Myelination of peripheral nerve fibres by Schwann cells is crucial for axonal function and long-term integrity. A longstanding question is whether these myelinating glia diversify according to axonal properties, such as motor versus sensory modality. However, modality-specific Schwann cell phenotypes remain unconfirmed, largely because technical constraints have prevented disentangling axon modality from the closely correlated variable of axon diameter. Here, by combining cell-size imaging with single-cell transcriptomics and myelin proteomics, we show that Schwann cells adopt molecular profiles that depend on axon calibre, rather than fibre modality. This molecular specialization includes a cell size-based tuning of myelin composition, which we found is critical for the long-term survival of specific fibre populations. Taken together, our data support a model in which myelinating Schwann cell diversity emerges as a tailored molecular adaptation to axonal surface cues, with relevance for fibre type integrity and function in the peripheral nervous system.
On this website we visualize data collected for this article. This includes single cell RNA sequencing data from peripheral nerves collected using the iCell8 plattform and SmartSeq2 technology alongside bulk RNA sequencing data from peripheral nerves and shotgun proteomics from purified dorsal and ventral root myelin.
Peripheral nerve single cell suspensions of lumbar dorsal roots, lumbar ventral roots, both sciatic nerves and both tibial nerves were prepared from individual animals. Peripheral nerves were freshly dissected and collected in chilled PBS (containing 1:1000 Hoechst 33342, abcam ab228551) and epineurial-perineurial connective tissue was carefully removed before gentle teasing. Tissue was transferred to 400 µl digest solution containing 0.0125% Trypsin (Gibco), 2 mg/ml Collagenase II (Worthington), 1:1000 Hoechst 33342 (abcam) and 1:20 Fluoromyelin green (ThermoScientific) in DMEM (1 g/l Glucose, with Pyruvate and Lactate, Gibco) and incubated for 20 min in a thermoshaker with 350 rpm at 37°C with additional manual agitation every 5 min. The enzymatic reaction was quenched by adding 200 µl fetal bovine serum (BioSell) and manual agitation before filling up the samples with 1400 µl chilled PBS and transfer to ice. The cells were pelleted by centrifugation at 300 × g at 4 °C for 10 min to remove supernatant and the cell pellet resuspended in 200 µl chilled PBS to obtain a concentrated single cell suspension. Cells were counted with a CASY cell counter (OMNI Life Sciences) for estimated cell numbers and verified by manual counting of cell numbers in droplets of 1 µl in quadruplets. Cells were diluted to 28,000 cells / ml and dispensed together with positive (10 ng/μl Control K-562 RNA) and negative controls (diluent) into 5,184-nanowell SmartChips (TAKARA, Single-Cell 350v Chips, 640300) with the iCELL8 cx Single-Cell System (TAKARA, 640000) prior dual colour imaging by the iCELL8 Imaging System. The plates were stored at -80 °C. The CellSelect software (TAKARA) allowed automated image processing and manual well selection. We manually inspected each individual well and measured length and thickness of every individual Schwann cell. For interpretation and correlation with the single cell transcriptomic data, we subsequently assigned the measured cells to three distinct Schwann cell classes: small (1st quartile: 111 µm, median: 141 µm, 3rd quartile: 170 µm), medium (1st quartile: 197 µm, median: 251 µm, 3rd quartile: 313 µm) and large (1st quartile: 334 µm, median: 397 µm, 3rd quartile: 445 µm). Of note, individual cells frequently did not align perfectly in a horizontal orientation within the cone-shaped wells, introducing a degree of measurement uncertainty in the determination of their precise length. We therefore consider classification into three size categories (small, medium and large) to be a more robust approach than directly correlating the measured cell length (in µm) with the transcriptional data. A mSC was considered broken when its nucleus appeared eccentric relative to the internode. In all selected wells, library preparation was performed according to the SMART-Seq iCELL8 Application Kit (TAKARA, 640219, 640220 & 640221) according to the manufacturer’s protocol. This included full-length cDNA synthesis, forward primer indexing, tagmentation, reverse primer indexing, library preparation, library amplification and library purification. 50 bp single-end sequencing was performed on a HiSeq 4000 (Illumina) to a read depth of approximatly 700 mio reads per chip.
Raw sequencing files (bcl) were converted into a single fastq file using bcl2fastq (Illumina). Each fastq file was de-multiplexed and analysed using the CogentAP analysis pipeline (Takara). The genome alignment was performed using mouse genome reference assembly GRCm39 and Ensembl gene annotation for GRCm39 version 113. The gene counts were summarized into gene matrices with the number of reads counted for each gene within each well. Statistical analysis of single-cell RNA sequencing data was performed with R 4.4.1 in RStudio using Seurat 5.2.0. Manual cell annotations from microscopic images were imported and simplified to nucleus, small, medium, and large Schwann cells. Wells with intact-appearing Schwann cells or individual nuclei and minimal debris were annotated as high quality (hq). Raw counts with and without intronic reads were imported from the Cogent gene-count matrices and merged into one Seurat object. The distributions of several high-level metrics were assessed manually while including positive and negative control wells. Cells were filtered to contain > 10000 counts, > 1500 genes, < 10 % mitochondrial counts, have a ratio of exonic to total counts per cell > 0.75, and a high-quality microscopical assessment. For further processing, counts were log1p-normalized using NormalizeData and centered and scaled using ScaleData. Highly variable genes (HVG) were selected using FindVariableFeatures using the mean.var.plot method. PCA was calculated using RunPCA on the HVGs. Batch effects between individual plates were reduced by integration with Harmony using the first 15 PCs. UMAP was calculated from Harmony dimensions 1:15. A KNN-nearest neighbor graph was also constructed from Harmony dimensions 1:15 using FindNeighbors and clustered with the Leiden algorithm using FindClusters at a resolution of 0.8. Clusters were assigned manually to cell types based on positive marker expression in each cluster. The distribution of cells across annotations of size and tissue was visualized using ggpointdensity. The correlation of PCA with size and tissue was calculated and visualized using PCAtools from the top 500 variable genes in a subset of cells identified as myelinating Schwann cells from lumbar roots. Genes specific for small and large myelinating Schwann cells were calculated on a subset of all cells identified as myelinating Schwann cells based on the size annotations using the Wilcoxon rank sum test as implemented in FindMarkers. As there were no biological replicates, we did not report p-values but instead visualized the expression of the top 6 markers for both small and large myelinating Schwann cells in each cell alongside the log2fold change using ComplexHeatmap.
The endoneurium of freshly dissected peripheral nerves from 6 months old C57Bl6/N mice was prepared in chilled PBS by careful removal of epineurial-perineurial connective tissue, except for peripheral nerve roots which stayed intact as they have only minimal connective tissue. Total RNA was extracted from the endoneurium of each peripheral nerve in a semi-automated fashion using a Maxwell RSC (Promega) and the Maxwell RSC simplyRNA Tissue Kit (Promega) according to the manufacturer’s instructions. In short, fresh-frozen nerve samples were mechanically homogenized in homogenization solution using the prechilled Bead Ruptor 24 (Omni). The tissue homogenate was then mixed with 200 μl lysis buffer and loaded onto the prepared cartridge. The cartridge was then placed into the Maxwell RSC (Promega) machine and the RNA isolated and purified automatically according to the manufacturer’s protocol. Integrity, purity and quantity of the isolated RNA were assessed using the Agilent 5200 Fragment Analyser System (Agilent) using the Fragment Analyser HS Total RNA Kit (Agilent). Library preparation for RNA-Seq was performed using a STAR NGS liquid handler (Hamilton) using the Illumina Stranded Total RNA with Ligation Ribo-Zero Plus Microbiome (96 reactions, Cat# 20072063) starting from 200 ng of total RNA. The size range of the final cDNA libraries was determined using the SS NGS Fragment kit (Agilent) on the Fragment Analyser (average 340 bp). Accurate quantification of cDNA libraries was performed by using the DeNovixDS-Series System. cDNA libraries were sequenced using the HiSeq4000 (Illumina) with 50 bp single-ended reads to 25 to 30 Mio reads/sample. Bcl files were demultiplexed using bcl2fastq 2.20. and sequencing quality determined using FastQC 0.11.
Reads were processed using the nf-core 77 [https://dx.doi.org/10.1038/s41587-020-0439-x] pipeline nfcore/rnaseq 3.18 [https://nf-co.re/rnaseq/3.18.0/]. Alignment was performed using STAR and transcripts were quantified using salmon using the mouse genome reference assembly GRCm39 and Ensembl gene annotation for GRCm39 version 113. Statistical analysis of bulk mRNA expression levels was performed using R 4.4.1 in RStudio. Individual transcript counts were imported from the nfcore/rnaseq pipeline and summarized to gene level using txi. Gene annotations were added using Biomart and the dataset subset to contain only genes annotated as protein-coding. Normalized expression levels were calculated using DEseq2. Strong outliers were assessed based on PCA and correlation matrix of the top 2000 variable genes and removed. The mean of the median axon diameters for each nerve was imported from our image analysis pipeline. We used the LRT test in DEseq2 to test for the association of gene expression changes with the differences in median axon diameter of each nerve. As we measured the median diameters in a separate experiment, we could only use their mean for each nerve. Therefore, we do not report the p-values as proof of differential expression. For genes that are not influenced by the differences in cell type composition between the nerves however, the resulting log2fold change of this test can still be interpreted as an estimated log2fold change of gene expression per 1 µm increase in median axon diameter. Using ComplexHeatmap, we plotted the normalized and scaled expression levels for the top 6 differentially-expressed genes for small and large myelinating Schwann-cells from scRNAseq, alongside the log2fold change from the LRT test, the scaled average expression in large, medium and small Schwann cells, as well as the means for median axon diameter and % motor axons for each nerve from our image analysis pipeline. We grouped the samples per nerve and ordered the groups by hierarchical clustering, manually changing the position of the final leaves after visual inspection.
Snap-frozen peripheral nerve tissue was homogenized in 1 ml of 0.29 M sucrose with freshly added Complete Mini protease inhibitors (Roche) for 3 × 10 s at 5000 rpm with 10 s break in a Precellys 24 tissue homogenizer (Bertin) in two cycles with 15 min incubation on ice. The homogenate was carefully layered over 1 ml of 0.85 M sucrose with protease inhibitors. Following centrifugation at 75,000 × g (TLS55 swinging bucket rotor, Beckman-Coulter) for 30 min at 4 °C, the interphases were transferred to fresh centrifugation tubes and clean water was added in excess to wash out the sucrose. Proteins were pelleted by centrifugation for 20 min at 75,000 × g at 4 °C and the supernatant carefully removed. The pellets were resuspended in clean water and incubated for 15 min on ice to dissolve cytoplasmic proteins by osmotic shock. After a final centrifugation at 12,000 × g for 15 min, supernatants were carefully removed and the pellet containing myelin proteins resuspended in 50 μl TBS (137 mM NaCl, 20 mM Tris-HCl, Roche Complete mini protease inhibitors, at pH7.4).
In-solution digestion of myelin proteins by filter-aided sample preparation (FASP)73 and LC-MS-analysis by different MSE-type data-independent acquisition (DIA) mass spectrometry approaches was performed as recently established for mouse PNS and CNS myelin 74,75. Briefly, protein fractions equivalent to 10 μg of myelin protein were solubilized in reducing lysis buffer (1% ASB-14, 7 M urea, 2 M thiourea, 10 mM DTT, 0.1 M Tris, pH 8.5) and alkylated with iodoacetamide. Samples were processed following a CHAPS-based FASP protocol using centrifugal filter units (30 kDa MWCO, Merck Millipore) and after detergent removal, the urea buffer was exchanged for digestion buffer (50 mM ammonium bicarbonate (ABC), 10% acetonitrile). Proteins were then digested overnight at 37 °C with 400 ng trypsin. Tryptic peptides were collected by centrifugation and subsequently extracted with 40 µl of 50 mM ABC and 40 µl of 1% trifluoroacetic acid (TFA). For quantification according to the TOP3 approach, combined flow-throughs were spiked with 10 fmol/μL of yeast enolase-1 tryptic digest or Hi3 EColi standard (Waters Corporation) and directly subjected to LC-MS-analysis. Tryptic peptides were separated by nanoscale reversed-phase UPLC and mass spectrometrically analysed on a quadrupole time-of-flight mass spectrometer with ion mobility option (Synapt G2-S/-Si, Waters Corporation) as recently described in detail 74,75. The samples were first analysed in the ion mobility-enhanced DIA mode with drift time-specific collision energies 76 referred to as UDMSE (providing increased proteome coverage at the cost of dynamic range for the accurate quantification of medium- to low-abundant proteins) and rerun in a data acquisition mode without ion mobility separation of peptides referred to as MSE (providing increased dynamic range at the cost of proteome coverage for the accurate quantification of exceptionally abundant myelin proteins) (see 74 for details).
Continuum LC-MS data were processed using Waters ProteinLynx Global Server (PLGS) and searched against a customized database comprising the UniProtKB/Swiss-Prot mouse proteome, supplemented with sequence data for porcine trypsin and the spike proteins yeast enolase 1/E. coli chaperone protein ClpB. Reversed sequences were included to enable false discovery rate (FDR) estimation, which was controlled at 1%. Post-identification analysis, including TOP3-based protein quantification, was carried out using the freely available software ISOQuant 76, as previously described 74,75. Proteins were quantified as parts per million (PPM), i.e., the relative abundance (w/w) of each protein with respect to the total protein content, but only if they were represented by at least two peptides (minimum length of six amino acids, score ≥ 5.5) identified in at least two runs. A false discovery rate (FDR) threshold of 1% was applied at both the peptide and protein levels, and at least one unique peptide was required. Known contaminants from blood (albumin, hemoglobin) and skin or hair (keratins) were excluded. Potential outliers were further assessed by examining peptide identification quality, quantification consistency, and distribution across isoforms. In the filtered protein datasets, protein abundance values from technical replicates were averaged per sample and log2-transformed. The resulting data were analysed using the Bioconductor R packages limma and q-value to detect significant differences in protein abundance via moderated t-statistics, as previously outlined 74,75. For data visualization, relative protein abundances in ppm alongside differential expression statistics were read into R. PCA was calculated from log2-transformed and scaled PPM values using prcomp. Non-compact myelin-associated genes were extracted from the gene ontology database with GO.db and AnnotationDbi. We searched for all terms that contained the strings: non-compact, node of ranvier, schmidt-lanterman incisiure, abaxonal, adaxonal, paranode, terminal loop, lateral loop and paranodal and extracted the genes associated with resulting GO-terms. We further augmented the results with a list of manually curated genes. We then visiualized the scaled log2-expression of all proteins that were measured in the UDMSE dataset transcribed from those genes alongside the differential expression statistics using ComplexHeatmap.