Map (top panel). Nodes represent individual GO term gene sets. Edges represent the relatedness between the genes within the gene sets. Genes are highly enriched for immune regulatory and injury-related functions. Top 10 significantly enriched GO terms are shown in the panel below. (c) Heatmap of jointly expressed injury-response genes. Color intensities represent row Z-score, red indicating higher expression and white indicating lower expression. (d) Upregulated transcription factors (TFs) with predicted binding motifs in the promoter regions of jointly differentially expressed injury-response genes (i.e., both upregulated and downregulated) in 3 dpc SNs compared to contralateral and uncrushed controls and in SNs of Dg and Dro at P1 compared to controls (fold change ≥2, FDR ≤ .05). Negative decadic logarithm of adjusted p value of TF target enrichment among injury induced genes (injury) and genes induced in Dgcr8 and Drosha cKO (Dg + Dro) is color-coded (white indicating high and blue low p values). Targeted (black) and nontargeted (yellow) jointly differentially expressed injury-response genes are shown for each TF. (e) Top 10 expressed miRNAs in P1 and P60 SNs of wild-type mice and their predicted target sites (based on TargetScan) among the jointly induced injury-response genes and upregulated TFs predicted to target jointly differentially-expressed injury-response genes. Target site conservation and quality is represented by the P CT score and nonpredicted targets (NT) in grey. Venn diagram below indicating 7 out 10 most abundant miRNAs overlap between P1 and P60 [Color figure can be viewed at wileyonlinelibrary.com] were affected morphologically, in agreement with the incomplete penetrance previously reported in TAM-induced Dgcr8 and Dicer mutants using inducible systems (Lin et al., 2015;Shin, Shin, McManus, Ptacek, & Fu, 2009;Viader et al., 2011). Assessment of toluidine bluestained sciatic nerve sections revealed 11 (out of 26) DGCR8 iKO animals without detectable changes between 3 and 6 mpT, and 4 (out of 11) Dicer iKO mice without detectable changes even at 14 mpT. Of note, incomplete penetrance was observed on both genders. As a consequence, we restricted the analysis to affected mice. As expected, levels of Dicer mRNA in Dicer iKO and Dgcr8 mRNA in Dgcr8 iKO were reduced in comparison to controls at 3 mpT (Figure 5b Successful radial sorting depends on an accurate pool of immature SCs to match the number of axonal segments to be myelinated. The observed deficit in SCs in P1 Dgcr8 and Drosha cKO is therefore likely to be causal for the early impairments in radial sorting. Such impairments have also previously been observed in response to increased SC death or decreased SC proliferation (Brinkmann et al., 2008;Messing et al., 1992;Porrello et al., 2014). As proliferation was not detectably changed in Dgcr8 and Drosha cKO, the increase in SC apoptosis likely contributes to the observed SC deficit. Additionally, given the developmental induction of injury-response genes in Dgcr8 and Drosha cKO, this may further lead to some form of early SC reprogramming, drifting away from the transcriptome boundaries normally required by immature SCs. In this way, such changes may also diminish the pool of immature SCs that are capable of performing radial sorting.
The remarkable ability of SCs to respond to injury and to activate a defined repair program is beneficial to the recovery of peripheral nerves and stands in sharp contrast to the CNS injury response (Brosius Lutz & Barres, 2014). However, such a program needs to be tightly controlled. Our results from developmental Dgcr8 and Drosha cKO reveal that the microprocessor is critically involved in this regulation of the injury-response program by preventing its inappropriate activity during PNS development. In support of this interpretation, we detected prominent induction of Shh and Gdnf upon loss of microprocessor in P1 SN, both of which are described markers of repair SCs (Jessen & Mirsky, 2016). The presence of these markers at this time point coincides with the onset of the cellular phenotype observed upon loss of microprocessor. Along with the presence of these repair SC markers, we verified the expression of other injury-response genes by comparing whole-transcriptome data after loss of microprocessor at P1 with data gathered from 3 dpc SNs, consistent with an early reprogramming of microprocessor-deficient SCs into repair-like SCs.
Nevertheless, microprocessor-dependent genes that are not induced upon injury might also contribute to impaired radial sorting.
In adult myelinating SCs, we found that induced deletion of Dgcr8 or Dicer leads to a failure in proper maintenance of the myelinating SC fate, resulting in the presence of demyelinated axons. Additionally, both Dgcr8 and Dicer iKO express a large fraction of genes induced upon active SC reprogramming. Notably, the requirement of both Dgcr8 and Dicer to counteract injury-response gene expression in adult SCs is in apparent contradiction to the requirement of the microprocessor, but not Dicer, in the suppression of these genes in developing SCs. However, the major relevant described functions of both the microprocessor and Dicer are in the biogenesis of canonical miR-NAs. In this context, we show that stage-specifically enriched miRNAs in developing and adult nerves are predicted to target a large fraction of the injury-induced transcripts that are enriched upon loss of the microprocessor in developing SNs and upon induced Dgcr8 or Dicer deletion in adult SCs. Upon closer inspection, however, we noticed that the genes induced in developing SCs in Dgcr8 and Drosha cKO are also mildly increased in Dicer cKO. In the context of miRNAsbased regulation, this might be interpreted in that miRNAs are less efficiently depleted and/or less functional in Dicer cKO than in Dgcr8 and Drosha cKO. Related to this, it has been reported that loss of Drosha compared to Dicer does not equally affect the population of expressed miRNAs (Kim, Kim, & Kim, 2016). While loss of Drosha caused a strong decrease of all expressed miRNAs, loss of Dicer more specifically resulted in reduced levels of miRNAs derived from the 3 0strand (3p) leaving the 5 0 -strand (5p) miRNAs less affected. The authors suggested a possible Dicer-independent mechanism to mature pre-miRNAs to mature 5p-species. While such a mechanism might contribute to the milder radial sorting defects and the less prominent upregulation of the microprocessor-dependent genes in Dicer cKO, differences in stabilities of the genetically targeted proteins-to-be-deleted may also take part in shifting miRNA depletion efficiencies.
In Dicer and Dgcr8 iKO SN, we found incomplete phenotypic penetrance. Individual miRNA mutants often lack a phenotype or show incomplete penetrance if not challenged by additional environmental stress factors (Mendell & Olson, 2012). Thus, the lack of such stress factors together with desynchronized miRNA depletion timings in individual SCs may contribute to the penetrance variability that we observed.
Taken together, we favor a general mechanism involving miRNAs that prevents inappropriate activity of the injury response in developing and adult SCs. However, we cannot exclude other distinct mechanisms contributing to or accounting for the defects that we observed in our mutants during development and myelin maintenance. For example, the microprocessor was shown to bind also to pri-miRNAlike structures present in other RNA species such as mRNAs, small nuclear-RNAs, and long non-coding RNAs (Macias, Cordiner, & Caceres, 2013). During adult neurogenesis, Drosha, but not Dicer or miRNAs, is required to favor neuronal differentiation over oligodendrogenesis by binding and cutting a pri-miRNA-like hairpin within Nfib mRNA (Rolando et al., 2016). Similarly, for dendritic cell development, Drosha is required to cleave hairpins within mRNAs encoding inhibitors of myelopoiesis (Johanson et al., 2015).
We envision that our data, including the bioinformatics analysis, constitute an important resource that motivates multiple future studies to individually evaluate the physiological impact of predicted miRNA-target and transcription factor relationships, which could act to safeguard early onset radial sorting and the myelinated adult SC state.
Our data highlight the multifaceted requirement of the miRNA biogenesis pathway in SC physiology during development and adulthood, potentially also with consequences in SC-related pathologies as seen in neuropathies, or in the processes following after nerve injury and in regeneration. These aspects warrant further investigations.

ACKNOWLEDGMENTS
The authors thank Dr Olivier Voinnet for intellectual input; Drs Andrea Erni, Tobias Kockmann, and Verdon Taylor for support and fruitful discussions; Dr Dies Meijer for Dhh-Cre mice, Dr Gregory Hannon for floxed Dicer mice, Dr Dan Littman for floxed Drosha mice, and Dr Markus Stoffel for floxed DGCR8 mice; Joanne Jeker for excellent support with data analysis. Technical assistance by the ETH Phenomics Center, Functional Genomics Center Zurich, and SCOPEM is gratefully acknowledged.

CONFLICT OF INTEREST
No conflict of interests declared.

DATA AVAILABILITY
RNA sequencing data have been deposited at the NCBI GEO database with the primary accession code GSE109075.