Abstract

   During peripheral nerve (PN) development, unmyelinated axons (nmAs)
   tightly fasciculate before being separated and enveloped by
   non-myelinating Schwann cells (nmSCs), glial cells essential for
   maintaining nmA integrity. How such a switch from axon-axon to
   axon-glia interactions is achieved remains poorly understood. Here, we
   find that inactivating SC-derived SEMA3B or its axonal receptor
   components in mice leads to incomplete nmA separation and envelopment
   by nmSCs, eliciting hyperalgesia and allodynia. Conversely, increasing
   SEMA3B levels in SCs accelerates nmA separation and envelopment. SEMA3B
   transiently promotes nmA defasciculation accompanied by cell adhesion
   molecule (CAM) endocytosis, subsequently facilitating nmA-nmSC
   association. Restoring SEMA3B expression following PN injury promotes
   nmA-nmSC re-association and alleviates hyperalgesia and allodynia. We
   propose that SEMA3B-induced CAM turnover facilitates a switch from
   axon-axon to axon-glia interactions promoting nmA envelopment by nmSCs,
   which may be exploitable for alleviating PN injury-induced pain by
   accelerating the restoration of nmA integrity.

   Subject terms: Glial development, Neuropathic pain
     __________________________________________________________________

   Nerve fibers relaying pain information from peripheral organs are
   normally supported by specialized glial cells called nonmyelinating
   Schwann cells. Here, the authors show that a secreted factor, SEMA3B,
   promotes nerve fiber-Schwann cell interactions necessary for
   suppressing pathological pain.

Introduction

   Schwann cells (SCs), the predominant glial cells within the peripheral
   nervous system (PNS), arise during embryonic development from neural
   crest cells in a well-defined sequence of events involving the
   formation of Schwann cell precursors (SCPs), followed by the generation
   of immature Schwann cells (iSCs), and finally, the differentiation into
   myelinating Schwann cells (mSCs) and non-myelinating Schwann cells
   (nmSCs). mSCs wrap individual axons with a diameter >1 μm to form a
   myelin sheath essential for rapid saltatory impulse propagation, while
   nmSCs envelop multiple axons with a diameter of <1 μm to form Remak
   bundles (RBs)^[46]1,[47]2. nmSCs are critical in maintaining the
   integrity and functionality of unmyelinated (nociceptive sensory and
   autonomous) axons (nmAs)^[48]3. Disruption of RB assembly and the
   association between nmSCs and nmAs due to genetic mutations or nerve
   injuries leads to impaired nmA function and hypersensitivity
   (allodynia)^[49]4–[50]10. Furthermore, in the absence of support from
   nmSCs, nmAs are prone to collateral sprouting and aberrant
   re-innervation of non-nociceptive end-organs^[51]11–[52]14, which are
   implicated in initiating chronic pain conditions^[53]15–[54]17.

   Peripheral nerve (PN) development is initiated by pioneer axons laying
   out principal nerve trajectories, which are subsequently joined by the
   bulk of developing axons before diverging toward their prospective
   end-organ targets^[55]18,[56]19. Peripheral axon fasciculation is
   mainly mediated by cell adhesion molecules (CAMs), including NCAM,
   L1CAM, TAG1, DSCAM, Cadherins, and Integrins^[57]20, and can be
   simultaneously influenced by ‘classical’ axon guidance molecules such
   as semaphorins and Ephrins^[58]21–[59]23. SCPs initially migrate
   alongside peripheral axons but do not penetrate the axon
   fascicles^[60]24. During the process known as axonal radial sorting,
   prospective mSCs insert lamellipodia into the axon fascicles and wrap
   around large-caliber (myelinated) axons (mAs), leaving fascicles with
   an increasing proportion of small-caliber nmAs. Ultimately, nmAs are
   separated and enveloped by nmSC processes to form RBs^[61]25. While the
   transition from nmA-nmA to nmA-nmSC interactions is critical for RB
   assembly, identifying key regulatory molecules is challenging due to
   the simultaneous occurrence of SCP proliferation, differentiation,
   myelination, and mA-mSC interactions during RB formation. For instance,
   molecules such as GABBR1, LRP1, and FGFR have been implicated in RB
   formation, but these molecules also contribute to other aspects of PN
   development, such as radial sorting and myelination^[62]26–[63]28,
   complicating the determination of their specific roles in mediating
   nmA-nmSC interactions.

   While it has been well-established that radial sorting and the
   specification of SCs into mSCs versus nmSCs is critically determined by
   the level of axonal NRG1 activating ERBB2/3 signaling in
   SCPs^[64]29,[65]30, how mSCs or nmSCs overcome pre-established
   axon-axon interactions mediated by CAMs in favor of axon-SC
   interactions remains to be elucidated. Reconfiguration of cell-cell
   adhesions is a common occurrence during embryonic development, tissue
   homeostasis, and pathology^[66]31,[67]32. For instance, tissue invasion
   by macrophages or trans-endothelial migration of leukocytes involves
   signals that promote localized removal by endocytosis of CAMs from
   cell-cell junctions^[68]33,[69]34. Myelination is preceded by the
   endocytosis-dependent clearance of internodal CAMs on mAs in response
   to mSC-derived signals^[70]35. During axon guidance, moreover, axon
   rerouting from pre-formed fascicles involves the regulation of adhesive
   forces between axons, such as through post-translational modifications
   weakening homophilic CAM interactions^[71]36,[72]37. Semaphorins,
   particularly SEMA3A and SEMA3E, have been implicated in regulating axon
   fasciculation at navigational choice points^[73]38, but how,
   mechanistically, Semaphorin signaling influences the action of CAMs
   remains to be determined.

   To identify mechanisms underlying the switch from axon-axon to
   axon-glia interactions during RB assembly, we employed single-cell
   transcriptomics of the murine sciatic nerve, thereby identifying
   SC-derived SEMA3B as the molecule facilitating the transition from
   nmA-nmA to nmA-nmSC interactions. Inactivation of SC-derived SEMA3B led
   to incomplete nmA separation and deficient RB assembly, leading to
   allodynia, which was phenocopied by loss of its putative axonal
   receptor components NRP1 and L1CAM. Alterations in SC-derived SEMA3B
   levels were associated with corresponding changes in axonal CAM levels
   during RB assembly. Additionally, recombinant SEMA3B promoted the
   endocytosis of axonal CAMs and the defasciculation of cultured nmAs. PN
   injury triggered a rapid decrease in SEMA3B expression, whereas its
   transgenic restoration accelerated nmA-nmSC re-association and
   alleviated allodynia. We propose that SEMA3B drives the transition from
   axon-axon to axon-glia interactions to facilitate RB assembly, which
   may provide a potential therapeutic avenue for alleviating PN
   injury-induced chronic pain by promoting nmA-nmSC re-association and
   restoring nmA integrity.

Results

Single-cell profiling of the murine sciatic nerve

   In mice, developing nmSCs begin extending lamellipodia into
   small-diameter axon fascicles around postnatal day 0 (P0) and
   progressively separate individual axons between P4 and P28. By P60,
   nearly all axons within the RB were segregated, leaving a minimal
   number of axons unwrapped (Fig. [74]1a), and these few unwrapped axons
   persisted until P120. Correspondingly, the Remark Bundle Mature (RBM)
   index, which quantifies the percentage of axons fully wrapped by nmSC
   cytoplasm within each RB, increased from 1.54 ± 0.29% at P0 to
   96.98 ± 0.81% at P60 and remained relatively stable thereafter
   (Fig. [75]1b), indicating the complete developmental maturity of the RB
   by P60. While prior single-cell studies have mapped in great detail
   cellular diversity and lineage relationships in developing and adult
   mouse peripheral nerves (PNs)^[76]39–[77]42, existing datasets do not
   completely cover the postnatal developmental window critical for RB
   assembly (P0–P60). To characterize the molecular underpinnings of RB
   assembly in more detail, we generated a single-cell RNA sequencing
   (scRNA-seq) atlas of murine sciatic nerves spanning five key stages of
   RB development (P0, P7, P14, P28, and P60). Our dataset comprises
   11,443 high-quality whole-cell transcriptomes (10x Genomics), enabling
   high resolution analysis of transcriptional dynamics during RB assembly
   (Fig. [78]1c, Supplementary Fig. [79]1a, b and Supplementary
   Table [80]1). Unsupervised clustering with uniform manifold
   approximation and projection (UMAP) identified 19 transcriptomically
   distinct clusters (Supplementary Fig. [81]1c), which were assigned to
   endothelial cells, fibroblasts, smooth muscle cells, monocytes,
   macrophages, and SCs (Fig. [82]1c and Supplementary Fig. [83]1d).
   Pseudotime analysis of Sox10^+/Plp1^+ SCs revealed a clear transition
   from dividing progenitor SCs to bi-potential intermediate SCs, followed
   by a divergence into two populations of maturing SCs (Fig. [84]1d, e).
   These populations were further classified into six transcriptionally
   distinctive clusters: SCPs, iSCs, pro-myelinating SCs (pmSCs),
   pro-nonmyelinating SCs (pnmSCs), mSCs, and nmSCs (Fig. [85]1f, j,
   Supplementary Fig. [86]1e–j and Supplementary Table [87]1), confirmed
   by RNA fluorescence in situ hybridization (FISH) of sciatic nerves from
   varying developmental stages (Fig. [88]1g–i).

Fig. 1. Single-cell transcriptome analysis of SCs identifies SEMA3B as a
potential molecule regulating RB assembly.

   [89]Fig. 1
   [90]Open in a new tab

   a Representative TEM images of the mouse sciatic nerve showing the
   spatiotemporal assembly of RB. nmSC lamellipodia either ensheath axons
   (closed arrowheads) or separate adherent axons (open arrowheads).
   Asterisks and circles indicate axons entirely or partially wrapped by
   nmSC lamellipodia, respectively. Data are representative of 3 mice
   across all developmental stages. Abbreviations: n nucleus, mA,
   myelinated axon. b Quantitative assessment of RB assembly progression,
   as measured by the Remark Bundle Maturity (RBM) Index. Data are
   presented as mean ± SEM. Sample size: n = 21/5 (number of RBs/number of
   mice), 23/4, 18/5, 27/6, 26/5, 28/5, and 16/4 for P0, P7, P14, P21,
   P28, P60, and P120, respectively. c UMAP plot illustrating the
   diversity of cell subtypes. d UMAP plot illustrating the cell-cycle
   phase. e UMAP plot illustrating the pseudotime assigned to each SC. f
   UMAP plots illustrating the expression of marker genes for SCPs
   (Mki67), iSCs (Ednrb), mSCs (Ncmap), and nmSCs (Scn7a). g-i FISH on
   cryosections of the mouse sciatic nerve showing indicated gene/protein
   expression in SOX10^+ SC subsets. Arrowheads and arrows indicate
   MKI67^+Ednrb^+ SCPs and MKI67^-Ednrb^+ iSCs in (g), Ednrb^+Scn7a^+ iSCs
   and Ednrb^-Scn7a^+ nmSCs in (h), Ncmap^+Scn7a^- mSCs and Ncmap^-Scn7a^+
   nmSCs in (i), respectively. Data are representative of 3 mice. j
   Relative proportions of SC lineages at each stage. k scFates-based
   pseudotemporal UMAP analysis of SC lineage progression. l UMAP plot
   indicating SC subtype diversity. m Schematic of genetic programs and
   potential time frames for RB assembly and myelination underlying SC
   development. n Volcano plot for differentiated expressed genes (DEGs)
   in SCs compared to non-SCs. DEGs were identified using a two-tailed
   Wald test with Benjamini-Hochberg adjustment for multiple comparisons
   (FDR-adjusted p value < 0.05 and |log₂ fold change|> 0.25). o Dot plot
   of the 15 most significantly over-represented KEGG pathways in SCs.
   Pathway enrichment was analyzed by a two-tailed Wald test with
   Benjamini-Hochberg correction (FDR-adjusted p value < 0.05 and |log₂
   fold change|> 0.25). p Dot plot showing relative expression of selected
   genes in the axon guidance pathway across different SC types and ages.
   Source data are provided as a Source Data file.

SEMA3B is highly expressed in developing nmSCs

   Our scRNA-seq of SCs aligned with previous single-cell mapping
   efforts^[91]39,[92]42, with comparatively deep transcriptome coverage
   and clear separation of mSC and nmSC lineages (Fig. [93]1k–m). We
   further exploited the additional coverage of the time window of RB
   assembly to identify novel features of nmSC maturation relevant to the
   transition from nmA-nmA to nmA-nmSC interactions. Among the top 15 most
   abundant pathways identified by gene ontology term (GOT) analysis were
   “axon guidance”, “cell adhesion molecules” and “ECM-receptor
   interaction” (Fig. [94]1n, o), which, we reasoned, might encompass
   molecules directly involved in axon-axon and/or axon-glia interactions
   during RB assembly^[95]18,[96]43. Of note, within the axon guidance
   pathway, Sema3b and Sema6d expression levels were significantly
   upregulated in SCs, with slightly higher levels in pnmSCs and nmSCs
   compared to pmSCs and mSCs (Fig. [97]1p and Supplementary Fig. [98]1k).
   Given the role of other class-3 semaphorins in regulating axon-axon
   interactions^[99]38, we focused our subsequent attention on Sema3b as a
   potential molecule influencing axon-axon and/or axon-glia interactions
   during RB assembly. Consistent with our scRNA-seq data (Fig. [100]1p),
   Sema3b transcripts were confirmed through real-time quantitative PCR
   (qPCR) and FISH in SOX10^+ SCs at P0, the expression of which peaked at
   P14, while high expression levels were maintained until P28, before
   gradually declining, yet not completely receding, towards adulthood
   (Fig. [101]2a and Supplementary Fig. [102]2a). Lower levels of Sema3b
   transcripts were detected in SOX10^+ SCPs and iSCs prior to P0
   (Supplementary Fig. [103]2a). Notably, although scRNA-seq detected
   Sema3b transcripts in both nmSCs and mSCs, with relatively higher
   levels in nmSCs (Fig. [104]1p and Supplementary Fig. [105]2b), SEMA3B
   protein was exclusively detected in SOX10^+MBP^- SCs at P14, and these
   SEMA3B^+SOX10^+MBP^- SCs wrapped multiple nmAs (Fig. [106]2b). While
   the post-translational mechanisms underlying the differences in Sema3b
   mRNA and SEMA3B protein distribution in mature SCs remains to be
   elucidated, it is tempting to speculate that this may involve
   differential cell surface-retention reflecting nmSC-specific synthesis
   of class III Semaphorin-binding ECM components (Supplementary
   Fig. [107]1l)^[108]44,[109]45. Taken together, these data identified
   SEMA3B as a highly selective marker for pnmSCs and nmSCs (collectively
   referred to as nmSCs).

Fig. 2. SC-derived SEMA3B regulates RB assembly.

   [110]Fig. 2
   [111]Open in a new tab

   a Quantitative real-time PCR (qPCR) analysis of Sema3b mRNA expression
   in sciatic nerves across postnatal development (P0, P7, P14, P28, and
   P60). n = 3 sciatic nerves per time point. b Transversal section of the
   P14 sciatic nerve showing that SEMA3B^+SOX10^+MBP^- SCs ensheath
   unmyelinated TUJ1^+ axons. The arrowhead and arrow indicate
   SEMA3B^+SOX10^+MBP^- nmSCs and SEMA3B^-SOX10^+MBP^+ mSCs, respectively.
   The experiment was repeated 3 times on 3 mice. c, d Representative TEM
   images of the sciatic nerve from P60 control and Sema3b^∆SC mice. nmSC
   lamellipodia are pseudo-colored to highlight their locations. Asterisks
   denote adjacent axons and circles denote partially wrapped axons.
   Arrowheads indicate lamellipodia protruding into the endoneurial space.
   The experiment was repeated 3 times on 3 mice. e, f Lateral views of
   3D-reconstructed RB from the sciatic nerve of P60 control and
   Sema3b^∆SC mice. g Quantification of the RBM indexes from P60 control
   (n = 7) and Sema3b^∆SC (n = 7) mice. h Quantification of the RBM
   indexes from P7 (n = 5 for both groups) and P28 Sema3b^i∆SC (n = 5 for
   both groups) mice with coil (CO) and tamoxifen (TM) treatment at P0. i
   Measurements of sensitivity to laser heat in control (n = 6 for all
   groups) and Sema3b^∆SC (n = 7 for all groups) mice. j Measurements of
   sensitivity to von Frey filaments in control (n = 6) and Sema3b^∆SC
   (n = 6) mice. All data are presented as mean ± SEM. **p < 0.01,
   *p < 0.05, ns not significant. Detailed statistics are provided in
   Supplementary Table [112]2. Source data are provided as a [113]Source
   Data file.

SEMA3B is dispensable for PN axon fasciculation, radial sorting, and
myelination

   To elucidate the contribution of SC-derived SEMA3B to PN development,
   we generated Sox10^Cre;Sema3b^loxp/loxp mice, in which Sema3b was
   selectively ablated in NCCs and the entire SC lineages (Supplementary
   Fig. [114]2c)^[115]46. Because the inactivation of other class-3
   semaphorins resulted in PN axon defasciculation^[116]47–[117]49,
   potentially complicating the deciphering of RB assembly impacts, we
   thoroughly investigated PN axon fasciculation in
   Sox10^Cre;Sema3b^loxp/loxp embryos. Corresponding with the low
   expression levels of SEMA3B/Sema3b in SCs during embryonic stages
   (Supplementary Fig. [118]2a), no alterations in PN axon fasciculation
   were observed in E12.5 and E14.5 embryos (Supplementary
   Fig. [119]2d–g), thereby ruling out a significant contribution of
   SC-derived SEMA3B to embryonic peripheral axon fasciculation. To
   further delineate the requirements of SEMA3B within SCs, we generated
   Dhh^Cre;Sema3b^loxp/loxp mice (Sema3b^∆SC) to ablate Sema3b
   specifically in SC lineages (Supplementary Fig. [120]2h)^[121]50.
   Sema3b^∆SC mice appeared behaviorally indistinguishable from control
   littermates in adulthood. The survival, proliferation, and
   differentiation of SCs in these mice were comparable to their control
   counterparts (Supplementary Figs. [122]2i–o, [123]3a). Light microscopy
   and transmission electron microscopy (TEM) analysis of sciatic nerves
   revealed no differences in cellular composition or organization,
   including myelination status, between control and Sema3b^∆SC mice
   (Supplementary Fig. [124]3b–g). Thus, SEMA3B is dispensable for SC
   specification, radial sorting, and subsequent SC differentiation and
   myelination, indicating that Sema3b^∆SC mice would allow us to
   selectively study the contribution of SEMA3B to RB assembly.

SEMA3B is required for nmA separation, envelopment, and integrity

   In control mice, TEM analysis of sciatic nerves and TEM-based partial
   3D-reconstruction of RBs revealed that nmSC processes physically
   separated axons and enveloped nearly all small-diameter (nmA) axons
   within RBs at P60 (Fig. [125]2c, e). In contrast, P60 Sema3b^∆SC mice
   displayed significantly altered RB structures, with some nmAs partially
   enveloped and many nmAs remaining unseparated (Fig. [126]2d, f).
   Additionally, some nmSC processes protruded into the endoneurial space
   (Fig. [127]2d). Consequently, the contact surface between nmAs and
   nmSCs within the RBs of Sema3b^∆SC mice was significantly reduced
   (Fig. [128]2g and Supplementary Video [129]1). However, no evident
   defects in radial sorting were observed, as almost no axons greater
   than 1.0 μm in diameter were detectable within the RBs of Sema3b^∆SC
   mice (Fig. [130]2d, f). We next asked if the defective RB assembly was
   an indirect effect of low SEMA3B expression in the embryonic SC lineage
   in Sema3b^∆SC mice. Therefore, we established
   Plp1^CreERT2;Sema3b^loxp/loxp mice (Sema3b^i∆SC) to temporally ablate
   Sema3b in postnatal SCs via tamoxifen (TM) administration at
   P0^[131]51. Sema3b^i∆SC mice exhibited RB assembly defects, including
   failed nmA separation and envelopment (Fig. [132]2h and Supplementary
   Fig. [133]3h, i), similar to those observed in Sema3b^ΔSC mice, thereby
   supporting the conclusion that SC-derived SEMA3B is essential for
   proper nmA-nmSC association and RB assembly. The nmAs within the RBs
   predominantly comprise nociceptive axons^[134]52, and defective RB
   formation can trigger chronic (neuropathic) pain conditions,
   manifesting as hypersensitivity to both noxious and non-noxious stimuli
   (hyperalgesia and allodynia, respectively)^[135]26. Consistent with the
   observed nmA-nmSC association failures, Sema3b^∆SC mice exhibited
   significantly reduced thermal and mechanical withdrawal thresholds,
   indicating the development of thermal and mechanical hyperalgesia and
   allodynia (Fig. [136]2i, j). Thus, SC-derived SEMA3B is required for
   nmA-nmSC association necessary for RB assembly and nmA integrity.

Elevating SEMA3B expression accelerates nmA separation and envelopment

   RB assembly progresses over an extended postnatal period, coinciding
   with a peak in Sema3b expression around P14, following a phase of
   relatively weak expression (Fig. [137]2a and Supplementary
   Fig. [138]2a), raising the question of whether SEMA3B levels could
   serve as a rate-limiting factor in initiating nmA-nmSC association. To
   investigate whether SEMA3B is not only required but also sufficient for
   promoting RB assembly, we generated a mouse line facilitating the
   Cre-activated transgenic overexpression of Sema3b from the Rosa26 locus
   (Rosa26^floxed-stop-Sema3b) (Supplementary Fig. [139]3j, k). We then
   established Dhh^Cre;Sema3b^floxed-stop-Sema3b (Sema3b^tgSC) mice to
   selectively activate transgenic SEMA3B expression in SCs
   (Fig. [140]3a). FISH on the sciatic nerve of P14 Sema3b^tgSC mice
   confirmed significantly elevated Sema3b expression level in SCs
   compared to control mice, accompanied by the simultaneous expression of
   TOM in the SCs of Sema3b^tgSC mice (Fig. [141]3b–d). Ultrastructural
   examination of RBs in Sema3b^tgSC mice through TEM revealed
   significantly increased nmA separation and envelopment at P7 and P28
   compared to control counterparts (Fig. [142]3e–g), indicating that
   elevated SEMA3B levels expedited RB assembly. Collectively, SC-derived
   SEMA3B is not only necessary but also sufficient for directing nmA
   separation and envelopment during RB assembly.

Fig. 3. Elevating SEMA3B expression level in SCs accelerates RB assembly.

   [143]Fig. 3
   [144]Open in a new tab

   a Schematic illustration of the strategy for generating Sema3b^tgSC
   mice. b, c FISH on the sciatic nerve showing an increased Sema3b
   expression level in SOX10^+ SCs of P14 Sema3b^tgSC mice. Arrowheads and
   arrows in b indicate Sema3b^+SOX10^+ SCs and Sema3b^-SOX10^+ SCs. Note
   that in Sema3b^tgSC mice, Sema3b expression is ubiquitous in all
   SOX10^+ SCs, which concurrently express TOM. d Quantification of Sema3b
   transcript level with qPCR in the sciatic nerve of P14 control (n = 6)
   and Sema3b^tgSC(n = 6) mice. e, f Representative TEM images of the
   sciatic nerve from P7 and P28 control and Sema3b^tgSC mice. Asterisks
   and circles denote axons that are adjacent and partially wrapped,
   respectively. Arrowheads indicate lamellipodia protruding into the
   endoneurial space. g Quantification of the RBM indexes from P7 and P28
   control (n = 6 for both P7 and P28) and Sema3b^tgSC (n = 6 for both P7
   and P28) mice. All data are presented as mean ± SEM. *p < 0.05.
   Detailed statistics are provided in Supplementary Table [145]2. Source
   data are provided as a [146]Source Data file.

Potential SEMA3B receptor components are present on nmAs

   The involvement of SEMA3B in RB assembly suggested the presence of
   interaction partners on nmAs underlying its apparent influence on
   nmSC-nmA interactions. Class-3 semaphorins typically signal through a
   receptor complex composed of Neuropilins (NRP) and Plexins
   (PLXN)^[147]53, but can encompass interactions with several other
   protein family members, such as L1CAM and VEGFR^[148]54–[149]56. To
   investigate the presence of potential SEMA3B receptor components on
   nmAs during RB assembly, we conducted snRNA-seq of dorsal root ganglia
   (DRGs), comprising primary somatosensory neurons projecting mAs and
   nmAs through the sciatic nerves, from P14 mice (Fig. [150]4a,
   Supplementary Fig. [151]4a and Supplementary Table [152]1). Based on
   established distinct SN marker genes^[153]57, we compared the
   expression levels of potential SEMA3B receptor components in Nefh^+
   myelinated sensory neurons (mSNs) and Nefh^- unmyelinated sensory
   neurons (nmSNs). Our analysis revealed high Nrp1 expression in nmSNs,
   whereas Nrp2 was expressed at relatively low levels in nmSNs but at
   high levels in spinal motor neurons, which contribute a significant
   proportion of peripheral mAs (Fig. [154]4b, c and Supplementary
   Fig. [155]4b–e). FISH on P14 DRGs confirmed high Nrp1 expression in
   CGRP^+ and TH^+ nmSNs, with negligible expression in IB4^+ nmSNs,
   PRPH^+ nmSNs, NF200^+ mSNs, and PV^+ mSNs (Supplementary
   Fig. [156]4f–h). Immunodetection of NRP1 specifically in CGRP^+ and
   TH^+ nmAs in P14 sciatic PNs further supported these findings
   (Fig. [157]4d, e). While all Plexin family members were expressed at
   relatively low levels in nmSNs (Fig. [158]4b), L1CAM, which was
   previously implicated in directly binding NRP1/2 to affect class-3
   semaphorin signaling^[159]54,[160]58, was exclusively expressed at high
   levels in nmSNs (Fig. [161]4b, c, f and Supplementary Fig. [162]4i–l).
   Moreover, high levels of both L1CAM and NRP1 were co-detected on
   sciatic nmAs (Fig. [163]4g). Thus, during the developmental period of
   RB assembly, peripheral nmAs express at least two SEMA3B receptor
   components at significant levels: NRP1 and L1CAM.

Fig. 4. SEMA3B interacts with axonal NRP1/L1CAM receptor component to
regulate RB assembly.

   [164]Fig. 4
   [165]Open in a new tab

   a UMAP plot illustrating cell subtype diversity from P14 DRG. b Dot
   plot comparing expression levels of axon guidance pathway genes in
   Nefh^− nmSNs and Nefh^+ mSNs. c UMAP plots illustrating the expression
   levels of Nefh, Nrp1, Nrp2, and L1cam in SNs. d–g Transverse sections
   of P14 sciatic nerve. NRP1 is expressed in TH^+ (d) and CGRP^+ (e)
   unmyelinated axons, and L1CAM is expressed in almost all unmyelinated
   axons (f). NRP1 and L1CAM are simultaneously expressed in subsets of
   unmyelinated axons (g). The experiment was repeated 3 times on 3 mice.
   h–k TEM images of the sciatic nerve from P60 control, Nrp1^∆SN,
   L1cam^∆SN, and Nrp1/L1cam^∆SN mice. l Quantification of the RBM indexes
   from control, Nrp1^∆SN, L1cam^∆SN, and Nrp1/L1cam^∆SN mice. n = 6 for
   all groups. m STED images illustrating the distribution of NRP1 and
   L1CAM particles within cultured sensory axon shafts following Fc and
   SEMA3B treatment. Arrowheads indicate particles with NRP1 and L1CAM
   co-localization. n Quantification of the number of L1CAM particles in
   the axon shaft with Fc (n = 7 axon shafts from 3 DRGs) and SEMA3B
   (n = 8 axon shafts from 3 DRGs) treatment. o Quantification of the
   numbers of particles with NRP1 and L1CAM co-localization in the axon
   shaft with Fc (n = 7 axon shafts from 3 DRGs) and SEMA3B (n = 8 axon
   shafts from 3 DRGs) treatment. p Representative images of the sensory
   axon shafts from control, Nrp1^∆SN, L1cam^∆SN, and Nrp1/L1cam^∆SN
   embryos showing the lateral filopodia of axon shaft. q Quantification
   of the percentages of collapsed sensory growth cones following Fc
   (n = 5 DRGs from 3 control embryos) and SEMA3B (control, n = 6 DRGs/3
   embryos; Nrp1^∆SN, n = 6 DRGs/3 embryos; L1cam^∆SN, n = 5 DRGs/3
   embryos; Nrp1/L1cam^∆SN, n = 6 DRGs/3 embryos) treatment. r
   Quantification of the filopodia numbers of cultured DRG shafts
   following treatment with Fc (n = 7 axon shafts/3 DRGs) and SEMA3B
   (control, n = 5 axon shafts/3 DRGs; Nrp1^∆SN n = 4 axon shafts/3 DRGs;
   L1cam^∆SN, n = 6 axon shafts/3 DRGs; Nrp1/L1cam^∆SN, n = 4 axon
   shafts/3 DRGs) treatment. All data are presented as mean ± SEM.
   *p < 0.05, **p < 0.01. Detailed statistics are provided in
   Supplementary Table [166]2. Source data are provided as a [167]Source
   Data file.

NRP1 and L1CAM are required for nmA separation and envelopment

   Previous studies have implicated the roles of L1CAM in PN axon
   fasciculation and RB assembly^[168]37,[169]59, as well as in mediating
   class-3 semaphorin signaling through the forming receptor complexes
   with NRP1/2^[170]55,[171]60, prompting us to ask whether SEMA3B could
   be the ‘missing’ nmSC-derived L1CAM signaling partner affecting nmA-nmA
   interactions. To determine the requirements of nmA-expressing L1CAM and
   NRP1 for RB assembly, we established
   Advillin^Cre;Nrp1^loxp/loxp;L1cam^loxp/loxp (Nrp1/L1cam^∆SN) mice, in
   which both Nrp1 and L1cam are selectively ablated in SNs^[172]61. The
   RBs of Nrp1/L1cam^∆SN mice exhibited incomplete nmA separation and
   envelopment, including nmSC processes projecting into the endoneurial
   space (Fig. [173]4k, l), closely resembling the RB defects observed in
   Sema3b^∆SC mice (see Fig. [174]2d). The RB assembly defects in
   Nrp1/L1cam^∆SN mice were significantly more pronounced than those in
   Nrp1^∆SN mice and somewhat stronger than those in L1cam^∆SN mice alone
   (Fig. [175]4h–l), likely due to the expression of additional,
   functionally redundant receptor components such as NRP2 and PLXNA4 by
   nmAs (Fig. [176]4b, c and Supplementary Fig. [177]4e)^[178]62,[179]63.
   However, simultaneous elimination of both Nrp1 and Nrp2 induces
   pronounced embryonic axon defasiculation^[180]49,[181]64, thus
   impairing the PN development as a whole and precluding the assessment
   of NRP2 inactivation as a probable nmA-derived SEMA3B receptor
   component in RB assembly. Therefore, the requirement for SC-derived
   SEMA3B in RB assembly is mirrored by the similar necessity for NRP1 and
   L1CAM, but we presently cannot rule out the involvement of other
   potential nmA-derived SEMA3B receptor components present at low levels
   in nmAs, including NRP2 and PLXNA4^[182]62,[183]63.

SEMA3B requires NRP1 and L1CAM to modulate axonal morphology

   To determine whether the requirements of L1CAM, NRP1, and SEMA3B for RB
   assembly indicate their involvement in common or distinct signaling
   pathways, we investigated acute sensory axonal responses to SEMA3B
   stimulation in vitro. Treatment of sensory axons extending from
   cultured DRG neurons with recombinant SEMA3B resulted in rapid
   retraction of axonal shaft filopodial branches and axonal growth cone
   collapse (Supplementary Fig. [184]4m–o and Supplementary Video [185]2).
   Meanwhile, multicolor super-resolution stimulated emission depletion
   (STED) microscopy revealed that SEMA3B-induced axonal responses were
   accompanied by reconfiguration of axonal distribution patterns of NRP1
   and L1CAM, including a significant increase in NRP1^+L1CAM^+ foci
   (Fig. [186]4m–o). To further elucidate the involvement of L1CAM and
   NRP1 in SEMA3B-induced axonal responses, we cultured SNs from Nrp1^∆SN,
   L1cam^∆SN, and Nrp1/L1cam^∆SN embryos and subsequently stimulated them
   with SEMA3B. The axonal responses, specifically axon shaft filopodial
   retraction and growth cone collapse induced by SEMA3B, were markedly
   reduced in Nrp1^∆SN or L1cam^∆SN axons and even further reduced in
   axons lacking both components (Fig. [187]4p–r and Supplementary
   Fig. [188]4o). While these findings are consistent with the role of
   NRP1 and L1CAM in mediating SC-derived SEMA3B effects on nmAs, the
   slightly additive effect of Nrp1^∆SN and L1cam^∆SN mutations could also
   indicate actions in parallel pathways or alternatively reflect the
   continued expression of potentially redundant signaling components in
   these mutants (see Figs. [189]1p, [190]4b and [191]5e).

Fig. 5. SC-derived SEMA3B promotes localized axon deadhesion and CAM
clearance.

   [192]Fig. 5
   [193]Open in a new tab

   a STED images showing TUJ1 signals on axon bundles treated with Fc and
   SEMA3B. Arrowheads indicate axonal beading. b Fasciculation index is
   calculated as the axonal area divided by the total area. c
   Quantification of the fasciculation indexes for axon bundles treated
   with Fc and SEMA3B. n = 6 axon shafts from 3 DRGs for both groups. d
   KEGG pathway analysis of up-regulated DEGs in nmSN. Pathway enrichment
   was analyzed by a two-tailed Wald test with Benjamini-Hochberg
   correction (FDR-adjusted p-value < 0.05 and |log₂ fold change|> 0.25).
   e Dot plot illustrating genes in the cell adhesion molecules pathway.
   f, g STED images of P7 Plp1^CreERT2;R26^tdTom RBs. Dotted lines
   demarcate RB boundaries. Asterisks denote axons ensheathed by TOM^+
   mSCs. The experiment was repeated 3 times on 3 mice. f’, g’ Magnified
   views of the boxed areas in (f) and (g). Dotted lines demarcate
   unmyelinated axon boundaries. Arrowheads indicate lamellipodia
   undergoing separating fasciculated axons. The asterisk in (g’) denotes
   an axon wrapped by SC lamellipodia, with strong CDH2 signals in the
   contact region. h STED images showing TUJ1 signals on axon bundles. i
   Quantification of the fasciculation indexes for axon bundles with
   SEMA3B or SEMA3B+Dynasore treatment. n = 6 axons bundles from 3 DRGs
   for both groups. j, k L1CAM and CDH2 particles on the membrane of
   sensory axon shafts from E14.5 DRG. l Quantification of artificial
   fluorescence intensity of L1CAM on the membrane of sensory axon shafts.
   n = 7 axon bundles from 3 DRGs for all groups. m Quantification of
   artificial fluorescence intensity of CDH2 on the membrane of sensory
   axon shafts. n = 6 axon bundles from 3 DRGs for all groups. n–p STED
   images of the RBs from P7 control, Sema3b^tgSC, and Sema3b^∆SC mice.
   Dotted lines delineate the assumed RB boundaries. q Quantification of
   the numbers of L1CAM particles per RB. n = 6 RBs from 3 mice for all
   groups. r Schematic summary: nmSC-derived SEMA3B induces the
   endocytosis of axonal CAMs, leading to axonal segregation. All data are
   presented as mean ± SEM. **p < 0.01, *p < 0.05. Detailed statistics are
   provided in Supplementary Table [194]2. Source data are provided as a
   [195]Source Data file.

SEMA3B promotes localized axon deadhesion and CAM clearance

   Given that secreted SEMA3s are known to influence axon-axon
   interactions in embryonic PNs and other contexts^[196]48,[197]65, we
   reasoned that SEMA3B might indirectly facilitate nmA-nmSC association
   by modulating nmA-nmA adhesion rather than directly mediating nmA-nmSC
   association during RB assembly. To test this hypothesis, we cultured
   dissociated SNs from E14.5 DRGs for three days to promote axon
   bundling. SN axons treated with SEMA3B, but not with the control
   protein, exhibited a modified morphology characterized by bulges
   separated by thinner segments, a phenomenon termed axonal
   beading^[198]66, ultimately leading to the localized segregation of
   bundled axons at beading sites (Fig. [199]5a–c), suggesting that SEMA3B
   promotes localized loosening of axon-axon adhesion. To identify
   potential mediators of nmA-nmA adhesion regulated by SEMA3B, we
   analyzed snRNA-seq data of SNs, and found high expression levels of
   genes encoding several CAMs, including L1cam and Cdh2 (also known as
   N-Cadherin), previously shown to be associated with nmAs during RB
   assembly, in Nefh^- nmSNs (Fig. [200]5d, e and Supplementary
   Fig. [201]4i–k, [202]5a-c)^[203]67,[204]68. Additionally, both L1CAM
   and CDH2 exerted significant attractive effects on axon filopodial
   branches (Supplementary Fig. [205]5d, e), suggesting that axonal CAMs
   may mediate axonal adhesion, thus potentially serving as mediators of
   nmA-nmA adhesion regulated by SEMA3B. Supporting our hypothesis, STED
   imaging of sciatic nerves from Plp1^CreERT2;R26^tdTom mice, which
   allowed simultaneous visualization of nmSC processes and axons in
   immature RBs, revealed high expression levels of L1CAM and CDH2 in the
   contact regions among nmA fascicles (Fig. [206]5f, g). Notably, L1CAM
   and CDH2 signals in the axonal region proximal to nmSC lamellipodia
   were relatively weaker (Fig. [207]5f, g), suggesting that nmSC
   lamellipodia may diminish the expression levels of axonal CAMs, thereby
   disrupting CAM-induced heterophilic and/or homophilic interactions
   between nmAs.

   Given that the impact of SEMA3s on axon growth involves
   endocytosis^[208]69,[209]70, and recruitment of axonal L1CAM promotes
   receptor endocytosis induced by SEMA3A^[210]54,[211]71, we subsequently
   investigated whether SEMA3B-mediated axon deadhesion similarly involved
   the endocytic clearance of CAMs. Activation of the endocytosis pathway
   was observed in Nefh^− nmSNs during RB assembly (Fig. [212]5d).
   Pre-treatment of SN axons with Dynasore, an inhibitor of
   Dynamin-mediated endocytosis^[213]72, significantly reduced
   SEMA3B-promoted localized beading and deadhesion of cultured SN axons
   (Fig. [214]5h, i). To assess whether SEMA3B could induce endocytosis of
   axonal CAMs, we performed L1CAM and CDH2 immunostaining on sensory
   axons at 4 °C to halt endocytosis^[215]73, thus allowing visualization
   of signals expressed on the membrane. Indeed, Dynasore pretreatment
   significantly lowered the SEMA3B-mediated reduction of L1CAM and CDH2
   cell surface expression levels in cultured SN axons (Fig. [216]5j–m).
   Furthermore, Dynasore pre-treatment significantly reduced L1CAM
   co-localization with the early endosome marker EEA1 in cultured SN
   axons (Supplementary Fig. [217]5f, g). Thus, SEMA3B may mediate
   localized loosening of axon-axon adhesion by triggering the
   endocytosis-dependent clearance of axonal CAMs. To establish the in
   vivo relevance of these actions, we altered SC-derived SEMA3B levels
   and examined corresponding changes in CAM levels on nmAs in developing
   RBs. Elevating SEMA3B levels in nmSCs of Sema3b^tgSC mice resulted in a
   small but significant decrease in L1CAM and CDH2 levels in nmAs in
   developing RBs. Conversely, reducing SEMA3B levels in nmSCs of
   Sema3b^∆SC mice resulted in a small but significant increase in L1CAM
   and CDH2 levels in nmAs (Fig. [218]5n–q and Supplementary
   Fig. [219]5h–k). These results are consistent with the role of SEMA3B
   in the endocytic clearance of nmA-derived CAMs.

   In summary, these data suggest that SEMA3B operates by loosening
   axon-axon interactions in preformed nmA fascicles through localized
   endocytic clearance of CAMs (Fig. [220]5r).

SEMA3B-mediated deadhesion is transient

   Following the shift from axon-axon to axon-glia interactions, RB
   assembly requires the formation of stable contacts between nmAs and
   nmSCs to maintain nmA integrity. Consistent with previous observations
   demonstrating the role of CAMs in mediating nmA-nmSC interactions
   during RB assembly^[221]68,[222]74–[223]76, our analysis of scRNA-seq
   data and in situ hybridization revealed high expression levels of
   several CAMs, including L1cam and Cdh2, in pnmSCs and nmSCs, and their
   expression levels notably increased starting from P14, correlating with
   the onset of nmSC wrapping of axons within the RB (Fig. [224]6a, b and
   Supplementary Fig. [225]6a–d). Concurrently, a progressive decline of
   Sema3b expression was observed in SCs during the later stages of RB
   assembly (Fig. [226]2a and Supplementary Fig. [227]2a), suggesting the
   recruitment of axonal CAMs to the axonal membrane, thereby potentially
   facilitating interactions with nmSC-expressing CAMs to regulate their
   reciprocal attraction. Supporting this speculation, STED imaging of the
   mature RBs from Plp1^CreERT2;R26^tdTom mice confirmed the high
   expression levels of axonal L1CAM and CDH2 in the contact area between
   nmSC lamellipodia and axon (Fig. [228]6c–e and Supplementary
   Fig. [229]6e–g). We next asked how the eventual establishment of stable
   nmA-nmSC contacts during RB assembly could be consolidated with the
   continued presence of SC-derived SEMA3B, which might potentially
   promote CAM clearance. We cultured primary SCs or SNs on surfaces
   coated with L1CAM or CDH2, to which both cell types are stably attached
   (Fig. [230]6f–i and Supplementary Fig. [231]6h–o). Acute (1-hour)
   treatment of these cultures with SEMA3B promoted the detachment of SCs
   and SNs from both the L1CAM- and CDH2-coated surfaces. However, chronic
   (24-hour) treatment with SEMA3B led to the reattachment of SCs and SNs
   to L1CAM and CDH2-coated surfaces (Fig. [232]6j–n and Supplementary
   Fig. [233]6p, q), suggesting that the effects of SEMA3B-induced
   deadhesion are transient and are followed by a period of
   reestablishment of adhesions.

Fig. 6. SEMA3B transiently interrupts axonal adhesion to facilitate nmSC-nmA
adhesion.

   [234]Fig. 6
   [235]Open in a new tab

   a UMAP plot illustrating L1cam expression in iSCs and nmSCs. b FISH on
   transverse section of P28 sciatic nerve. Arrowheads indicate
   L1cam^+Scn7a^+SOX10^+ nmSCs and arrows indicate L1cam^-Scn7a^-SOX10^+
   mSCs. The experiment was repeated 3 times on 3 mice. c STED image of
   the RB from P28 Plp1^CreERT2;R26^tdTOM sciatic nerves. The experiment
   was repeated 3 times on 3 mice. c’ Magnified view of the boxed area in
   (c) illustrating a high L1CAM expression level in the contacting area
   between TUJ1^+ axons and TOM^+ SC lamellipodia. Note the presence of
   L1CAM signals in TOM^+ SC lamellipodia. d Drawing a line to calculate
   the fluorescence intensity across the section. e Quantification of the
   fluorescence intensities of TOM, TUJ1, and L1CAM signals. f
   Representative E14.5 DRG SNs cultured on Fc- and L1CAM-coated
   coverslips. g Quantification of the numbers of attached sensory
   neurons. n = 6 independent experiments for both group. h SCs purified
   from P0 sciatic nerves cultured on Fc- or L1CAM-coated coverslips. i
   Quantification of the numbers of attached SCs. n = 6 independent
   experiments for both groups. j Assay for mimicking the acute and
   chronic effect of SEMA3B on attached SNs and SCs on L1CAM-coated
   coverslip. k Images showing attached SNs under different conditions. l
   Quantification of the numbers of attached SNs under different
   conditions. n = 6 independent experiments for all groups. m Images
   showing attached SCs under different conditions. n Quantification of
   the numbers of attached SCs under different conditions. n = 6
   independent experiments for all groups. o Schematic representation of
   the three biological events involved in RB assembly: nmA-nmA adhesion,
   nmA-nmA deadhesion, and nmA-nmSC adhesion. The corresponding molecular
   mechanisms are depicted. nmSC-derived SEMA3B interacts with axonal
   NRP1/L1CAM receptor complex to shift CAM-induced nmA-nmA adhesion to
   CAM-induced nmA-nmSC adhesion. The lower panel presents STED images
   illustrating the dynamic distribution patterns of CAM during the three
   events of RB assembly. All data are presented as mean ± SEM.
   **p < 0.01, *p < 0.05. Detailed statistics are provided in
   Supplementary Table [236]2. Source data are provided as a [237]Source
   Data file.

   Collectively, these data support the notion that SEMA3B promotes
   nmA-nmSC interactions and RB assembly by transiently disrupting nmA-nmA
   interactions to facilitate subsequent nmA-nmSC interaction
   (Fig. [238]6o).

SEMA3B accelerates RB reassembly and alleviates pain following PN injury

   SC dedifferentiation is an inherent part of the regeneration response
   following PN injury^[239]77,[240]78, which is also accompanied by RB
   disassembly, nmA dysfunction, and aberrant pain
   perception^[241]8,[242]79. Following partial sciatic nerve ligation
   (PSNL) in mice, RB disassembly is reflected by the (re)appearance of
   fascicles of small-caliber axons, in which most axons are not or only
   partially enveloped by nmSC processes (Fig. [243]7d, i). Consistent
   with injury-induced SC dedifferentiation, PN injury was followed by
   prolonged downregulation of nmSC markers, including Scn7a, L1cam, and
   Sema3b, in the SCs as revealed by the analysis of previously
   established transcriptome profiling data (Fig. [244]7a–c and
   Supplementary Fig. [245]7a, b)^[246]80. FISH on sciatic nerves
   confirmed the downregulation of Sema3b persisting 7 days post-ligation
   (dpl), which slightly recovered at 21 dpl but remained notably lower
   compared to control mice (Fig. [247]7e, g). Additionally, analysis of
   previous transcriptome profiling data from DRGs of injured mice
   revealed sustained expression of CAMs, including L1cam and Cdh2, in
   nmSNs (Supplementary Fig. [248]7c–e and Supplementary
   Table [249]1)^[250]81, suggesting the reformation of nmA fascicles
   through CAM-mediated axon-axon interactions after PN injury.

Fig. 7. Elevating SEMA3B expression level in SCs alleviates PN injury-induced
pain.

   [251]Fig. 7
   [252]Open in a new tab

   a UMAP visualization of SCs from scRNA-seq data in sham mice and
   nerve-injury mice. b Relative proportions of SC lineages in total SCs
   of sham mice and nerve-injury mice. c Dot plot showing expression
   levels of Scn7a, Sema3b, and L1cam in SCs of sham mice and nerve-injury
   mice. d Cartoon depicting that the distal part of the injury site is
   harvested for analysis. e Sema3b expression levels in SOX10^+ SC
   following different days post ligation (dpl) of control mice. f Sema3b
   expression levels in SOX10^+ SCs following different days post-ligation
   (dpl) of Sema3b^tgSC mice. Sema3b transcript levels demonstrated
   relatively stable after nerve injury. g Quantification using qPCR
   showing Sema3b transcript levels in sciatic nerves without injury
   (sham) and with 7 and 21 days post ligation (dpl). n = 6 mice for all
   groups. h–j Representative TEM images of the sciatic nerves from
   control and Sema3b^tgSC mice. Note that the contact area between nmSC
   lamellipodia and axons is significantly decreased in the RB of control
   mice after injury. In contrast, the deficient RB assembly is noticeably
   reduced in Sema3b^tgSC mice with injury. Asterisks and circles indicate
   axons that are adjacent and partially wrapped, respectively. Arrowheads
   indicate lamellipodia protruding into the endoneurial space. k
   Quantification of RBM indexes of control and Sema3b^tgSC mice without
   and with nerve crush. n = 6 mice for all groups. l Measurements of
   sensitivity to von Frey filaments in control and Sema3b^tgSC mice
   without and with nerve injury. n = 6 mice for all groups. m
   Measurements of sensitivity to hot laser in control and Sema3b^tgSC
   mice without and with nerve injury. n = 6 mice for all groups. All data
   are presented as mean ± SEM. **p < 0.01, *p < 0.05. Detailed statistics
   are provided in Supplementary Table [253]2. Source data are provided as
   a [254]Source Data file.

   We next asked whether augmenting the slow post-PSNL recovery of SEMA3B
   expression in re-differentiating SCs could accelerate RB reassembly and
   aid the restoration of nmA integrity. To test this idea, we performed
   PSNL on Sema3b^tgSC mice. At 21 dpl, SCs along the sciatic nerve of
   Sema3b^tgSC mice showed significantly increased Sema3b expression
   (Fig. [255]7e–g). In Sema3b^tgSC mice following PSNL, TEM analysis of
   the sciatic nerves revealed a significant decrease in the number of
   small-caliber axons not enveloped by nmSCs and an apparent acceleration
   of RB reassembly (Fig. [256]7j, k). Moreover, Sema3b^tgSC exhibited
   significantly reduced allodynia and hyperalgesia following PSNL
   compared to control mice (Fig. [257]7l, m). Thus, restoring SC-derived
   SEMA3B following PN injury can accelerate RB reassembly and alleviate
   pain, presumably by facilitating the re-envelopment of nmAs by nmSCs in
   post-injury nmA fascicles.

Discussion

   The intimate association with glia has long been established to provide
   essential metabolic and functional support for axons, but glia-derived
   signals are also emerging as key mediators of axonal maturation during
   neuronal development^[258]43. In the present study, we attempted to
   shed light on the steps preceding the establishment of axon-glia
   contacts, which so far had received comparatively little attention. We
   provided evidence that CAM turnover elicited by a glia-derived factor,
   SEMA3B, facilitates a switch from axon-axon to axon-glia interactions
   during PN development (Fig. [259]6o), which we further exploited to
   accelerate the restoration of axon-glia contacts following nerve
   injury, thus alleviating axonal dysfunction and neuropathic-like pain.

Transitioning from axon-axon to axon-glia interactions

   The axon-axon interactions dominating the early stages of neural
   development underlie axon fasciculation, segregation, navigation, and
   pre-target sorting, and are mediated by CAMs as well as other protein
   families, including ‘classical’ axon guidance
   molecules^[260]18,[261]36. During PN development, most axons, including
   nmAs, remain fasciculated before radial sorting and the eventual
   transition to axon-glia interactions. Several CAMs have been implicated
   in axon fasciculation, as well as in the initial establishment of
   SC-axon contacts^[262]82. Specifically, interactions involving L1CAM
   have been implicated to mediate both axon
   fasciculation^[263]37,[264]83,[265]84, and axon-glia contacts during
   early radial sorting and RB assembly^[266]59,[267]85, raising the
   question how axon-axon interactions effectively become swapped for
   axon-glia interactions when both cell types express the same CAM. It
   has been proposed that axonal L1CAM may engage in a heterophilic manner
   with NRP1 to mediate SEMA3A-induced axon response^[268]54,[269]71, but
   its binding partners on SCs have not been identified^[270]75. Our
   findings indicate that one of such interaction partners is SEMA3B (via
   axonal NRP1), but this alone is insufficient to explain the eventual
   requirements of L1CAM for establishing axon-glia adhesion during RB
   assembly. Additionally, our results indicate that L1CAM inactivation in
   SCs at least partially phenocopies the RB assembly defects observed
   upon axonal L1CAM removal (see Supplementary Fig. [271]6r, s),
   suggesting a significant contribution of homophilic L1CAM interactions
   to the eventual formation of nmSC-nmA adhesion, consistent with the
   high levels of L1CAM localizing to axon-glia contacts in maturing RBs
   (Fig. [272]6c). While not ruling out some contributions of heterophilic
   CAM interactions between SCs and axons, such a model would necessarily
   imply mechanism/s facilitating the switch from mediating axon-axon to
   mediating axon-glia interactions by the same CAMs.

Glia-derived factors and clearance of axonal CAMs

   Our data suggest that SC-derived SEMA3B engages axonal NRP1 and
   (directly or indirectly) L1CAM to loosen axon-axon adhesion temporally,
   thus facilitating nmA-nmSC interactions and RB assembly. Previous
   studies have illustrated that L1CAM directly interacts with NRP1, but
   not NRP2, to form a receptor complex^[273]54,[274]71. This complex is
   endocytosed and activates a signaling cascade involving focal adhesion
   kinase (FAK) and mitogen-activated protein kinase (MAPK), leading to
   disassembly of adhesion points, resulting in growth cone collapse and
   axon retraction^[275]86, which provides a plausible route through which
   SEMA3B-NRP1/L1CAM could trigger endocytosis not only of L1CAM but other
   nmA-resident CAMs, including CDH2, to disrupt axon-axon adhesions.
   Supporting this hypothesis, our scRNA-seq analysis revealed MAPK
   pathway activation in unmyelinated sensory neurons coinciding with RB
   assembly (see Fig. [276]5d), implicating this signaling axis in the
   spatiotemporal regulation of axon-glia interactions. In addition to
   SEMA3B, other semaphorin family members are expressed in nmSCs (see
   Fig. [277]1p and Supplementary Fig. [278]1k), which may hint at an even
   more profound involvement of SEMA signaling in RB assembly. Similarly,
   NRP2 and some PLXNA members were significantly expressed in CGRP^+ nmAs
   (see Fig. [279]4b, c and Supplementary Fig. [280]4e). However, its
   potentially overlapping role with NRP1 remains masked by the PN
   defasciculation in Nrp1/2 double-deficient mice. Adding to the
   complexity, other CAMs such as NrCAM, which are highly expressed in
   nmAs during RB assembly (see Supplementary Figs. [281]5b, [282]6a), can
   associate with NRP1/2 to transmit SEMA3B- and SEMA3F-induced
   signaling^[283]87,[284]88. The combination of these interactions may
   account for the relatively mild deficiencies observed in RB assembly in
   Nrp1 and L1cam mutant mice.

   Recent evidence has shown that axonal CAMs, including L1CAM and CDH2,
   are spatially organized within the periodic membrane skeleton (PMS),
   which is crucial for establishing and maintaining axon-axon
   interactions and fasciculation^[285]89. Therefore, elucidating how
   SEMA3B impacts PMS organization, potentially reflected by axonal
   beading induced by SEMA3B, will be highly informative. The restriction
   of SEMA3B to nmSCs raises the question of whether mSCs would similarly
   release factors that prime fasciculated mAs for the transition to
   mA-mSC interactions. Indeed, immature mSC-derived factors, which remain
   to be identified, have recently been shown to promote widespread
   clathrin-dependent endocytic clearance of CAMs from mA internodal
   segments prior to myelination^[286]35, while mSC-derived prostaglandin
   2 has been separately shown to organize the distribution of
   voltage-gated sodium channels along mAs^[287]90. Whether the same or
   similar mSC-derived factors also play a role in redistributing axonal
   membrane proteins during radial sorting, thus priming fasciculated mAs
   for engagement with mSCs, remains to be addressed.

   Currently, we cannot strictly rule out that other actions by SC-derived
   SEMA3B, in addition to promoting axonal CAM turnover, could be involved
   in RB assembly. Notably, secreted semaphorins have been implicated in
   either counteracting or promoting cell-cell adhesions depending on the
   specific cellular context, suggesting a potential supplementary role
   for SEMA3B signaling in stabilizing axon-glia contacts. However,
   disentangling these contributions from its temporary effects on initial
   axon-axon interactions remains challenging, particularly given our
   current inability to directly observe the glia-axon interactions
   involved in RB assembly in situ. Moreover, although our findings
   indicate that SEMA3B signaling partially relies on an axonal receptor
   complex incorporating L1CAM, consistent with observations in other
   contexts, it is also plausible that L1CAM plays an indirect role in
   transmitting or modulating SEMA3B signaling. Finally, it will be
   important to test the ability of SEMA3B to restore nmA integrity when
   applied after PN injury, which would more closely resemble ‘clinical
   reality’. In any case, the effectiveness of restoring SC-derived SEMA3B
   will likely be restricted to early intervention following PN injury,
   before the onset of the complex, long-term mechanisms that drive
   chronic (neuropathic) pain.

Methods

Mouse lines

   All experimental procedures were conducted according to protocols
   approved by the Animal Care and Use Committee of Zhejiang University
   (Protocol AIRB-2021-970). Experiments were performed using age- and
   sex-matched animals. Animals were group-housed in facilities with
   controlled temperature and humidity and maintained on a 12-hour
   light-dark cycle. All transgenic mice used in these studies were of the
   C57BL/6 J strain. All the mouse lines used in our study were maintained
   on a C57BL/6 genetic background. The morning a vaginal plug was
   detected was designated as embryonic day (E) 0.5. Detailed information
   on mouse lines and primer sequences for genotyping is available in
   Supplementary Table [288]2. Sema3b^loxp/loxp were obtained from Dr.
   Allan Bradley (Wellcome Trust Sanger Institute)^[289]91. The following
   mouse lines were used in our study: Dhh^cre (Jax stock# 012929);
   L1cam^loxp/loxp (GemPharmatech stock# T017494); vGlut2^Cre (Jax stock#
   016963); Plp1^CreERT2 (Jax stock# 005975); R26^tdTom(Jax stock#
   007909); Sox10^Cre(SMOC: NM-KI-220066); Chat^cre (Jax stock# 006410);
   Nrp1^loxp/loxp (Jax stock# 005247). Mice were anesthetized via
   inhalation of 3-4% isoflurane (flow rate: 1 L/min O₂) until loss of
   toe-pinch reflex. Euthanasia was performed by cervical dislocation
   under deep anesthesia, followed by immediate tissue dissection.

   The Sema3b^tgSC transgenic mice were developed by Shanghai Model
   Organisms Center, Inc. (Shanghai, China). Sema3b^tgSC mice were
   generated with CRISPR/Cas9-mediated homologous recombination with the
   guide RNA 5’-GGGGACACACTAAGGGAGCTTGG-3’ targeting the Rosa26 gene. A
   freshly prepared microinjection mixture composed of Cas9 mRNA, guide
   RNA, donor vector, and targeting vector, diluted in microinjection
   buffer, was injected into the cytoplasm and pronucleus of zygotes,
   which were subsequently implanted into pseudopregnant C57BL/6 female
   mice to yield live offspring. Founders were identified through PCR
   genotyping utilizing the primers specified in Supplementary
   Table [290]2. Hemizygous founders were then bred with Dhh^Cre mice to
   generate Dhh^Cre;Sema3b^tg mice.

Tamoxifen and EdU injections

   For experiments involving inducible CreERT mouse strains, tamoxifen
   (T5648; Sigma-Aldrich) was dissolved in a mixture of ethanol and corn
   oil (1:9 by volume) to a concentration of 50 mg/ml and administered via
   intraperitoneal injection at a dose of 50 μg TM/g body weight. For cell
   cycle study, mice were injected intraperitoneally with EdU (100 mg/kg,
   Invitrogen). The injected mice were euthanized 2 h after injection for
   analysis. EdU incorporation was subsequently detected using Alexa Fluor
   488 azide following the manufacturer’s instructions (Invitrogen).

Single-cell RNA sequencing

Single-cell preparation for single-cell analysis

   Mice were anesthetized via inhalation of 3–4% isoflurane, and
   euthanasia was performed by cervical dislocation under deep anesthesia.
   Sciatic nerves from mice at various developmental stages (P0, P7, P14,
   P28, and P60) were pooled for each experimental analysis as follows: 12
   males and 10 females for P0, 10 males and 11 females for P7, 10 males
   and 8 females for P14, 9 males and 6 females for P28, and 5 males and 8
   females for P60, to facilitate scRNA-seq library preparation. DRGs at
   the lumbar level of P14 mice were meticulously dissected from 4 males
   and 4 females to facilitate snRNA-seq library preparation. The
   harvested neural tissues were placed in a culture dish containing an
   appropriate volume of calcium-free and magnesium-free PBS on ice,
   sectioned into 0.5 mm² pieces, and subsequently dissociated into single
   cells using a dissociation medium containing 0.35% collagenase IV,
   2 mg/ml papain, and 120 Units/ml DNase I, maintained in a 37 °C water
   bath. The enzymatic digestion was terminated using 1×PBS with 10% fetal
   bovine serum. The resulting cell suspension was passed through a 70-30
   μm cell strainer and centrifuged at 300 × g for 5 min at 4 °C. The cell
   pellet was then resuspended in 100 μl of 1× PBS (0.04% BSA) and
   supplemented with 1 ml of 10× red blood cell lysis buffer, allowing it
   to incubate at RT for 2–10 min. Subsequent centrifugation at 300 × g
   for 5 min at room temperature was performed. The cells were resuspended
   in 100 μl of a Dead Cell Removal MicroBeads solution, and non-viable
   cells were eliminated using the Miltenyi® Dead Cell Removal Kit. The
   final cell suspension was prepared in 50 μl of 1× PBS (0.04% BSA).
   Overall cell viability was verified via trypan blue exclusion, yielding
   a viability rate exceeding 85%. Single-cell suspensions were quantified
   using a hemocytometer or a Countess II Automated Cell Counter, with the
   cell concentration adjusted to 700-1200 cells/μl.

Chromium 10x Genomics library and sequencing

   Single-cell suspensions were introduced into the 10x Chromium system to
   isolate 5000 individual cells following the manufacturer’s guidelines
   for the 10x Genomics Chromium Single-Cell 3’ kit (V3). The ensuing
   steps of cDNA amplification and library construction adhered rigorously
   to the standard protocol. Sequencing of the libraries was performed
   using an Illumina NovaSeq 6000 system (paired-end multiplexing run,
   150 bp) by LC-Bio Technology Co., Ltd (Hangzhou, China), achieving a
   minimum sequencing depth of 20,000 reads per cell.

scRNA-seq data processing and quality control

   Transcripts were aligned to the mm10 reference genome utilizing
   CellRanger Version 3.1.0 from 10x Genomics. The “mkfastq” function
   within CellRanger was employed for the purposes of demultiplexing and
   generating FASTQ files. Subsequently, single-cell gene counts were
   obtained using the “count” function with the parameters “–id,
   –transcriptome, –fastqs, and –sample”. Following an initial assessment
   of metrics provided by CellRanger, cells of low quality were excluded
   to mitigate cell-specific biases. Additionally, nuclei identified as
   potential doublets—specifically, cells expressing more than 8000 genes
   per cell—were excluded from further analysis. The filtered cells and
   counts data from Cell Ranger were then analyzed using Seurat version
   3.1.2, implemented in R version 3.6.2. Quality control metrics were
   stringent, including cells with fewer than 8000 UMIs per cell, more
   than 2000 UMIs per cell, fewer than 2500 genes per cell, and no more
   than 5% mitochondrial gene expression. Mitochondrial genes and genes
   expressed in fewer than three cells were removed from the dataset.
   After completing the quality control procedures, 11,819 cells remained
   available for subsequent bioinformatics analyses.

Multiple sample integration with reciprocal principal component analysis

   To overcome batch effects related to freshly dissociated samples and
   nuclei isolated from fresh frozen samples, including higher
   mitochondrial and ribosomal transcripts in the fresh samples and more
   intronic and long non-coding reads in the frozen nuclei, Seurat’s
   reciprocal principal component analysis (RPCA) was used to integrate
   the scRNA-seq datasets. Briefly, a SeuratObject was created for each
   sample. Each sample was subsequently normalized with Seurat’s
   “NormalizeData” function. Variable features, totaling 3000 for each
   sample, were identified using “FindVariableFeatures”. To facilitate
   integration, features were selected through the
   “SelectIntegrationFeatures” function, with nfeatures set to 3000. The
   RPCA integration was then executed by sample using the
   “FindIntegrationAnchors” function in Seurat. Following this, the
   “IntegrateData” function was used to integrate the data, with k-nearest
   neighbors (k.weight) configured to 50. The integrated values were
   returned for all genes within the SeuratObject

Downstream analyses of scRNA-seq data

   The integrated SeuratObject underwent further scaling utilizing the
   “ScaleData” function. Subsequently, graph-based clustering was executed
   on the integrated object with a resolution parameter set at 0.5.
   Differential expression analysis of genes was conducted for the
   clusters of the “integrated Assay” on the “RNA Assay” using the
   “FindAllMarkers” function, specifically targeting upregulated genes
   (only.pos = T). For subsequent analyses, only differentially expressed
   genes (DEGs) with an adjusted p value (p.adj) ≤ 0.05 were considered
   significant. Highly variable genes were identified via the
   “MeanVarPlot” function using default settings. Principal component
   analysis (PCA) was performed on the top 3000 variable genes,
   incorporating the top 50 principal components into downstream analyses.
   Additionally, the top 20 principal components were derived using the
   “RunPCA” function with default parameters. A K-nearest neighbor graph
   was generated based on Euclidean distances. Subsequent to performing
   PCA, the first 20 dimensions were utilized to compute a (UMAP) with an
   initial clustering resolution of 0.5 for all nuclei and sub-clustering
   using the “RunUMAP” function in Seurat. Utilizing the “FindNeighbors”
   and “FindClusters” functions in Seurat, cells were clustered via a
   graph-based methodology employing the Louvain algorithm, with the
   resolution parameter set at 0.03 for cell grouping. To identify
   differentially expressed genes in each cluster, Wilcoxon rank sum tests
   were conducted using the “FindAllMarkers” function in Seurat. In this
   analysis, gene expression is defined as raw counts normalized to 10,000
   transcripts per cell, with logarithmic expression represented as
   log2-transformed expression values. Annotation of the clusters was
   performed manually employing the “SingleR” package based on marker gene
   expression, classifying clusters into various cell types: fibroblast
   (Pdgfra, Mfap5), macrophage (Aif1, Itgam), monocyte (Cd3g, Il7r),
   smooth muscle cell (Acta2, Mylk), Schwann cell (Sox10, Dhh), and
   endothelial (Vwf, Pecam1).

   Post-clustering cells that met the quality control benchmarks detailed
   above were analyzed further. Nuclei likely representing doublets
   between glial and non-glial cells were excluded from downstream
   analysis. Specifically, cells exhibiting marker gene expression (>0.5
   standard deviations from the mean) from multiple cell types were
   classified as doublets and subsequently excluded. The marker genes
   employed for doublet identification included: smooth muscle
   cell = Acta2, endothelial = Pecam1, macrophages = Aif1, Schwann cell = 
   Sox10, monocyte = Cd3g, and fibroblasts = Pdgfra. Clusters derived from
   DRGs expressing the marker gene tubb3 were classified as neurons, while
   clusters showing enrichment for non-neuronal marker genes such as
   Pdgfra, Pecam1, Arhgap15, or Sox10 were designated as non-neuronal
   clusters.

Analysis of sub-clusters

   For specific cell types or classes of interest, their transcriptomic
   data were extracted for focused subsequent analysis. For instance, SCPs
   and their derivatives were isolated using the “SubsetData” function
   following initial annotation. Subsequent clustering analysis was
   performed using the “FindClusters” and “FindAllMarker” functions, and
   UMAP was employed to re-cluster the SCPs and their derivatives.
   Sub-clusters were annotated based on the expression of marker genes,
   including SCPs (Ednrb), iSCs (Sostdc1, Mant2), pnmSCs (Sox2), nmSCs
   (Scn7a), pmSCs (Pou3f1), and mSCs (Ncmap, Prx). The criteria for
   identifying marker genes for each cluster included an adjusted P value
   of less than 0.01 and a fold change greater than 0.5.

Trajectory Inference Using scFates in Single-Cell Transcriptomics

   To delineate the differentiation trajectory of Schwann cell precursors
   (SCPs), we applied scFates, a computational framework for branching
   trajectory inference in single-cell transcriptomic data^[291]92. The
   workflow began with data preprocessing involving gene filtering using
   scanpy and scFates, where over dispersed genes were identified via the
   Pagoda2 method followed by PCA-based dimensionality reduction. Cells
   were then mapped to a diffusion space, and a tree-like structure
   comprising 3,000 nodes was reconstructed using the SimplePPT algorithm
   (σ = 0.0005, λ = 200). Root nodes were selected based on CytoTRACE
   differentiation potential scores, SCPs were identified as the initial
   differentiation nodes. Branch-specific gene dynamics were analyzed
   through generalized additive models (GAMs) incorporating cubic spline
   regression, with interaction terms identifying genes exhibiting
   divergent expression patterns across trajectories (FDR < 0.0001).
   Differentially expressed genes between branches were further validated
   using linear regression models (FDR < 0.05). Cell states along
   differentiation paths were partitioned via Leiden clustering on UMAP
   embeddings, with Wilcoxon rank-sum tests identifying state-specific
   marker genes. This approach enabled high-resolution reconstruction of
   SCPs differentiation trajectories while accounting for multi-branching
   events, as demonstrated in neural crest lineage studies. Methods were
   implemented in Python using scFates v1.0.0 with default parameters
   unless specified. Statistical significance thresholds followed
   Benjamini-Hochberg correction.

Comparison of sham SC sample and post-injury SC samples

   The scRNA-seq datasets ([292]GSE198582) were retrieved utilizing the
   “getGEO” function from the GEOquery package (v2.58.0). To extract
   contrasts between “sham SC” and “post-injury SC” samples, the
   “makeContrasts” function from the Limma package was employed.
   Differential expression analysis between “sham SC” and “post-injury SC”
   samples was conducted using empirical Bayes statistics, facilitated by
   the “ebayes” function from the Limma package. The resulting moderated
   t-statistics were categorized into “upregulated,” “downregulated,” or
   “no change” groups using the “decideTests” function from Limma. The top
   50 differentially expressed single-cell markers from various cells were
   visualized using a scaled matrix. ComplexHeatmap (v2.11.1) was utilized
   to annotate differential expressions and clusters.

Cell cycle discrimination analyses

   We employed cell cycle-related genes, including a previously
   established core set comprising 43 G1/S phase genes and 54 G2/M phase
   genes. For each cell, a specific cell cycle phase (G1, S, G2/M) was
   assigned based on the expression patterns of the G1/S or G2/M phase
   genes, utilizing the scoring methodology implemented in the
   “CellCycleScoring” function within the Seurat package. Additionally,
   the quantification of cells within each cell cycle state was performed
   using the ‘WhichCells’ function in Seurat

Pseudotime and lineage inference

   Pseudotime and lineage analysis were executed utilizing Monocle3 and
   SoptSC, respectively. In brief, pseudotime was determined by
   calculating the shortest path distance between individual cells and a
   designated root cell on a cell-to-cell graph, which was constructed
   based on the similarity matrix. The root cell was identified by the
   user within the Monocle3 framework. The visualization of cell
   trajectories was achieved using UMAP. Cell states were represented
   through abstract lineage trees, which were derived by computing the
   minimum spanning tree of the cluster-to-cluster graph predicated on the
   shortest path distance between cells. Pseudotime was subsequently
   projected onto the lineage tree in a manner that defined the order of
   each state (or cluster) based on the average distances between cells
   within that state and the root cell. For Diffusion Pseudotime (DPT),
   the root cell was chosen from the SCP cluster.

Developing Schwann cell trajectory

   To explore the potential relationships among various SC subpopulations,
   we utilized pseudotime trajectory analysis with the “Monocle2” package.
   Initially, the prerequisite dataset was constructed from a Seurat
   object comprising all Schwann cells via the “importCDS” function in the
   “Monocle3” package. The gene expression matrix was imported into
   Monocle, followed by dimensionality reduction through parameter
   adjustment. Then, established marker genes were selected to perform
   pseudotime trajectory analysis using the “reduceDimension” function
   along with the “DDRTree” method. The outcomes of the analysis were
   visualized by plotting cellular trajectories via the
   “plot_cell_trajectory” function, enabling the construction of a cell
   differentiation trajectory. The cell trajectories, illustrated as
   tree-like structures with tips and branches, depicted different
   differentiation states, samples, and cell subsets.

KEGG analysis

   To explore the potential candidates expressed by SCs during RB
   assembly, Gene Ontology (GO) enrichment analysis and Kyoto Encyclopedia
   of Genes and Genomes (KEGG) pathway enrichment analysis were conducted
   using the R package ClusterProfiler. Only terms with a false discovery
   rate (FDR) of less than 0.05 were considered statistically significant.
   The top 26 pathways were selected based on their adjusted p-value
   rankings.

Immunostaining

   Whole-mount staining was performed as described previously^[293]93.
   Briefly, mice were anesthetized via inhalation of 3-4% isoflurane until
   loss of toe-pinch reflex. Euthanasia was performed by cervical
   dislocation under deep anesthesia. Tissues were isolated and fixed in
   4% paraformaldehyde/ PBS for 6–24 h according to tissue sizes at RT,
   followed by further fixation with Dent’s solution (methanol:dimethyl
   sulfoxide 4:1) for 3–24 h at RT. Primary antibodies and secondary
   antibodies were applied in 1% BSA/PBS-Triton solution (0.5% Triton
   X-100; 9002-93-1; Sigma-Aldrich, St. Louis, MO) for 36–48 h and
   12–24 h, respectively. After washing, the samples were cleared with
   BABB solution (benzyl alcohol:benzyl benzoate 1:2; Sigma) for imaging.
   For immunofluorescence in frozen sections, tissues embedded in OCT were
   sectioned (30–60 mm) using a Leica Cryostat (LeicaCM1860; Leica
   Biosystems, Nussloch, Germany). Primary and secondary antibody
   immunodetections were performed at RT for 24 and 2 h, respectively, in
   1% BSA/PBS-Triton solution. Slides were washed PBST (1X PBS, 0.01%
   Tween20) and mounted with mounting medium containing DAPI (Vector
   Laboratories) for nuclei detection.

   The following primary and second antibodies were used: rat anti-L1CAM
   (R&D, MAB5674, 1:3000), Rabbit anti-L1CAM (ABclonal, A8555, 1:3000),
   mouse anti-TUJ1 (Abcam, 78078, 1:3000), guinea pig ant-TUJ1 (Oasis
   Biofarm, OB-PGP031, 1:500), rabbit anti-SEMA3B (Abcam, AB48197, 1:200),
   rabbit ant-SOX10 (Abcam, [294]AB227680, 1:1000), goat anti-SOX10 (R&D,
   AF2864, 1:1000), guinea pig anti-SOX10 (Oasis Biofarm, OB-PGP001,
   1:500), rabbit anti-dsRed (Oasis Biofarm, OB-PRB013, 1:3000), rat
   anti-dsRed (Oasis Biofarm, OB-PRT017, 1:3000), rat anti-MBP (Millipore,
   MAB386, 1:250), goat anti-NRP1 (R&D, AF566, 1:250), goat anti NRP2
   (R&D, AF2215, 1:250), rabbit anti-RAB5 (CST, 3547S, 1:500), mouse
   anti-EEA1(BD, 610456, 1:200), rabbit anti-MKI67 (Abcam, AB5320,
   1:1000), rabbit anti-TH (Millipore, 657012, 1:2000), sheep anti-TH
   (Millipore, AB1542, 1:1000), rabbit anti-CGRP (Bachem, T-4032, 1:3000),
   goat anti-CGRP (Abcam, AB36001, 1:3000), B-FABP7(Oasis Biofarm,
   OB-PGP011, 1:500), rabbit anti-CASPASE3 (CST, 9661, 1:1000),
   biotin-conjugated Isolectin GS-IB4 (Sigma, L2140, 1:1000), chicken
   anti-PRPH (Abcam, AB39374, 1:1000), rabbit anti-NF-200 (sigma, N4142,
   1:1000), rabbit anti-PV (Swant, PV214, 1:3000), goat anti-PV (Swant,
   PV25, 1:3000), rabbit anti-SOX2 (Merck, AB-5603, 1:500), goat anti-OCT6
   (Santa-cruz, sc-11661, 1:250), rabbit anti-CDH2 (Abcam, ab19348,
   1:2000). The secondary antibodies used were Alexa-405, Alexa-488,
   Alexa-555, and Alexa-637 (all at 1:1000; Life Technologies, Carlsbad,
   CA).

Fluorescence in situ hybridization

   Single-molecule fluorescence in situ hybridization (FISH) was performed
   using RNAscope Fluorescent Multiplex kit (ACD: #320850; Newark, CA)
   according to manufacturer’s instructions, with the following probes:
   Sema3b probe (Cat# 522091), Scn7a (Cat# 548561), Ncmap (Cat# 577231),
   Nrp1 (Cat# 471621), L1cam (Cat# 577241), Nrp2 (Cat# 5006614), Matn2
   (Cat# 5006614), and Ednrb (Cat# 473801). Briefly, freshly dissected
   tissues were fixed overnight with 4% PFA, cryoprotected in 30% sucrose
   overnight at 4 °C, and then sectioned at 20 µm using a Leica Cryostat
   (LeicaCM1860; Leica Biosystems, Nussloch, Germany). After
   hybridization, the sections were stained with distinct antibodies
   following the same protocol as described above.

RNA isolation and quantitative real-time PCR

   Mice were anesthetized via inhalation of 3–4% isoflurane, and
   euthanasia was performed by cervical dislocation under deep anesthesia,
   followed by immediate tissue dissection. RNA from control and
   Sema3b^tgSC mouse sciatic nerves was extracted using TRIzol^TM Reagent
   (Invitrogen, Cat# 15596026). According to the manufacturer’s
   instructions, cDNA was synthesized from 1 μg RNA HiScript III RT
   SuperMix for qPCR (Vazyme, Cat# R323-01). Primers and probes were
   designed with the manufacturer’s online software. PCR primer sequences
   are available upon request. Real-time PCR was performed with a
   LightCycler 480 (Roche) using Universal Probe-Library reagents (Roche).
   A linear pre-amplification step was performed using Bestar^TM qPCR
   MasterMix (DBI, Cat# DBI-2043). The quantity of PCR products was
   normalized with internal control α-tubulin mRNA levels, compared with a
   standard curve generated by serial dilutions of input DNA.

Transmission electronic microscopy (TEM)

   Sciatic nerves were harvested from mice following transcardial
   perfusion with 4% PFA and 2.5% glutaraldehyde in 0.1 M PBS. The nerves
   were subsequently fixed in 2.5% glutaraldehyde and 4% PFA in PBS at
   4 °C overnight. After fixation with 1% osmium tetroxide for 1 h, the
   specimens were stained for 30 min with 2% uranyl acetate. Following
   dehydration through a graded series of ethanol and 100% acetone, the
   samples were embedded in epoxy resin (Serva). Ultrathin sections in
   100 nm were cut using a Leica RM 2155 microtome with a diamond knife
   (Histo HI 4317, Diatome) and counterstained with uranyl acetate and
   lead citrate. Images were obtained using a 100 kV transmission electron
   microscope (Canon, Japan). The g-ratio refers to the proportion of the
   fiber diameter relative to the same fiber’s diameter, including the
   myelin sheath, measured on electron micrographs for at least 100
   randomly selected axons per animal and nerve (n = 3 animals per
   genotype). The Remak bundle maturity (RBM) index was defined as the
   percentage of wrapped axons within the total axons in a Remak bundle.
   At least 10 randomly selected Remak bundles per animal were analyzed to
   compute the RBM index per animal (n = 3 animals per genotype).

3D reconstruction of the Remak bundle

   After fixation with 2.5% glutaraldehyde for a minimum of 24 h at 4 °C,
   the sciatic nerves were processed with a solution comprising equal
   volumes of 2% osmium tetroxide and 3% potassium ferrocyanide for a
   duration of 1 h. After rinsing with double-distilled water (ddH[2]O),
   the specimens were incubated with 1% thiocarbohydrazide for 20 min at
   RT. Thereafter, the specimens underwent additional rinsing with ddH[2]O
   and were incubated with 2% osmium tetroxide for 30 min at RT. Following
   a subsequent wash with ddH[2]O, the tissues were immersed in 1% uranyl
   acetate overnight at 4 °C. After incubation in 0.66% lead nitrate
   diluted in 0.03 M L-aspartic acid for 30 min at 60 °C, samples were
   dehydrated through an ascending series of ethanol concentrations and
   100% acetone, and then embedded in EPON 812 resin for 48 h at 60 °C.

   Resin blocks were meticulously trimmed using a Leica EM trimmer until
   the surface of the black tissue within the block was discernible. A
   scanning electron microscope (Thermo Fisher, Teneo VS) equipped with an
   ultramicrotome in its specimen chamber was employed for serial imaging.
   Data was collected in serial-surface view mode with a slice thickness
   of 100 nm. Each serial section was imaged using a 2.5-kV acceleration
   voltage and a 0.2-nA current in backscatter mode (BSE). The image
   storing resolution was configured to 6144×6144 pixels with a pixel
   dwell time of 1μs. For each sample, 225 slices were obtained,
   facilitating the reconstruction of RBs, with one whole sensillum
   contained in the image stack measuring 22.5 µm in depth.

   RBs were reconstructed using Amira software version 3D 2020.2 (Thermo
   Fisher Scientific, USA). The individual channels of the Z-stacks were
   imported into the digital environment. The “Align Slices” module
   facilitated the automatic alignment of a stack of 2D cross-sections.
   Employing the segmentation tool, contours of the nuclei and cytoplasm
   of Schwann cells and axons were meticulously drawn on each virtual
   section of the stack. The “Generate Surface” tool was used to produce a
   surface view of the RB. Volume rendering was applied to elucidate the
   interaction between SCs and axons within the internal structure of the
   RB. The video was generated utilizing the “Animation” tab, with the
   “Camera Path” module used to determine the focal angles. The “Clipping
   Plane” module facilitated the examination of each interaction layer
   between SCs and axons within each RB, with the display mode set to
   volume rendering.

Mouse behavior assays

   Behavioral tests were performed on mice aged between 8 and 12 weeks
   utilizing a double-blind methodology. The von-Frey test is used to
   evaluate baseline mechanical nociception. Mice were initially
   acclimated on a wire rack for at least 40 min, followed by the
   application of non-noxious graded stimuli (von Frey filaments ranging
   from 0.008 to 6 grams of force) perpendicularly to the plantar surface
   of the hind paw, positioned between the footpads, via wire-mesh
   observation cages. The paw withdrawal threshold (PWT) was ascertained
   by systematically increasing and decreasing the stimulus intensity,
   followed by analyzing withdrawal responses utilizing a Dixon
   non-parametric test. The test was undertaken three times at
   thirty-minute intervals, and these baseline scores were then averaged
   for each experimental group across. The thermal sensitivity evaluation
   was conducted using the hot plate assay, with the temperature plate set
   to 50 °C, 53 °C, and 56 °C, respectively. The latency of mouse
   responses to heat, indicated by behaviors such as jumping or paw
   licking, was meticulously documented. A 30-s cutoff time was rigorously
   and strictly enforced for the assay to prevent paw injury. The
   experimenters were blinded to the mouse genotype during the experiment.

Mouse nerve crush injury surgery

   A sciatic nerve crush injury was performed under sterile conditions.
   Mice were anesthetized with 3% isoflurane and maintained with 2%
   isoflurane throughout the procedure. An incision was made along the
   longitudinal axis of the femur to expose the sciatic nerve. The sciatic
   nerve underwent two 30-s crush injuries using flat forceps. The injury
   site was marked with an epineural suture on the surface of the muscle.
   The muscle and skin layers were subsequently closed using 6.0 silk
   sutures. In sham-operated animals, the nerve was exposed at the
   mid-thigh level without being crushed. For sciatic nerve crush, the
   nerve was crushed for 15 s, using a fine forceps (Dumont #55). The
   proximal section of the sciatic nerve at the injury site was collected
   from the mice at 7 and 21 days post-injury for analysis.

Sensory neurite outgrowth assay

   Neurite outgrowth assays were executed in accordance with previously
   established protocols^[295]94. Briefly, DRGs at the lumbar level of
   E12.5 embryos were dissected and cultured on PDL/laminin double-coated
   coverslips and subsequently cultured in culture medium prepared with
   neurobasal medium (Invitrogen, 21103-049) supplemented with 10% fetal
   bovine serum (Gibco, 10091130), 50mM L-Glutamic acid (Invitrogen,
   25030-024), 1X B-27 (Invitrogen, 17504-044), and 1%
   penicillin/streptomycin (Gibco, 15140-122). Nerve growth factor (NGF)
   (R&D Systems, 256-GF) was administered at 100 ng/ml concentration.
   After a 24-h culture period, SEMA3B (R&D Systems, 5440-S3-025/CF), with
   or without Dynasore, was introduced into the medium for a 2-h
   incubation duration. A clustered human IgG-Fc fragment (Jackson
   ImmunoResearch) was used as a control.

Sensory axon fasciculation assay

   DRGs at the lumbar level of E12.5 embryos were dissected and
   subsequently dissociated by incubating them in a 0.25% trypsin
   (Invitrogen, cat# 25300-054) prepared in the neurobasal medium for
   5–10 min at 37 °C until complete dissociation was achieved. Primary
   sensory neurons were then plated on PDL/laminin-coated coverslips and
   cultured in the identical culture medium as described above with
   supplemented with nerve NGF at a high concentration of 500 ng/ml to
   facilitate sensory axon outgrowth. The medium was replenished every
   48 h, and samples were collected after a 96-hour incubation period when
   fasciculated axons were visible under a light microscope.

Primary culture of Schwann cells

   Dissected sciatic nerves from P0 mice were cut into short pieces
   (2–3 mm in length) and placed on PDL/laminin-coated coverslips.
   Approximately 10 nerve segments were placed on each coverslip spaced to
   facilitate the expansion of SCs. The nerves were cultured in the
   culture medium for the sensory neuron culture supplemented with 3%
   horse serum, 100 ng/ml NRG1β1/HRG1β1 (R&D, cat. no. 396-HB-100), and
   0.1 mM db-cAMP (Sigma-Aldrich, D0260-5mg). After 48 h of culture, the
   attached SCs were dissociated with 0.25% trypsin and transferred to
   PDL-coated coverslips with the same medium for an additional 24 h
   before collection for analysis. For L1CAM and CDH2 assay, the
   coverslips were double-coated with L1CAM (200 µg/mL, R&D, 5674-NC) or
   CDH2 (200 µg/mL, R&D, 6626-NC) through overnight incubation at 4 °C.

Multicolor super-resolution STED microscopy

   Immunofluorescence staining of the sciatic nerves was performed
   following the same procedure as described above, except for the
   secondary antibodies used. The following secondary antibodies were used
   for STED microscopy: Star orange goat anti-rabbit (Abberior,
   STORANGE-1002-500 UG, 1:100), Star 460 L goat anti-guinea pig
   (Abberior, ST460L-1006-500 UG, 1:100), Star red goat anti-rat
   (Abberior, STRED-1007-500 UG, 1:100), as well as donkey anti-goat 647
   (Jackson ImmunoResearch, 705-005-147, 1:1000), and donkey anti-rat 555
   (Jackson ImmunoResearch, 712-565-150, 1:1000). Multicolor STED images
   of the RB on the sections were acquired using Abberior Facility Line
   (Abberior Instruments GmbH, Göttingen, Germany) fluorescence microscope
   with an oil immersion objective (UPlanSApo 100/1.4 oil, Olympus, Japan)
   in the presence of an imaging buffer. For the ultrastructural analysis
   of the locations of distinct protein particles, three different
   avalanche photodiodes and three different color filters were utilized
   (500–570 nm; 605–625 nm; 650–730 nm). Meanwhile, one confocal channel
   was used to detect DAPI signals to ensure that STED and confocal
   signals are acquired in parallel. Raw images were processed to correct
   for the chromatic shift between the STED and confocal channels. The
   number of protein clusters, as well as clusters with co-localized
   proteins, within the region of interest (ROI) was quantified using Fiji
   software. The data presented are representative of at least two
   independent experiments.

Quantification and statistical analysis

   The number of biological replicates differed across the experiments,
   with a minimum of three samples analyzed from at least six mice. All
   data and graphs generated in this study were analyzed using GraphPad
   Prism 9.0 software (GraphPad Prism software, San Diego, CA).
   Experimental groups were compared employing a two-tailed unpaired
   t-test for two-group comparisons and one-way ANOVA followed by Tukey’s
   multiple comparisons for three or more groups. The Mann-Whitney test
   was utilized for non-parametric comparisons between two groups with
   unmatched distributions. Significance was determined at p < 0.05.
   Detailed statistical parameters, including definitions, precise values
   of n, and p values, can be found within the figures, figure legends,
   and STAR Methods. In behavioral experiments, all procedures were
   conducted blindly, and analyses were performed on approximately equal
   numbers of male and female mice randomly selected from populations. No
   sex-specific differences in disease progression were identified.

Reporting summary

   Further information on research design is available in the [296]Nature
   Portfolio Reporting Summary linked to this article.

Supplementary information

   [297]Supplementary Information^ (12.2MB, pdf)
   [298]41467_2025_61318_MOESM2_ESM.pdf^ (90.7KB, pdf)

   Description of Additional Supplementary Files
   [299]Supplementary Video 1^ (3MB, mp4)
   [300]Supplementary Video 2^ (7.4MB, mp4)
   [301]Reporting Summary^ (75.2KB, pdf)
   [302]Peer Review file^ (519.5KB, pdf)

Source data

   [303]Source Data^ (114.9KB, xlsx)

Acknowledgements