Testing SIPA1L2 as a modifier of CMT1A using mouse models

Abstract

Charcot-Marie-Tooth disease type 1A (CMT1A) is a demyelinating peripheral neuropathy caused by the duplication of peripheral myelin protein 22 (PMP22), leading to muscle weakness and loss of sensation in the hands and feet. A recent case-only genome-wide association study of CMT1A patients conducted by the Inherited Neuropathy Consortium identified a strong association between strength of foot dorsiflexion and variants in signal induced proliferation associated 1 like 2 (SIPA1L2), indicating that it may be a genetic modifier of disease. To validate SIPA1L2 as a candidate modifier and to assess its potential as a therapeutic target, we engineered mice with deletion of exon 1 (including the start codon) of the Sipa1l2 gene and crossed them to the C3-PMP22 mouse model of CMT1A. Neuromuscular phenotyping showed that Sipa1l2 deletion in C3-PMP22 mice preserved muscular endurance assayed by inverted wire hang duration and changed femoral nerve axon morphometrics such as myelin thickness. Gene expression changes suggest involvement of Sipa1l2 in cholesterol biosynthesis, a pathway that is also implicated in C3-PMP22 mice. Although Sipa1l2 deletion did impact CMT1A-associated phenotypes, thereby validating a genetic interaction, the overall effect on neuropathy was mild.

INTRODUCTION

Charcot-Marie-Tooth disease (CMT) is a collection of inherited peripheral neuropathies that result in demyelination and axon degeneration in motor and sensory axons of the peripheral nervous system (1). Although over 100 loci have been associated with CMT in the human genome, CMT1A is by far the most common form, accounting for about half of all cases (2). CMT1A is caused by unequal crossover at repeat sequences that flank 1.5 megabases of DNA on chromosome 17p12 resulting in a duplication of the intervening sequence (3–5). This duplication includes the peripheral myelin protein 22 (PMP22) gene and the increased gene dosage of PMP22 underlies the demyelinating neuropathy of CMT1A (2).

Although CMT is commonly considered to be a monogenic (Mendelian) disease, genetic and environmental modifiers can influence the clinical presentation and severity. This is particularly evident in CMT1A in which the patient duplications are quite homogeneous but clinical signs are still variable. A recent CMT1A case-only genome-wide association study (GWAS) performed by the Inherited Neuropathy Consortium identified 4 suggestive loci that associated with outcomes including difficulty in eating with utensils, hearing loss, decreased ability to feel, and the CMT neuropathy score (6). However, the strongest association was between weakness in foot dorsiflexion and 4 SNPs in introns 6, 7, 8, and 12 of the SIPA1L2 gene (7). Interestingly, this genome-wide analysis did not find associations for 2 previously implicated modifiers, LITAF, which can cause CMT on its own, and miRNA 149 (8–10).

SIPA1L2 contains a RAP/RAN GTPase activating domain (GAP), a PDZ domain, and a C-terminal coiled-coil domain. It associates with postsynaptic density proteins and is enriched at synapses in cultured hippocampal neurons. Unlike other SIPA1 family members, however, its overexpression does not alter synaptic spine morphology (11). It also associates with myosin heavy chain 9 and beta-actin (7). SIPA1L2 has been shown to regulate retrograde trafficking of BDNF/TrkB amphisomes, and Sipa1l2 knockout mice show impaired BDNF-dependent presynaptic plasticity (12). Importantly for myelination, SIPA1L2 is expressed in Schwann cells and is regulated by SOX10. Knockdown of Sox10 results in lower Sipa1l2 levels, and knockdown of Sipa1l2 results in decreases in other relevant genes including Pmp22, Mpz, and Egr2, suggesting that Sipa1l2 is in the same SOX10 co-expression network as other myelin genes (7). In addition to the recent association with CMT1A, it is also associated with Parkinson disease in multiple (but not all) patient cohorts (13–18).

The GWAS findings suggest that SIPA1L2 may be a modifier of CMT1A (7). Based on the idea that reducing SIPA1L2 expression may concomitantly decrease the expression of other myelin genes including PMP22, this also suggests that SIPA1L2 may be a therapeutic target for CMT1A. While the GWAS association of SIPA1L2 with foot dorsiflexion and CMT1A is impressive for the size of the study, it is still below the normal statistical threshold for genome-wide significance and therefore requires additional validation. Rare disease GWAS are very challenging because of low patient numbers, making additional studies in patients impractical. Finally, the effect of Sipa1l2 expression levels on myelin gene expression was identified in vitro in the S16 rat Schwannoma cell line (7). Reproducing the effect in vivo would strongly support the involvement of SIPA1L2 in the co-expression network.

We first searched sequencing data from 50 CMT1A patients to look for up- or down-stream variation in the vicinity of the SIPA1L2 gene that could impact gene expression, focusing on haplotypes that contained the SIPA1L2 SNPs reported in the original GWAS. No additional coding variants, variants in canonical splice junctions, or structural changes were detected. Because introns 6, 7, 8, and 12 of Sipa1l2 have low conservation between mice and humans (18.0%, 28.5%, 11.6%, 46.9%, respectively), engineering the human variants into the mouse genome is not straightforward and might have subtle effects that provide little insight into Sipa1l2 function. Therefore, we decided that engineering a truncated or null allele with a larger deletion in Sipa1l2 would provide better insight into gene function and disease-relevant genetic interactions. We used CRISPR/Cas9 genome editing to delete 1877 bp at the 5′-end of the Sipa1l2 gene, including the start codon. We also obtained the C3-PMP22 transgenic mouse model of CMT1A (19). These mice carry a human yeast artificial chromosome (YAC) transgene containing the PMP22 gene, and thus accurately model the increased gene dosage of CMT1A. They develop a relevant and progressive demyelinating phenotype (19). Mice with the Sipa1l2 deletion were bred to C3-PMP22 transgenic mice to assess the effects of altering Sipa1l2 levels on the severity of the demyelinating phenotype of C3-PMP22. We performed an analysis of neuromuscular phenotypes, including neurophysiology and histopathology. We also performed RNA-seq to assess changes in myelin gene expression and to identify altered or interacting pathways.

Here we show that although Sipa1l2 deletion does not induce neuromuscular phenotypes on its own, it does impact neuromuscular phenotypes associated with CMT1A, including muscular endurance, body weight, and myelin thickness in the C3-PMP22 mice. Our gene expression analysis identifies repression of several SOX10/EGR2 network genes, a potential role for Sipa1l2 in cholesterol biosynthesis, and an interaction with C3-PMP22 gene expression signatures. However, we did not detect changes in expression of Sox10 and Egr2 themselves at 6 months of age, suggesting that Sipa1l2 may interact with cholesterol biosynthesis and a subset of SOX10/EGR2 network genes in the absence of expression changes in myelination-related genes such as Pmp2, and Mpz. Alternatively, gene expression changes of these key transcription regulators and myelination genes may only occur developmentally with no detectable effects at 6 months while repression of cholesterol biosynthesis genes is more persistent. Taken together, our in vivo findings support Sipa1l2 as a modifier of CMT1A severity. However, the interaction does not appear to markedly rescue CMT1A phenotypes in this model.

MATERIALS AND METHODS

Sequence analysis of the human SIPA1L2 locus

All patient samples were collected as part of NINDS U54 NS065712 in accordance with World Medical Association guidelines with proper informed consent and data privacy protections. CMT1A duplication diagnosis was obtained by clinical genetic testing. Whole genome sequencing was performed from peripheral blood using the Illumina TruSeq DNA library preparation kit. DNA was sequenced to a mean depth of 30×. The generated sequencing data were then aligned to the GRCh37 reference genome using the Burrows-Wheeler aligner (version: 0.7.12) and variants were called using GATK (version 4.1.4.1) and then imported to the GENESIS genome database and analysis platform (20). Additional inferences with functional elements and haplotypes were made using the UCSC genome browser.

SIPA1 gene family expression in peripheral nerves from public resources

To assess the possibility of redundant functions among SIPA1 gene family members (SIPA1, SIPA1L1, SIPA1L2, SIPA1L3) gene expression levels from human tibial nerves were retrieved from the GTEx Portal (21). The data used for the analyses described in this section were obtained on January 10, 2024 and correspond to dbGaP Accession phs000424.v8.p2. Expression levels of these genes from mouse sciatic nerve were retrieved on January 10, 2024 from the Sciatic Nerve Atlas (SNAT) (22). Target gene expression in TPM (GTEx) and RPKM (SNAT), and the ratio of target gene expression to expression of GAPDH are reported (Supplementary Data Table S1). Similarity of the amino acid sequences of the proteins encoded by each SIPA1 family member were compared in a pairwise fashion using the online EMBOSS Needle webtool provided by EMBL-EBI (23). Percent similarity for each comparison is reported (Supplementary Data Table S2).

Experimental mice: C3-PMP22 and Sipa1l2 knockout

The C3-PMP22 mouse model of CMT1A was obtained from Dr Frank Baas of Amsterdam University. The official strain designation of these mice is B6.Cg-Tg(PMP22)C3Fbas/J, (MGI:5817396), but they will be referred to as C3-PMP22 for brevity. These mice carry a human YAC containing the PMP22 gene. They are a derivative of the C22-PMP22 line (which carries 7–8 copies of the transgene) and have undergone a spontaneous reduction of transgene copy number (now 3–4 copies) (19, 24). To create a Sipa1l2 loss-of-function allele, we used CRISPR/Cas9 genome editing, employing 2 guides flanking the first coding exon of mouse Sipa1l2 (exon 1 in NM_001081337.2; guide sequences: upstream sgRNA1: TTAAAGAGTGCTGCCGAAGC, downstream sgRNA1: CTATTAGACTGACAAAGCGT). The mouse gene consists of 21 coding exons. The region we targeted includes 5′-UTR sequence, the start codon, and the first 1486 bp of coding sequence in exon 1. Two double strand breaks were repaired with the deletion of the intervening 1877 bp including the entire exon sequence. These mice were bred for 3 generations to eliminate mosaicism and reduce possible off target mutations. C3-PMP22 transgenic mice were crossed with Sipa1l2+/− heterozygotes to produce Sipa1l2+/− mice carrying the PMP22 transgene. These C3-PMP22::Sipa1l2+/− mice were then crossed to WT::Sipa1l2+/− mice to produce experimental genotypes: WT::WT, WT::Sipa1l2+/−, WT::Sipa1l2−/−, C3-PMP22::WT, C3-PMP22::Sipa1l2+/−, C3-PMP22::Sipa1l2−/−. Experimental cohorts were aged to 4 and 6 months for experiments. Mice were housed under standard conditions with 14:10-hour light/dark cycles and ad libitum access to food and water. All procedures were performed in accordance with The Guide on the Care and Use of Laboratory Animals and were approved by the Institutional Animal Care and Use Committee of The Jackson Laboratory.

Sipa1l2 deletion confirmation

To confirm the deletion in Sipa1l2, primers were designed that flank the deleted interval (forward = 5′-AAG ATC CCA GGG AGA AGG AA-3′, reverse = 5′-GAT CCC CAG GAG ACA CTC AA-3′). Amplification from genomic DNA was performed using standard procedures and is expected to yield a 2179-bp product from the wild-type allele and a 302-bp product from the targeted allele. Products were run on an agarose gel with molecular weight standard (Quick-Load Purple 1 kb DNA Ladder, NEB N0552S). Sipa1l2 mRNA was also amplified from reverse transcribed total RNA from lung and brain tissue collected from 5-month-old Sipa1l2+/+ and Sipa1l2−/− mice. RNA was prepared by first homogenizing in Trizol (Invitrogen 15596026 [Waltham, MA]). Following the addition of chloroform and centrifugation, the aqueous layer, containing the RNA, was mixed with 1.5 volumes of 100% ethanol and transferred to an RNeasy mini column (Qiagen 74104 [Hilden, Germany]). The manufacturer’s protocol was then followed to isolate total RNA. cDNA was synthesized using SuperScript III reverse transcriptase (Thermo 18080051 [Waltham, MA]). Sipa1l2 mRNA was amplified from the cDNA using primers located in exons 1 and 6 (forward = 5′-CGA TAG ACG CAT TGC CCT CT-3′, reverse = 5′-CGT TTT CAG TGC AGG GGT TG-3′). Products were run on an agarose gel with molecular weight standard (GeneRuler 100 bp DNA Ladder, Thermo SM0243).

Genotyping

Toe biopsies were obtained from mice during the first postnatal week or ear notches were collected >14 days of age. To extract DNA for PCR, toes or ear notches were digested in a 50 mM NaOH solution for 25 minutes at 95C, then cooled and neutralized with Tris-HCL buffer. Products were visualized using standard agarose gel electrophoresis with ethidium bromide staining to detect PCR products specific to the C3-PMP22 transgene or the Sipa1l2 deletion or used for Sanger sequencing. Primers are as follows: C3-PMP22 forward = 5′-CCC CTT TTC CTT CAC TCC TC-3′ and reverse = 5′-CCA ATA AGC GTT TCC AGC TC-3′; Sipa1l2 deletion forward = 5′-CAG CCT TGC ACA ACA GGA TA-3′, reverse = 5′-CTC CCA TCT GTG CAG CTA TCA-3′; Sipa1l2 wildtype forward = 5′-CAG CCT TGC ACA ACA GGA TA-3′, reverse = 5′-TGG AAG GGT CCT TGA GTT TG-3′. Gel electrophoresis of genomic DNA was run with a Quick-Load Purple 1 kb DNA Ladder (NEB N0552S).

Neuromuscular phenotyping

Muscular endurance was evaluated using the inverted wire hang duration test. Briefly, mice were placed on a wire mesh grid, which was then inverted, and the time they can remain suspended was recorded. Trials lasted up to 1 minute, and the average of 3 trials is reported (25). Sciatic motor nerve conduction velocity (NCV) was calculated using electromyography with electrode placement described previously (25–27). Mice were anesthetized with isoflurane and placed on a heating pad to maintain normal body temperature. Recording electrodes were placed in the left rear paw, and stimulating electrodes were placed near the ankle and then near the sciatic notch. NCV was calculated in m/s by dividing the distance between stimulation points by the difference in latency between the greatest compound muscle action potential amplitude elicited by stimulation at the sciatic notch and ankle. Body weights were recorded. Mice were euthanized using CO₂ and tissues were collected. Triceps surae muscles were collected, and weights were recorded for muscle-weight-to-body-weight measurements to assess atrophy. All statistical comparisons were performed using GraphPad PRISM 9. Two-way ANOVA was applied to test differences between all genotype means, and either Tukey multiple comparisons test (wire hang duration, nerve conduction velocity, muscle-to-bodyweight ratio) or Dunnett multiple comparisons test (body weight) were applied.

Histology

Motor and sensory branches of femoral nerves were dissected free and fixed overnight in 2% glutaraldehyde and 2% paraformaldehyde in a 0.1 M cacodylate buffer. Nerves were processed for histology by Electron Microscopy Services at The Jackson Laboratory. Sections were stained with toluidine blue and imaged using a Nikon Eclipse E600 microscope with a 40× objective or a Leica DM6 upright widefield microscope with a 63× oil objective. High magnification images were adjusted in Adobe Photoshop using the levels function to uniformly set black level and saturation to take advantage of the full dynamic range and improve image contrast. Semiautomated quantification was performed using lower magnification images in ImageJ using threshold function and the Analyze Particles plugin as described previously (28). The line tool in ImageJ was used to record axon diameter and axon diameter plus myelin for 50 randomly selected axons per nerve section to calculate G-ratio (axon diameter/axon+myelin diameter). Images were visually inspected, and the number of totally amyelinated axons was recorded. Statistical comparisons were performed using GraphPad PRISM 9. Two-way ANOVA was used to test differences between all means with either Dunnett multiple comparisons test (axon counts) or Tukey multiple comparisons test (percent amyelinated and G-ratios).

RNA-seq

Tissues were lysed and homogenized in TRIzol Reagent (ThermoFisher) using a Pellet Pestle Motor (Kimble), then RNA was isolated using the miRNeasy Micro kit (Qiagen), according to manufacturers’ protocols, including the optional DNase digest step. RNA concentration and quality were assessed using the Nanodrop 2000/Nanodrop 8000 spectrophotometer (Thermo Scientific) and the RNA 6000 Nano and Pico Assay (Agilent Technologies, Santa Clara, CA). Libraries were constructed using the KAPA mRNA HyperPrep Kit (Roche Sequencing and Life Science), according to the manufacturer’s protocol. Briefly, the protocol entails isolation of polyA containing mRNA using oligo-dT magnetic beads, RNA fragmentation, first and second strand cDNA synthesis, ligation of Illumina-specific adapters containing a unique barcode sequence for each library, and PCR amplification. The quality and concentration of the libraries were assessed using the D5000 ScreenTape (Agilent Technologies) and Qubit dsDNA HS Assay (ThermoFisher), respectively, according to the manufacturer’s instructions. Libraries were sequenced 150 bp paired-end on an Illumina NovaSeq 6000 using the S4 Reagent Kit v1.5. Raw sequencing data are available (GSE249764). Analysis of RNA-seq data was performed using an open source Nextflow pipeline (v3.5) at The Jackson Laboratory comprising tools to perform read quality assessment, alignment, and quantification (29). The pipeline takes as input a single sample and outputs read counts. FastQC (RRID: SCR_014583) was used for quality checks and then Trim Galore! (RRID: SCR_011847) was used to remove adapters and sequences with low quality (Fred<20). Sequence reads that passed quality control were aligned to mouse reference (GRCm38) using STAR (v2.7) (RRID: SCR_004463) and gene expression estimates were made using RSEM (v1.3) (RRID: SCR_013027) with default parameters. BAM files were loaded into Integrative Genomics Viewer (IGV) to visually inspect alignments (30). Differential expression analysis was performed using a differential expression script in R developed by Computational Sciences at The Jackson Laboratory based on the EdgeR package (31). We selected FDR<0.05 and absolute Log₂FC > 1.5 cutoffs to determine differential expression. We normalized raw counts values using trimmed mean of m-values (TMM) at this stage for follow-up Gene Set Enrichment Analysis.

Over-representation and gene set enrichment analyses

ENSEMBL identifiers for differentially expressed genes were uploaded to MouseMine.org to perform over-representation analyses through: Reactome pathways, Mammalian Phenotype Ontology, and Gene Ontology (32). We recognized at the outset that the relative paucity of differentially expressed genes would temper any conclusions drawn from this analysis. We only reported annotations with a Holm-Bonferroni corrected p value <.05. To improve our ability to detect changes in pathway regulation driven by coordinated low magnitude changes across multiple gene sets, we performed gene set enrichment analysis (GSEA) using TMM counts, an approach that would be less susceptible to bias due to arbitrary cutoffs. We used default GSEA settings, which we determined were appropriate for our dataset, including a weighted enrichment statistic, the Signal2Noise method for ranking genes, and 1000 permutations. We queried against the MSigDB Reactome pathway gene lists. After a comprehensive analysis of parameterizations, we determined that an Experimental_versus_Control design focused on negative normalized enrichment scores was most informative to address our experimental question related to the interaction of Sipa1l2 deletion with CMT1A phenotypes. We performed a GSEA leading edge analysis to identify the core driver genes associated with cholesterol biosynthesis pathways for all genotypes (33). Each gene list was analyzed against the ENCODE database through the Network Analyst web tool to identify upstream regulators (34).

RESULTS

Sequence analysis of the human SIPA1L2 locus reveals no additional variants

We performed whole genome sequencing on 50 patients with CMT1A. The goal was to evaluate the presence of variation in the 117 kb genomic space of the SIPA1L2 gene that could impact gene expression. The coding and noncoding sequences of 50 CMT1A patients were studied including 2 kb upstream and downstream of the SIPA1L2 gene (NM_020808). Special attention was given to the haplotypes that contained the associated SNP (rs10910527, rs7536385, rs4649265, rs1547740) reported in the original GWAS (7). No coding variation with CADD scores >18, MAVERICK>0.1 were identified (35). No variation was identified by SpliceAI that disrupts canonical splice junctions or that was predicted to have splicing effects. No structural changes were present in these CMT1A samples. While this does not present an exhaustive screen, it remains unclear whether the associated marker SNPs in SIPA1L2 are correlated with pathogenic or quantitative trait loci in the vicinity of SIPA1L2. More comprehensive future studies, including long read sequencing, will potentially clarify the presence of functionally active DNA variation.

Sipa1l2 deletion mice do not exhibit neuromuscular phenotypes

Given the poor conservation of introns between mice and humans in the Sipa1l2 gene, rather than introducing the human SNPs identified in the GWAS, we used CRISPR/Cas9 with 2 guides flanking the first coding exon of the Sipa1l2 gene (exon 1 in NM_001081337.2) on mouse chromosome 8. This produced a 1877-bp deletion including the first 1486 bp of coding sequence (Fig. 1A). These mice were bred for at least 3 generations to eliminate mosaicism and reduce possible off target mutations. We verified the deletion in homozygous Sipa1l2−/− mice by PCR of genomic DNA, which produced PCR products of the anticipated size (Fig. 1B). The 1877-bp deletion was also verified with Sanger sequencing of PCR products spanning the deleted region in genomic DNA (Fig. 1C). We performed RT-PCR using lung and brain mRNA from Sipa1l2−/− mice and found Sipa1l2 is not detected in homozygous knockout mice when using a primer combination that includes the coding sequence deleted in exon 1 (Fig. 1D). RNA-seq read alignment showed that, indeed, exon 1 of Sipa1l2 lacks coverage in homozygous knockouts; however, exons downstream of the first exon have coverage at approximately wild-type levels (Fig. 1E).

Figure 1.

CRISPR engineered deletion of exon 1 of Sipa1l2. (A) CRISPR/Cas9 genome editing was used to introduce a 1877-bp deletion containing the first 1486 coding nucleotides of Sipa1l2. The gene model is read left-to-right, inverted from its original orientation on the reverse strand for readability. The red box indicates the deletion and red half arrows indicate the position of primers used to sequence and amplify genomic DNA. Blue half arrows indicate primers used for RT-PCR. (B) Gel electrophoresis of PCR products from genomic DNA showing deletion of 1877 bp in homozygous Sipa1l2 knockout mice. The exposure of the lane containing the ladder has been increased to improve readability. (C) Sanger sequence chromatogram of genomic DNA defining the break points of the deletion in homozygous Sipa1l2 knockout mice. (D) RT-PCR from lung and brain mRNA indicates that Sipa1l2 amplicons are absent in homozygous Sipa1l2 knockout mice using a primer in the deleted region. Intervening lanes between experimental samples and negative control lane have been removed. (E) RNA-seq alignment in IGV from representative Sipa1l2 knockout and wild-type control mice reveals an absence of read coverage along exon 1 of Sipa1l2 in knockout mice. The normal wildtype coverage is indicated by the red box. However, expression was detected in downstream exons at levels similar to wild type. The gene model in this IGV view is read right-to-left reflecting that Sipa1l2 is on the reverse strand.

Sipa1l2 deletion alters neuromuscular phenotypes relevant to CMT1A

Grip strength and endurance were assayed using the inverted wire hang test (see Methods). Testing at 4 months identified significantly (p = 0.0007) reduced wire hang duration in C3-PMP22 mice compared to wild-type littermates. Wire hang duration of C3-PMP22 mice with either heterozygous or homozygous Sipa1l2 deletions did not differ significantly from either wild type or C3-PMP22 littermates, indicating an intermediate phenotype (Fig. 2A). At 4 months, body weights did not differ significantly between genotypes (Fig. 2B). Sciatic motor nerve NCV was significantly reduced in all genotypes carrying the C3-PMP22 transgene compared to wild-type littermates, and Sipa1l2 deletion did not modify this phenotype (Fig. 2C): Wild type vs C3-PMP22 (p < 0.0001), wild type vs C3-PMP22::Sipa1l2+/− (p < 0.0001), and C3-PMP22::Sipa1l2−/− (p = 0.0003). These tests were also performed at 6 months of age. Wire hang duration was significantly decreased in C3-PMP22 (p = 0.0076) and C3-PMP22::Sipa1l2+/− genotypes (p = 0.0086) compared to wild-type littermates, while the C3-PMP22::Sipa1l2−/− mice were again not significantly different than control, suggesting some persistent benefit of homozygous Sipa1l2 deletion in the C3-PMP22 background (Fig. 2D). Body weights were significantly (p = 0.012) decreased only in the C3-PMP22::Sipa1l2−/− genotype compared to wild-type littermates (Fig. 2E). NCV was diminished (p < 0.0001) in all genotypes containing the C3-PMP22 transgene, and Sipa1l2 deletion did not modify nerve conduction velocity (Fig. 2F). Triceps surae muscle-to-body weight ratio did not differ significantly between any genotype (not shown).

Figure 2.

Neuromuscular phenotyping in 4- and 6-month-old mice. (A) Bar plot of wire hang duration of 4-month-old mice as measured by average hang duration of 3 consecutive trials. Each point represents one mouse. C3-PMP22 mice have significantly reduced wire hang duration (p = 0.0007) compared to littermate wild-type mice. (B) Bar plot of body weights of 4-month-old mice. Each datapoint indicates one mouse. No significant differences are detected between genotype. (C) Bar plot of sciatic motor NCV of 4-month-old mice. Each data point represents one mouse. Significant differences are detected between wild type and C3-PMP22 (p < 0.0001), wild type and C3-PMP22::Sipa1l2+/− (p < 0.0001), wild type and C3-PMP22::Sipa1l2−/− (p = 0.0003). (D) Bar plot of wire hang duration of 6-month-old mice as measured by average hang duration of 3 consecutive trials. Each datapoint represents one mouse. C3-PMP22 and C3-PMP22::Sipa1l2+/− mice have significantly reduced wire hang duration (p = 0.0076 and p = 0.0086, respectively) compared to wild-type littermates. (E) Bar plot of bodyweight of 6 months mice. Each data point represents one mouse. Only C3-PMP22::Sipa1l2−/− mice differed significantly from wild types (p = 0.012). (F) Bar plot of sciatic motor NCV of 6-month-old mice. Each datapoint presents one mouse. Significant differences were detected between wild-type mice and C3-PMP22, C3-PMP22::Sipa1l2+/−, and C3-PMP22::Sipa1l2−/− (all p < 0.0001). All comparisons are 2-way ANOVA with either Tukey multiple comparisons test (wire hang duration, nerve conduction velocity) or Dunnett multiple comparisons test (body weights). Error bars depict standard deviation.

Nerve histology reveals interactions between Sipa1l2 deletion and myelination

We collected motor and sensory branches of the femoral nerve at 6 months of age. Nerves were sectioned, stained with toluidine blue, and visualized at 40× and 63× by light microscopy. Genotypes expressing the C3-PMP22 transgene exhibit evident demyelination in the motor branch of the femoral nerve compared to those genotypes without the transgene, but this effect is not clear in the sensory branch (Fig. 3). We counted the number of axons in both motor and sensory branches of the femoral nerve and identified no significant differences in the number of axons of femoral nerve motor branches between any genotypes (Fig. 4A). We detected significantly fewer axons in femoral nerve sensory branch for C3-PMP22 (p = 0.0001) and C3-PMP22::Sipa1l2−/− (p < 0.0001) mice, but interestingly, this reduction was not observed in C3-PMP22::Sipa1l2+/− mice (Fig. 4B). We next examined the ratio of totally amyelinated axons to total axons in both femoral nerve branches. All genotypes expressing the C3-PMP22 transgene exhibited significantly more amyelinated axons in the motor branch of femoral nerve (Fig. 4C). No difference between wild-type mice and other genotypes in the percentage of amyelinated axons was detected in the sensory branch (Fig. 4D). It should be noted that any degree of myelination, even partial myelination, was scored as “myelinated.” Therefore, this analysis could miss effects on the degree of myelination in thinly myelinated axons present in both motor and sensory branches.

Figure 3.

Representative histopathology of femoral nerve branches. Representative images of toluidine blue-stained sections from motor and sensory branches of femoral nerve visualized at 40× on a light microscope for all genotypes: (A) motor branches of Sipa1l2 genotypes in a wild-type background; (B) motor branches of Sipa1l2 genotypes in the C3-PMP22 background; (C) sensory branches of Sipa1l2 genotypes in a wild-type background; (D) sensory branches of Sipa1l2 genotypes in the C3-PMP22 background. Inset images were visualized at 63×. Scale bars: A = 50 µm and applies to all 40× images; inset = 10 µm and applies to all inset images.

Figure 4.

Nerve morphometrics from femoral nerve motor and sensory branches. (A) Bar plot of the total number of axons in the motor branch of the femoral nerve. No differences are detected among genotypes. (B) Bar plot of the total number of axons in the sensory branch of the femoral nerve. Significantly reduced numbers of axons are detected between both C3-PMP22 (p = 0.0001) and C3-PMP22::Sipa1l2−/− (p < 0.0001). (C) Bar plot depicting the percentage of totally amyelinated axons in the motor branch of the femoral nerve. Significant reduction is detected for all genotypes expressing the C3-PMP22 transgene (C3-PMP22 and C3-PMP22::Sipa1l2−/− p < 0.0001, C3-PMP22::Sipa1l2+/− p = .0002). (D) Bar plot of the percentage of amyelinated axons in the sensory branch of the femoral nerve. No differences were detected. (E) Dot plot depicting G-Ratio of axons from the motor branch of the femoral nerve. Significant differences are detected between wild type and Sipa1l2+/− (p = 0.0063), wild type and C3-PMP22 (p = 0.0005), wild type and C3-PMP22::Sipa1l2−/− (p = 0.0009). (F) Dot plot depicting G-ratio of axons from the sensory branch of the femoral nerve. Significant differences are detected between wild type and Sipa1l2+/−, C3-PMP22, C3-PMP22::Sipa1l2+/−, and C3-PMP22::Sipa1l2−/− (all p < 0.0001). For axon counts and myelination ratios, each datapoint indicates one mouse. For g-ratios, 50 axons were randomly selected for quantification from each of 5 mice per genotype. All comparisons are 2-way ANOVA with either Dunnett multiple comparisons test (axon counts) or Tukey multiple comparisons test (percent amyelinated axons and G-ratios). Error bars depict standard deviation.

We next quantified g-ratios (defined as axon diameter/axon+myelin diameter) for both branches of the femoral nerve. Thus, thinner myelin leads to a higher g-ratio, closer to a value of 1. In the motor branch, Sipa1l2+/− mice exhibit an increased g-ratio (p = 0.0063), as do C3-PMP22 mice (p = 0.0005), whereas C3-PMP22::Sipa1l2−/− exhibit a g-ratio significantly (p = 0.0009) lower than that of wild types. Sipa1l2−/− and C3-PMP22::Sipa1l2+/− did not differ significantly from wild types (Fig. 4E). A similar increase above wild type is observed in the sensory branch of Sipa1l2+/− mice (p < 0.0001). All genotypes carrying the C3-PMP22 transgene have significantly (p < 0.0001) lower g-ratios in the sensory branch compared to wild type mice (Fig. 4F).

When the relationship between axon diameter and myelin thickness is plotted, the effect is clearer. Mice without the C3-PMP22 transgene have a normal, upward slope, with larger axons having thicker myelin. Mice with the C3-PMP22 transgene have a downward slope, with larger axons having thinner myelin and hypermyelination of small diameter axons. Notably, while the downward slope is similar, C3-PMP22::Sipa1l2−/− mice have thicker myelin across the range of axon diameters (Prism Simple Linear Regression Y-Intercept ANCOVA p < .0001) (Supplementary Data Fig. S1A). Myelin thickness was also impacted by Sipa1l2 deletion in the sensory branch of the femoral nerve; however, the effect is most obvious in Sipa1l2+/− mice on a wild-type background, where heterozygous deletion reduces myelin thickness across all axon diameters (Supplementary Data Fig. S1B).

These results are complex as g-ratios depend on changes in both myelin thickness and axon diameter. To summarize: First, there was axon loss in the sensory branch of femoral nerve in genotypes with the C3-PMP22 transgene, except C3-PMP22::Sipa1l2+/−. Second, there were more totally amyelinated axons in all genotypes with the C3-PMP22 transgene in the motor branch of femoral nerve. Further, in the motor branch, the C3-PMP22 transgene causes hypermyelination of small diameter axons and demyelination of large diameter axons, while Sipa1l2 deletion increased myelin thickness across all diameters in the C3-PMP22 background. In the sensory branch, myelination trends were less obvious but Sipa1l2+/− decreased myelin thickness in the wild-type background.

Assessing possible compensatory effects of SIPA1 family members

The effects of deleting Sipa1l2 on its own or in the C3-PMP22 transgenic background were significant but modest. It is possible that other Sipa1 family members compensated for the loss of Sipa1l2, or that in mice and humans, different Sipa1 family members have different expression patterns. To address this, we performed a comparison of SIPA1 gene family expression using publicly accessible datasets. We found that in human tibial nerve, SIPA1 is more highly expressed than SIPA1L2, while both SIPA1L1 and SIPA1L3 are expressed at lower levels. In mouse, Sipa1l1 is more highly expressed than Sipa1l2, while Sipa1 and Sipa1l3 are expressed at lower levels. The level of Sipa1l2 expression relative to Gapdh is doubled in mice compared to humans. Comparing mice to humans, the ratio of Sipa1 to Gapdh is much lower in mice while the ratio of Sipa1l1 to Gapdh is much higher (Supplementary Data Table S1). The percent similarity of amino acids from different SIPA1 family proteins compared to SIPA1L2 is very consistent between mice and humans. SIPA1L1 exhibits the highest similarity at 63% (Supplementary Data Table S2). In mice, the most similar Sipa1 family member, Sipa1l1, exhibits expression levels almost triple that of Sipa1l2 in sciatic nerve, whereas in human tibial nerve SIPA1L1 expression is half that of SIPA1L2. Therefore, in mice expression of Sipa1l1 in peripheral nerves may be masking some of the effect of changes in Sipa1l2, and this effect may be less pronounced in human peripheral nerve.

RNA-seq analysis of sciatic nerve implicates cholesterol biosynthesis

Library preparation and sequencing were performed using RNA isolated from sciatic nerves of wild type, Sipa1l2−/−, C3-PMP22, and C3-PMP22::Sipa1l2−/− mice because we presumed any interaction between SIPA1L2 and C3-PMP22 gene expression would be strongest in homozygous knockouts. Sciatic nerve was chosen because this tissue is rich in Schwann cells, the primary cell type affected in CMT1A. Our differential expression analysis compared all experimental genotypes against wild types and used a false discovery rate cutoff of FDR <0.05; we considered genes with an absolute Log₂FC>1.5 to be differentially expressed (Supplementary Data Table S3). We identified 59 significantly differentially expressed genes in C3-PMP22 mice, 7 differentially expressed genes in Sipa1l2−/− mice, and 88 differentially expressed genes in C3-PMP22::Sipa1l2−/− mice (Fig. 5A). Notably, we did not see changes in expression of any of the other Sipa1 gene family members that may indicate compensation for the loss of Sipa1l2. We also performed a targeted investigation of genes in the SOX10/EGR2 co-expression network and failed to identify any changes or differential expression in these genes in sciatic nerves at 6 months of age (Supplementary Data Fig. S2A). We also considered differentially expressed genes that were shared between multiple experimental genotypes. Only 3 genes overlapped between Sipa1l2−/−, C3-PMP22, and C3-PMP22::Sipa1l2−/− mice (Supplementary Data Fig. S2B). These genes are troponin I, ribosomal protein L34 pseudogene 1, and ribosomal protein S3A2. Their functions are related to calcium sensitivity in striated muscle, the large ribosomal subunit, and the small ribosomal subunit, respectively (36).

Figure 5.

Differential expression and gene set enrichment analyses. (A) Differentially expressed genes identified in comparisons between Sipa1l2−/−, C3-PMP22, and C3-PMP22::Sipa1l2−/− mice with wild type littermates were filtered by FDR<0.05 and absolute Log₂FC > 1.5. Gene lists were visualized with volcano plots. Upregulated genes are colored red and downregulated genes are colored blue. Gray dots indicate genes that were filtered from analysis. We identified 7 DEGs for Sipa1l2−/− mice, 59 DEGs for C3-PMP22 mice, and 88 DEGs for C3-PMP22::Sipa1l2−/− mice. (B) An enrichment plot depicting Reactome pathways identified for each experimental genotype based on TMM normalized counts data from RNA-seq. Pathways with negative normalized enrichment scores are in descending order from the most significant family-wise error rate across genotypes, with an FWER<0.05 cutoff. The x-axis reports the absolute value of the negative normalized enrichment score. In all cases, pathways are associated with a negative normalized enrichment score, or repressed. Genotypes are color coded.

We next performed a basic overrepresentation analysis using Mouse Mine. The gene list from C3-PMP22 mice was significantly associated with the Reactome pathway “striated muscle contraction” (Holm-Bonferroni p = 0.003211). The gene list from Sipa1l2−/− mice was also associated with “striated muscle contraction” (HB p = 3.19e-7) and “muscle contraction” (HB p = 1.60e-4). The C3-PMP22::Sipa1l2−/− mice gene list was not significantly associated with any Reactome pathways. Given the relatively low number of differentially expressed genes and the paucity of overrepresented pathways associated with them, we next performed GSEA using TMM normalized counts values of all genes, which would better suited for detecting the coordinated effects of subtle expression changes in multiple genes, rather than using sparse lists of differentially expressed genes.

GSEA Reactome pathways were filtered by family-wise error rate (FWER<0.05) and ordered by absolute normalized enrichment score (NES) (Supplementary Data Table S4). Across mutant versus wild type contrasts, for all experimental groups, we found that pathways associated with upregulated genes in the experimental group (positive enrichment scores) tended to vary with enrichment statistic and metric for ranking genes, and the pathways themselves were diverse. Alternatively, those pathways with negative normalized enrichment score, which are frequently interpreted as downregulated or repressed in the experimental genotype, were robust across enrichment statistics, ranking metrics, and mutant genotypes. For this reason, we chose to focus our analysis on negative normalized enrichment scores. We found that cholesterol biosynthesis and pathways related to the regulatory activity of Sterol Response Element Binding Proteins/Factors (SREBP/SREBF) on cholesterol biosynthesis have large negative NES across Sipa1l2−/−, C3-PMP22, and C3-PMP22::Sipa1l2−/− mice (Supplementary Data Fig. S2C). Both Sipa1l2−/− and C3-PMP22 are associated with a variety of other significant pathways with lower magnitude NES, though C3-PMP22::Sipa1l2−/− is associated with fewer (Fig. 5B). Leading edge analysis of all genotypes identified lists of core enrichment genes primary responsible for the identification of cholesterol-related pathways (Supplementary Data Table S4 and Fig. S2D). Query of the ENCODE transcription-factor database through Network Analyst revealed that many of these leading-edge genes are regulated by SREBP-1, providing a possible upstream regulator for the cholesterol-associated gene expression signatures.

DISCUSSION

Here we report a newly CRISPR-engineered strain of mice with exon 1 of Sipa1l2, which contains the initiation codon, knocked out. We did not detect any overt behavioral or neuromuscular phenotypes (e.g. NCV, wire hang performance, body weight) to be associated with the Sipa1l2 deletion in mice without the C3-PMP22 transgene but we did detect changes at the histological level in g-ratio and myelin thickness in the motor and sensory branches of the femoral nerve. These differences are most pronounced in mice heterozygous for the deletion although the biological basis of the heterozygous effect remains unclear. Gene set enrichment analysis identified a group of Reactome pathways related to cholesterol biosynthesis and its regulation among downregulated pathways in the Sipa1l2−/− mice. Given the absence of pronounced neuromuscular phenotypes in Sipa1l2−/− mice it is possible that another Sipa1 gene family member is performing a redundant function. The most likely candidate is Sipa1l1, which is highly expressed in mouse sciatic nerve and encodes a protein with 63.5% similarity to SIPA1L2. Importantly, SIPA1L1 is not highly expressed in human tibial nerve, suggesting any redundant function may be unique to mouse. However, differential expression of Sipa1l1 was not detected in our RNA-seq data.

We found several interactions between the Sipa1l2 deletion and C3-PMP22 phenotypes. These include wire hang duration, where the Sipa1l2 deletion promotes endurance in mice overexpressing human PMP22. We also detected a slight reduction of body weights in transgenic mice homozygous for the Sipa1l2 deletion. Notably, we did not detect an interaction between nerve conduction velocity deficits and the Sipa1l2 deletion. Nerve conduction velocity decrements are an important feature of CMT1A, and demyelinating CMTs generally (37, 38). We observed an interesting effect in nerve histology whereby heterozygous Sipa1l2 knockout seems to prevent axon loss in the sensory branch of the femoral nerve. A protective effect is also detected in the motor branch of the femoral nerve of heterozygous knockouts, leading to no significant difference in g-ratio between wild types and C3-PMP22::Sipa1l2+/− mice. Gene set enrichment analysis indicates repression of cholesterol biosynthesis across Sipa1l2−/−, C3-PMP22, and C3-PMP22::Sipa1l2−/− genotypes.

A previous study has demonstrated decrements in wire-hang performance of C3-PMP22 mice, which we also observed (39). The effect of the Sipa1l2−/− deletion on wire hang performance is perhaps most interesting because at both 4- and 6-month timepoints C3-PMP22::Sipa1l2−/− mice are not significantly different from wild types. C3-PMP22::Sipa1l2+/− mice, though not significantly different from wild type at 4 months, do show a significant decrease in endurance by 6 months. These data are quite variable and alone are not firm evidence of a protective effect but do suggest an interesting dose effect at both ages studied.

A previous study of histopathological changes in C3-PMP22 mice identified nonsignificant decreases in the number of myelin-competent axons in both dorsal (sensory neurons) and ventral (motor neurons) roots. These findings generally align with our observations in femoral nerves although we detected a significant decrease in axon number in the sensory branch of the femoral nerve. This same analysis identifies more amyelinated fibers in both dorsal and ventral roots of C3-PMP22 mice compared to wild types, which we also observe; however, we generally detected lower proportions of amyelinated fibers (39). This difference could be due to different ages at collection, different populations of neurons observed, or interobserver variability. However, variability in the extent of demyelination observed in the C3-PMP22 model between and within mice was reported in the original study describing these mice and other models such as Lama2 mutant mice show differences in myelination phenotypes between roots and more distal nerves (19, 40).

The Sipa1l2 deletion does appear to modify several histological measures in the C3-PMP22 background. C3-PMP22::Sipa1l2+/− mice do not exhibit reduction in axon number in the sensory branch of femoral nerve, nor significant differences in motor branch g-ratio, compared to wild type mice while C3-PMP22::Sipa1l2−/− do. In the motor branch of femoral nerve, homozygous Sipa1l2 deletion modestly increases myelin thickness. In this case, the increase in myelin thickness of transgenic mice with the homozygous Sipa1l2 deletion may cause significantly different g-ratios from wild type while a more modest effect in transgenic mice with a heterozygous deletion in Sipa1l2 maintain g-ratios closer to the wild-type mean. Our observation that heterozygous Sipa1l2 deletion in a wild-type background increases g-ratio due to thinner myelin in both branches of femoral nerve may indicate a role for Sipa1l2 in regulating myelin thickness, although this is speculative. While the interaction between Sipa1l2 deletion and myelination decrements in C3-PMP22 mice is interesting, it is not sufficient for functional rescue, as evidenced by our failure to detect differences in nerve conduction velocity. However, reduced nerve conduction velocity appears before other symptoms in PMP22 duplication carriers, often in early childhood, which begs the question if complete rescue of myelination could improve conduction velocity (38, 41, 42).

The repression of gene sets related to cholesterol biosynthesis and its regulators such as sterol response element binding protein (SREBP) in Sipa1l2−/− mice compared to wild types is worth noting given the importance of local cholesterol synthesis in Schwann cells during myelin growth (43–46). There is evidence that coordinate action between SREBP and EGR2 regulates peripheral nerve myelination (47, 48). Our finding that cholesterol biosynthesis genes, including some SOX10/EGR2 network genes, are repressed without persistent changes in Sox10 and Egr2 expression may suggest that either developmental changes in Egr2 are sufficient to repress cholesterol biosynthesis in adult mice, or that SIPA1L2 performs a regulatory function in cholesterol biosynthesis independent of EGR2. Interestingly, rare variants in SIPA1L2 have previously been associated with variable lipid response in humans, and rare variants in SREBF1, which encodes SREBP-1, have been implicated in Parkinson disease (49, 50). Though circumstantial, this may help explain why SIPA1L2 variants are associated with Parkinson disease in only some GWAS studies (13–18). It could be that SIPA1L2 variants are only associated with Parkinson subtypes that exhibit altered cholesterol metabolism and the prevalence of individual patient subtypes differed between studies (51). Repression of cholesterol biosynthesis has also previously been reported in the C3-PMP22 mouse model, which we detect in our analysis as well (52). An increase in pathway gene expression signature strength may explain why these same pathways are identified in C3-PMP22::Sipa1l2−/− mice, but without diverse lower enrichment score pathways that are also present in Sipa1l2−/− and C3-PMP22 mice. Alternatively, this may suggest that Sipa1l2 deletion attenuates these pathway endophenotypes in C3-PMP22 mice.

Our in vivo results support the putative interaction between SIPA1L2 and the severity of CMT1A phenotypes identified by patient GWAS (6). Further, they may indicate an interaction through the highly relevant cholesterol biosynthesis pathway, previously shown to transactivate SOX10/EGR2 pathway components (48). We did not, however, detect changes in many myelination-related genes of the SOX10/EGR2 network, or the transcription factors themselves, possibly due to developmental regulation or unknown activities of SIPA1L2. Though compelling, our neuromuscular phenotyping suggests that the effect of Sipa1l2 deletion is relatively modest in the C3-PMP22 mouse model, and that effects on myelination do not abide by a straightforward dose effect. It is possible that a larger effect may occur in other animal models of CMT1A. The failure to improve NCV while also decreasing body weight is disappointing from a potential therapeutic standpoint. Consistent with results from the patient GWAS (7), we conclude that the Sipa1l2 deletion modifies some CMT1A related phenotypes in mice but does not demonstrate a clear enough benefit to establish Sipa1l2 as a therapeutic target for CMT1A. Further study of Sipa1l2 may help untangle the complex relationship between cholesterol biosynthesis and myelination in the peripheral nervous system during disease.

FUNDING

This work was supported by National Institutes of Health (grant numbers R21 NS116936, R24 NS098523, and R37 NS054154 to R.W.B.). Patient sequencing was supported by National Institute of Neurological Disorders and Stroke U54 NS065712. The Scientific Services at The Jackson Laboratory are supported by National Cancer Institute (grant number CA34196). G.C.M. was supported by T32GM132006. T.J.H. was supported by funding from Uplifting Athletes’ Young Investigator Draft, the Charcot-Marie-Tooth Association, and NINDS K99 NS130151.

CONFLICT OF INTEREST

R.W.B. has served as a paid consultant for Roche in a capacity unrelated to this work. The other authors declare no conflicts of interest.

SUPPLEMENTARY DATA

Supplementary Data can be found at academic.oup.com/jnen.