Schwann Cell Transcript Biomarkers for Hereditary Neuropathy Skin Biopsies

Abstract

Objective

Charcot-Marie-Tooth disease (CMT) is most commonly caused by duplication of a chromosomal segment surrounding Peripheral Myelin Protein 22, or PMP22 gene, which is classified as CMT1A. Several candidate therapies reduce Pmp22 mRNA levels in CMT1A rodent models, but development of biomarkers for clinical trials in CMT1A is a challenge given its slow progression and the difficulty in obtaining nerve samples. Quantitative PCR measurements of PMP22 mRNA in dermal nerves were performed using skin biopsies in human clinical trials for CMT1A, but this approach did not show increased PMP22 mRNA in CMT1A patients compared to controls. One complicating factor is the variable amounts of Schwann cells (SC) in skin. The objective of the study was to develop a novel method for precise evaluation of PMP22 levels in skin biopsies that can discriminate CMT1A patients from controls.

Methods

We have developed methods to normalize PMP22 transcript levels to SC-specific genes that are not altered by CMT1A status. Several CMT1A-associated genes were assembled into a custom Nanostring panel to enable precise transcript measurements that can be normalized to variable Schwann cell content.

Results

The digital expression data from Nanostring analysis showed reproducible elevation of PMP22 levels in CMT1A vs. control skin biopsies, particularly after normalization to SC-specific genes.

Interpretation

This platform should be useful in clinical trials for CMT1A as a biomarker of target engagement that can be used to optimize dosing, and the same normalization framework is applicable to other types of CMT.

Introduction

Charcot Marie Tooth disease type 1A (CMT1A) is the most common inherited peripheral neuropathy affecting approximately 1:5,000 individuals¹. CMT1A is an autosomal dominant demyelinating neuropathy and is caused by a duplication of a 1.4 Mb segment on chromosome 17, containing the Peripheral Myelin Protein 22, PMP22, gene. Rodent models have shown that neuropathy is reversible upon reduction of PMP22 mRNA levels, as was shown in transgenic mice that over-expressed PMP22 via an inducible promoter². A recent study employing antisense oligonucleotides showed dramatic improvement in a severe mouse model of CMT1A as well as a more mild rat model of the disease³. While it remains to be seen if reversibility can be expected in CMT1A patients, therapeutic approaches are currently being developed to decrease PMP22 expression⁴.

Clinical outcome assessments such as the Rasch modified CMT Neuropathy or Exam Scores (CMTNS-R/CMTES-R) measure severity of the neuropathy in adults and children with CMT1A^{5, 6}. However, since neuropathy is slowly progressive, there is a clear need for sensitive biomarkers that measure disease progression and target engagement in patients. Recent studies have tested for biomarkers of disease impairment and/or progression in CMT1A. For example, MRI measurements of intramuscular fat accumulation (IMFA) fraction of calf muscles are a sensitive biomarker that increases over 12 months in patients with CMT1A^{7, 8}. In addition, neurofilament L protein (NEFL) levels are elevated in CMT1A plasma samples and correlate with severity as measured by the CMTNS-R and CMTES-R⁹.

Since several candidate therapeutic strategies are aimed to reduce PMP22 mRNA levels^{3, 10}, developing a robust method for establishing PMP22 mRNA levels from skin biopsies would be a valuable marker of target engagement for clinical trials in CMT1A. Dermal skin biopsies allow for the relatively non-invasive evaluation of myelinated nerve fibers innervating structures such as Meissner’s corpuscles¹¹. Immuno-EM showed elevated PMP22 protein in CMT1A patient skin samples compared to control, although the levels were more variable than in non-CMT1A samples ^{11, 12}. Quantitative PCR measurements of PMP22 RNA were employed in clinical trials for CMT1A ^{13, 14}, but these assays were unable to discriminate PMP22 levels in CMT1A patients from controls ¹⁵. However, in the antisense oligonucleotide study, the improvement of neuropathy was accompanied by a measurable reduction of Pmp22 by quantitative PCR of foot pad skin³. The loss of dermal nerve fibers in CMT1A patients^16–18 likely hampers the ability to establish a reliable assay, and Schwann cell numbers in skin can be affected by age and sex¹⁹. Therefore, it is important to develop optimal normalization criteria to address the inherent variability in skin biopsy analysis. In addition, several technologies are now available to digitally assess gene expression level, such as RNA-seq, as well as other methods with increased precision, including droplet digital PCR and Nanostring^20–23. For the relatively subtle differences in PMP22 levels between CMT1A and control Schwann cells (1.5 fold), the use of alternate technologies may present considerable advantages over more traditional methods.

Methods

Patient recruitment and consent

Ethics approval was obtained from University of Iowa and University of Wisconsin, and written informed assent/consent were provided by participants under a protocol approved by the ethics board of the NIH Rare Diseases Clinical Research Network (Protocol INC6611).

RNA-seq

Two 2 mm punch skin biopsies were performed 9 cm proximal to the ulnar crease and placed in RNAlater (ThermoFisher) prior to freezing as previously described ^{12, 13, 24–26}. Total RNA isolated from the samples were prepared for RNA-seq using NuGEN ovation library prep kits and the libraries were loaded on the Illumina HiSeq 2500 at the UW Biotechnology Center. Analysis of RNA-seq data was performed by the Bioinformatics Resource Center at Univ. of Wisconsin using methods that adhere to ENCODE guidelines and best practices for RNA-Seq (Encode Consortium, 2016). Illumina HiSeq reads were trimmed ²⁷ to remove sequencing adaptors and low quality bases. Reads shorter than 25 nucleotides after trimming were removed from downstream analysis. The trimmed sequences were aligned to the annotated (Ensembl 38.86) human reference genome using splice-junction aware alignment software, STAR²⁸.

Quantitative RT-PCR

RNA samples from skin biopsies (300ng) were converted to cDNA using an M-MLV reverse transcriptase kit (Invitrogen #28025–013), RNase inhibitor (Promega #N251A), and a 50 μM mix containing random hexamers and oligo dT primer. Subsequent to cDNA synthesis, samples were diluted 1/10 for use in quantitative PCR (qPCR) reactions. qPCR reaction mixes were prepared using 2X SYBR green master mix (Thermo Fisher #4367659) and 300 nM primers (final concentration), then aliquoted (13μl) onto a 384-well reaction plate. Each reaction received 2μl of diluted cDNA for a final volume of 15μl. qPCR was performed using the ViiA 7 platform (Thermo Fisher) under the following cycling conditions: Hold1 at 50°C for 2min; Hold 2 at 95°C for 10min; 40 cycles at 95°C for 15s and 60°C for 1min. Data were analyzed using the relative ΔΔCT method ²⁹.

PMP22 (total) F: GCTACAGTTCTGCCAGAGA R: CTCCTCCTGTTGCTGAGTATC

PMP22 (P1) F: GCTGCAAAGAAATCTGCTTGG R: GATGATACTCAGCAACAGGAGG

Gliomedin F: AACTGCCATTCAGAAGTGAGG R: GAGAGTCTGCTAACAAGAGTGA

Nanostring Analysis

The custom 54 gene Codeset was designed by Nanostring to maximize detection of most transcript variants, with the exception of the probes targeted to exons 1A and 1B of the PMP22 gene. SC-specific expression was also assessed using profiling of sorted cell types in mouse embryonic skin³⁰. Total RNA was extracted from skin biopsies by homogenization/purification with the Trizol reagent, and then further purified using the Qiagen RNeasy MinElute Cleanup Kit (Cat# 74204). The samples were assessed for concentration, purity, and quality by NanoDrop and Bioanalyzer (Agilent RNA6000 PicoChip). The nCounter gene expression assay was performed by Nanostring (Seattle, WA). Each RNA sample (100 ng) was added to the CodeSet in hybridization buffer and incubated at 65°C for 16 hours. The custom CodeSet consists of Reporter and Capture probes that hybridize to the target sequences of interest, forming a tripartite complex. After hybridization/washing, transcript counts were obtained using the MAX System Digital Analyzer. For most samples, the two punches were combined prior to purification of RNA, but for selected samples, the two punches were purified separately. The same reagent batches were used in 3 runs as follows: Round 1: 3 control, 21 CMT1A; Round 2: 3 control in duplicate (+1 carryover), 4 CMT1A in duplicate; Round 3: 5 control (+1 carryover), 7 CMT1A in duplicate.

Statistical analysis

Transcriptome alignments of RNA-seq data from STAR were used as RSEM (RNASeq by Expectation Maximization) input for expression estimation ³¹. Genes with low or undetectable expression were independently filtered by requiring at least 3 samples to have a minimum of 1 count per million reads sequenced (cpm>1). The un-normalized read counts from RSEM were provided to EdgeR ^32–34 for normalization and differential expression using the generalized linear model (GLM) method. Significance of the negative-binomial test was adjusted with a Benjamini-Hochberg FDR correction using a 5% threshold³⁵.

Skin biopsy Nanostring data were analyzed by nSolver Analysis Software v4.0. Raw counts were normalized to the geometric means of the specified combinations of normalization genes before assessing changes between CMT1A vs. control. P values were adjusted using the Benjamini-Yekutieli FDR correction method. Displayed normalizations used either two genes (PLLP, VGLL3), 5 genes (CADM4, CNP, GLDN, GPM6B, PLLP) or 6 genes (CUEDC2, ERBB3, GLDN, PLEKHA1, PLLP, VGLL3). Pearson correlation coefficients were calculated for each gene relative to neuropathy score (CMTES-R) to produce the correlogram using HMISC version 4.0 ³⁶ at a 95% (or 90%) confidence interval.

Results

RNA-seq analysis of Human Skin Biopsies

To improve mRNA analysis in skin biopsies, we attempted to identify changes in PMP22 and other Schwann cell genes by performing RNA-seq analysis of skin biopsies from nine CMT1A patients and seven controls. The punch biopsies were collected 9 cm from the ulnar crease, and Illumina sequencing and mapping to the hg38 genome yielded ~9–12 M mapped reads/sample. This analysis revealed relatively few genes that were found to be differentially expressed between groups with a FDR-adjusted p-value of <0.05 (Supplementary Table 1). Analysis of the most significantly induced genes in CMT1A samples only revealed one gene, SOX5, that is SC-specific by comparison to a gene expression database of sorted skin cell types ³⁰. This was somewhat expected given that Schwann cells are a minority component of skin. However, we analyzed these data sets to determine if changes in PMP22 and other SC-specific genes could be detected in CMT1A skin biopsy samples relative to controls. After normalizing the RNA-seq data to each sample’s read depth, the PMP22 levels (measured in FPKM) are ~1.55 fold higher in CMT1A samples than in control samples (Figure 1, p=0.016, t-test, not FDR corrected). The same difference was not observed for MBP (myelin basic protein). It is striking that using a more digital method of RNA detection—compared to quantitative PCR used in previous studies ¹⁵—revealed an elevated level of PMP22. However, levels were more variable in the CMT1A samples, which may reflect variable amounts of Schwann cells in CMT1A skin biopsies as described above.

Figure 1. RNA-seq analysis of CMT1A skin biopsies.

A) RNA was purified from skin biopsies from 9 CMT1A patients and 7 controls for RNA-seq analysis. After normalizing for read depth and identifying differentially expressed genes (See Supplementary Table 1), the FPKM’s (fragments per kilobase per million mapped fragments) are shown for PMP22 gene in a box and whisker plot (box shows 25%/75% values). PMP22 levels were ~1.5 fold higher in CMT1A patient samples compared to control skin biopsies.

B) FPKM levels of another major myelin protein, Myelin Basic Protein, are shown.

C) To normalize RNA-seq data for Schwann cell content, the FPKM ratio of PMP22 to gliomedin (GLDN, which encodes a nodal component) is shown for CMT1A and control skin biopsies. GLDN levels were more variable at this read depth since this gene is expressed at a lower level than PMP22.

D) Quantitative RT-PCR was used on a subset of the same samples (8 CMT1A, 5 control) to show the ratio of total PMP22 to GLDN in CMT1A vs. control skin biopsies. A second primer set for the Schwann cell-specific transcript of PMP22 (containing exon1A initiating from the P1 promoter) gave similar results when normalized to GLDN.

Because of the potential variations in Schwann cell content, we sought to identify ways to normalize the data using SC-specific genes. Several resources are available to identify SC-specific genes, including a) expression profiles of peripheral nerve development and response to injury ^37–44, b) profiling of sorted cell types in mouse embryonic skin ³⁰ and c) RNA-seq profiling of >400 human tibial nerve samples ⁴⁵ along with many other tissues (Broad Institute, gtexportal.org). These resources were used to filter the skin RNA-seq data to identify a series of SC-specific genes that are relatively unchanged in nerve injury and CMT1A models. In a trial analysis to normalize PMP22 levels to SC content, we used Gliomedin (GLDN), which is a SC-specific gene (Supplementary Figure 2) involved in the formation of nodes of Ranvier ^{46, 47}. Many Schwann cell genes associated with myelination change dramatically after peripheral nerve injury in rodent models ^{37, 42}, but Gldn expression does not change after injury. Interestingly, normalization to GLDN appeared to greatly decrease the variability in PMP22 levels in the control group. In addition, the normalized PMP22 level in the CMT1A samples is still more variable, yet statistically different from the control samples (p=0.02, t-test). The increased variability of the normalized PMP22 in CMT1A skin biopsies is similar to the relative variabilities of PMP22 protein observed in the immuno EM studies for CMT1A vs. control ¹². Nonetheless, our data suggest that employing more digital methods of analysis coupled with normalization to appropriate genes may create a more reliable and precise method for discriminating PMP22 levels in skin biopsies of CMT1A patients for clinical trials.

Validation of PMP22 Levels

The same samples were analyzed by qRT-PCR (Figure 1D), which showed a similarly increased level of PMP22 in the CMT1A samples after normalization to GLDN. While PMP22 is expressed at lower levels in most other cell types, we also measured levels of the Schwann cell-specific isoform containing exon 1A from the P1 promoter. While both exon 1A- and exon 1B-containing transcripts are developmentally induced in Schwann cells during the myelination period ⁴⁸, the exon 1A-containing transcript is exclusively expressed in Schwann cells, and accounts for ~50% of PMP22 in human peripheral nerve ⁴⁸, and this was corroborated by analysis of human tibial nerve data sets (gtexportal.org) provided by the Broad Institute ⁴⁵. Exon 1B-containing transcripts from the P2 promoter are the major form in fibroblasts ⁴⁹ and most other tissues ⁴⁵. Increased levels of the exon 1A-driven transcript of PMP22 was similarly elevated in CMT1A samples compared to controls after normalization to GLDN.

Nanostring Detection of SC-specific Transcripts in Skin Biopsies

With the ultimate goal of developing a clinically applicable platform for analysis of skin biopsies, we developed a custom gene expression panel using the Nanostring platform, which uses a barcode-based probe design to count RNA transcripts ²³. Since no amplification or cDNA synthesis is involved, the precision of the assay is substantially better than qPCR techniques, and it can detect transcripts that are present at <1 copy per cell. This would suggest that it would also be able to detect transcripts in the minor fraction of Schwann cells that are found in skin biopsies. Nanostring technology has been used in clinical trial biomarker assays ^{20, 21, 50}.

Our initial RNA-seq studies were used to assemble a pilot 52 gene panel of SC genes that are a) dysregulated in CMT1A rodent models, or b) SC-specific genes that are relatively stable and therefore will serve as normalization controls (Table 1). SC-specific expression was assessed using the aforementioned profiling of sorted cell types in mouse embryonic skin ³⁰ and RNA-seq datasets of >400 human tibial nerve samples ⁴⁵. The panel includes probes to detect total PMP22 and also the two major transcripts with alternate first exons (exons 1A and 1B, driven by P1 and P2 promoters, respectively). The set includes a number of known myelin genes (Mpz, Mag, Mbp, Pmp2, S100B), many of which are regulated by the SOX10/EGR2 transcriptional network that is required for Schwann cell differentiation ^{51, 52}. Most of the selected Schwann cell genes are not expressed in other skin cell types, although several are expressed in oligodendrocytes of the CNS, partly due to the common regulation by the SOX10 transcription factor ^{53, 54}. Several of the genes are mutated in other types of CMT (e.g. MPZ, PMP2, NDRG1, PRX, EGR2). We also included transcripts of Schwann cell injury genes, many of which are normally induced through the action of JUN in Schwann cells after demyelinating nerve injury ⁴⁰: NGFR, GDNF, OLIG1, SHH, BDNF, RUNX2, HMGA2, FGF5, CCL2, GFAP, and UCN2.

Table 1.

Selection of Nanostring Genes for Skin Biopsy Analysis

Myelin Genes
PMP22	MPZ
-exonla	MAG
-exonlb	MBP
CLDN19	PMP2
MAL	PLP1
CD9	CNTF
IL16	PRX
AATK	NDRG1
L1CAM	SOX10
UGT8	EGR2
GPM6A	S100B
GPR37L1	DHH
Injury Genes
GDNF	OLIG1
SHH	BDNF
RUNX2	HMGA2
FGF5	CCL2
GFAP	UCN2
TMPRSS5
CMT1A Genes
POU3F1	NGFR
CXCL14	ID2
EDNRB	SOX4
SOX5	CUEDC2
TTYH1	BZW2
PLEKHA1
Stable Genes
GPM6B	CNP
CADM4	PLLP
PLEKHB1	LGI4
NR4A2	VGLL3
ERBB2	ERBB3
GLDN

Table 1. Custom Nanostring panel design.

The indicated list of genes were submitted to Nanostring for capture/detection probe design. Indicated categories are major myelin genes, genes induced in Schwann cells after nerve injury, genes shown to be induced rodent CMT1A models, and also Schwann cell genes that are relatively unchanged in nerve injury and/or neuropathy model profiles. Most genes were chosen based on evidence that their expression was Schwann cell-specific using published profiles of sorted cells and/or specific to human tibial nerve. Many peripheral nerve genes are also expressed in oligodendrocytes in the CNS. The complete list of genes and Refseq accession number used for probe design is in Supplemental Table 1.

As shown in Table 1, we incorporated additional candidate biomarkers that robustly reflect PMP22 expression levels and their effect on Schwann cell function. In the recent study employing antisense oligonucleotides (ASO), RNA-seq analysis of the mouse model revealed several hundred differentially expressed genes (DEG) compared to nerves from wild type animals³. Interestingly, ASO treatment to reduce PMP22 mRNA in the C22 mice resulted in a significant normalization of only a subset of DEG, suggesting that there are “rapid response” genes that are sensitive to restoration of normal levels of PMP22. Many of these genes correlated with disease severity as defined by electrophysiology in the mice ³. We prioritized those that are Schwann cell-specific as described above, including BZW2, POU3F1, NGFR, CXCL14, ID2, EDNRB, SOX4, CUEDC2, TTYH1 and PLEKHA1.

Finally, the panel was designed to normalize PMP22 levels to those of other Schwann cell specific genes (e.g. Gliomedin) that are not affected by CMT1A status, and we identified several candidate genes that are not apparently modulated in rodent models of CMT ^{3, 55}, but nonetheless are SC-specific. In addition to gliomedin, we selected several such genes that are preferentially expressed in Schwann cells relative to other skin cell types: DHH, GPM6B, CNP, CADM4, PLLP, PLEKHB1, LGI4, NR4A2, VGLL3, and ERBB3.

Optimizing Detection of Differential PMP22 Levels in CMT1A Skin Samples vs. Controls

For the initial Nanostring run, we obtained a set of 21 CMT1A skin biopsies (ages 23–68, median 50) and 4 controls. Although the number of controls was relatively low, our previous experience using RNA-seq suggested that PMP22 levels are relatively consistent among controls. The resulting data (Figure 2A) showed that the CMT1A skin biopsies had elevated levels of total PMP22 and its SC-specific exon 1A containing transcript relative to controls. The elevation was observed with both raw count data as well as after normalization to GLDN. Consistent with previous analysis of RNA-seq data, the normalization to GLDN appeared to reduce the variability among control samples, although there was still enhanced variability of the CMT1A samples relative to controls.

Figure 2. Nanostring analysis of CMT1A skin biopsies.

A) Equal amounts of skin biopsy RNA’s were applied to the custom Nanostring panel, and the plots show raw counts of total PMP22 as well as the Schwann cell-specific Exon 1A form of PMP22. In addition, the ratio of both forms to the Schwann cell-specific GLDN is shown. (21 CMT1A and 4 control, avg. age 48.2 and 51.2 yr, respectively)

B) The volcano plot shows each of the genes (dots) ranked with the most statistically significant changes between control and CMT1A skin biopsies in the upper right quadrant. The results are normalized to the geometric mean of six SC-specific normalizing genes including GLDN using Nanostring nSolver software, and the plot shows levels (log 2, x-axis) plotted against the significance of the changes (negative log10 p value) between control and CMT1A. The plot shows elevation of total PMP22 in CMT1A samples along with specific probes for its two major transcripts (Ex 1A and Ex 1B), and also increased levels of NGFR and L1CAM. (21 CMT1A, 4 controls).

After normalizing to six SC-specific genes including GLDN, the Nanostring panel results are shown as a volcano plot in Figure 2B, in which the most statistically significant increases in CMT1A biopsies relative to controls are shown in the upper right quadrant. Consistent with our transcript data, we saw an ~2-fold increase in PMP22 levels (and its two major transcripts in Schwann cells) in CMT1A vs. control, which were statistically significant even after FDR correction of the p-values (adj. p<0.01). In addition, we also saw elevated levels of L1CAM, which is a potentially novel biomarker of CMT1A, as well as increased levels of two treatment responsive SC-specific genes from the recent ASO study ³: SOX4 and NGFR/p75.

To provide a more robust set for testing of the panel, we obtained additional CMT1A and control skin samples for a total of 32 cases and 11 controls. The distribution of age/sex, and neuropathy score across the sample groups is shown in Table 2. The overall data set was normalized to different combinations of SC-specific genes, and we found that several independent normalization methods highlighted total PMP22 (and the exon1A/1B transcripts) as being the most significant changes in the CMT1A samples relative to controls (see volcano plots in Figure 3). The other most consistently elevated gene across different normalizations was L1CAM.

Table 2.

Distribution of Age/Sex and Neuropathy Score Across Sample Groups

	Age (range)	Sex	CMTES-R (range)
control	42.8 (25–58)	6 F, 5M	NA
CMT1A	46.4 (17–70)	18F, 14 M	11 (0–19)

CMTES-R = Rasch modified CMT Exam Score; NA = not applicable.

Figure 3. Nanostring analysis with alternate normalization methods.

Nanostring results from skin biopsies are shown for 11 controls and 32 cases, from three separate runs. The volcano plot shows levels (log 2, x-axis) plotted against the significance of the changes (negative log10 p value) between control and CMT1A. The results are normalized to the geometric mean of two Schwann cell specific genes (A), 5 SC-specific genes (B) or six SC-specific normalizing genes (C).

Analyzing Variability of PMP22 levels in CMT1A Skin Samples

As noted above, the increased variability among CMT1A samples¹² could arise from inherent variability of sample collection sites within a given subject. Since the samples were originally pooled from two separate skin punches, we also purified each skin punch independently for some selected cases and controls. For these samples, we could use them to assess intrasubject variability of skin biopsy analysis as well as test if our normalization methods are able to account for differential Schwann cell content of the two skin punches. As shown in Figure 4, there was fairly poor correlation between raw counts of duplicate samples. However, after normalization to SC-specific genes, the correlation was much improved. This demonstrates the importance of normalization to SC content and shows that the apparent variability of normalized PMP22 levels among CMT1A patients is not due to sampling variability in skin biopsies.

Figure 4. Nanostring Analysis of PMP22 levels in duplicate skin biopsies.

For selected control and CMT1A cases, the two skin punches were processed independently and submitted for Nanostring analysis in order to test the intra-subject variability of the results. The correlation plots show the PMP22 values for the two samples from each subject using A) raw counts or B-D) using several alternate normalization methods employing different combinations of Schwann cell-specific genes. The line shows the best fit of the data, and the r² value is indicated. The circled points show those that were obtained from control subjects, which have consistently lower levels of PMP22 than the remaining CMT1A cases.

The normalized data were tested for correlations at a 95% confidence interval between individual genes and the CMTES-R neuropathy score and age (Figure 5). There was no correlation between PMP22 level and neuropathy score, consistent with previous immuno-EM data ^{11, 12}. However, we did observe an inverse correlation between the neuropathy score and the levels of EGR2 and NDRG1. Both genes are mutated in other forms of CMT, and in fact EGR2 is a transcription factor that controls the level of NDRG1 ⁵⁶. The NDRG1 and EGR2 correlations with CMTES-R were also observed with the alternately normalized data sets that were used in Figure 3. The correlation analysis performed at a 90% confidence interval (not shown) revealed an additional positive correlation between neuropathy score and the TTYH1 gene. In the C22 CMT1A model, TTYH1 responded to treatment with antisense oligonucleotides for PMP22 ³. Finally, there is an inverse correlation between age and the EGR2, NDRG1, CCL2, and CADM4 genes at the 95% confidence interval, consistent with the known age-dependent progression of CMT1A.

Figure 5. Correlation analysis of CMTES-R with gene expression levels.

Nanostring gene expression levels normalized to PLLP/VGLL3 were correlated with CMTES-R to determine significant correlations. Correlation values using Pearson’s method are represented in the correlogram. The strength of the correlation is indicated by the heat map scale for ellipses, showing positive (red, right leaning), and negative correlations (blue, left leaning). The row labeled as CMTES-R shows two inverse correlations with EGR2 and NDRG1, which are significant at a 95% confidence interval. There was no significant correlation of CMTES-R with PMP22 levels.

Discussion

Development of biomarkers for clinical trials in CMT1A is a major challenge given the slowly progressive nature of the disease. Biomarkers can be used to not only measure disease process (NEFL levels) and disease burden (MRI) but also target engagement for a given therapy ⁵⁷. NEFL assays in CMT has provided an important marker of axonal degeneration in CMT1A ⁹, but there is currently a lack of biomarkers to assess disease processes in the Schwann cells that are most directly affected by the PMP22 gene duplication. Several candidate therapies reduce PMP22 levels in CMT1A rodent models and thereby ameliorate the symptoms of PMP22 overexpression ^{3, 10, 58}. However, assessing PMP22 mRNA reduction in human trials is problematic given that such analysis has not demonstrated clear elevation of PMP22 mRNA in CMT1A patients compared to controls ¹⁵.

The analysis of skin biopsies is hampered by variable amounts of Schwann cells (SC) in skin, as well as the variable amount of PMP22 in SC as previously established by immuno-EM in CMT1A skin biopsies ^{11, 12, 16, 17}. Therefore, it is important to develop optimal normalization criteria to address the variability inherent in skin biopsy analysis. Previous RT-PCR studies of human skin biopsies have used SC-specific genes (S100B or MBP) or ubiquitous genes (GAPDH, SDHA, TBP) for normalization ^{13, 15, 59}. The ideal normalization gene would be Schwann cell-specific and be unaffected by injury or CMT1A status. Therefore, we employed data sets from CMT1A rodent models ^{3, 55, 60} that have been filtered to identify SC-specific genes ³⁰ so that gene expression analysis can focus on the Schwann cell compartment of skin.

Another important consideration in assay development is the use of more digital gene expression platforms. The theoretical increase of PMP22 levels is 1.5 for CMT1A, and the precision of quantitative PCR is limited in this range ⁶¹. This is a particularly important point as the precision must be adequate for measurement of both PMP22 and normalization genes. Compared to hybridization-based microarrays, RNA-seq has much improved dynamic range as well as better precision particularly for less abundant transcripts. The pilot RNA-seq analysis of skin biopsies showed that PMP22 levels were ~1.5 fold higher in CMT1A patient samples compared to control skin biopsies after normalization to read depth, indicating that a precise method of gene expression analysis combined with robust normalization to Schwann cell-specific genes could provide a much improved technology. To construct such an assay, we turned to another digital gene expression platform based on Nanostring technology, since the performance/analysis costs of RNA-seq analysis are substantial. This platform directly detects RNA transcripts without cDNA synthesis and amplification, and its sensitivity has obvious advantages for development of a robust analysis platform for skin biopsies^{20, 21, 50}.

Using various combinations of SC-specific genes for normalization, the custom Nanostring platform was able to reduce the apparent variability and optimize the differential levels of PMP22 between CMT1A and control skin biopsy samples. We obtained similar results with probes designed for total PMP22 as well as the two major Schwann cell transcripts with alternate first exons (1A and 1B). Nanostring data were analyzed with several independent normalization methods, all of which have been able to distinguish PMP22 levels in CMT1A vs. control skin biopsies. Based on our data, we suggest that this platform could be a valuable tool in clinical trials for CMT1A as a measure of target engagement that could be used within several weeks. In addition, this platform could also be used to establish optimal dosing, since over-suppression of PMP22 would have the risk of inducing another neuropathy phenotype similar to HNPP. In addition, while it would seem optimal to target a ~30% reduction of PMP22 levels in CMT1A to bring it to the normal level, the significance of the results suggests that it would be possible to detect more subtle changes in PMP22 levels (e.g. 10–20%), which could lead to improvement in the neuropathy, although perhaps a somewhat larger patient sample could be required to establish the significance of the effect. The normalization framework may also be applicable to other types of CMT, although it will need to be validated in a larger sample set.

The increased variability of PMP22 measurements found previously in CMT1A skin samples ¹² was also evident in our study, and it has been proposed that other mechanisms besides elevated PMP22 mRNA per se are triggers for the neuropathy ^{15, 62}. To test if increased variability could be due to sampling variability, we processed duplicate skin biopsies from the same patient and found very good concordance between them after normalization to SC-specific genes. Therefore, the range of PMP22 levels in CMT1A patients appears to reflect true variability between cases rather than variability between sites of biopsy.

In addition to measuring PMP22, we incorporated analysis of other SC-specific genes that may reflect the dysregulation of CMT1A Schwann cells. Although there are hundreds of altered genes in rodent models of CMT1A ³, the most effective biomarkers will be those that respond quickly to normalization of PMP22 levels. The recent identification of an effective PMP22 ASO provided an opportunity to rank dysregulated genes that are most responsive to therapeutic reduction of PMP22 expression levels. RNA-seq analysis of sciatic nerve from the C22 mouse model of CMT1A identified a small subset of 76 differentially expressed genes between wild type and mutant nerves that returned towards wild type levels following treatment with ASO to reduce PMP22 mRNA levels³. These genes were mined to identify those that are SC-specific and several of these were included in the Nanostring panel. Of the candidate markers included in the panel, the most promising were L1CAM and NGFR, which were found to be increased in CMT1A skin. In rodent studies, Ngfr and L1cam are elevated in CMT1A models^{3, 63, 64}. In normal nerve, both genes are normally expressed in nonmyelinating Schwann cells, but both L1CAM and NGFR are expressed in myelinating Schwann cells of human skin¹⁹. It is anticipated that reduction of PMP22 in a clinical trial could be assessed relatively early through this method, and normalization of PMP22 levels and restoration of Schwann cell homeostasis would also hopefully reduce the levels of L1CAM and/or NGFR. It is notable that most of the Schwann cell injury genes were not elevated in CMT1A. The JUN transcription factor is a driver of demyelination after injury ⁴⁰, and while it is detected in CMT1A skin biopsies ⁶⁵, its levels may not be sufficient to drive an injury program. JUN was not included in this panel since it is not specific to Schwann cells in skin, but these and other data ³ suggest that CMT1A Schwann cells are distinct from injury-associated Schwann cells. Indeed deletion of Jun from a CMT1A mouse model does not mitigate neuropathy but rather exacerbates it ⁶⁶.

Using the CMTES-R score, we did not see correlation of PMP22 mRNA levels (or NGFR/L1CAM) with the severity of the neuropathy. This was expected given that impairment in patients correlates more with axonal degeneration than demyelination ⁶⁷. In addition, the distribution of PMP22 protein levels as measured by immuno-EM in CMT1A skin biopsies did not correlate with the severity of the neuropathy¹². Nonetheless, we did observe inverse correlations of neuropathy score with EGR2 and NDRG1. EGR2 is a transcription factor that activates many myelin genes during Schwann cell development and itself is mutated in CMT1D and some more severe peripheral myelinopathies^{51, 56, 68, 69}. NDRG1 is also mutated in a recessive form of CMT (CMT4E). While not described before, it could be that progression of CMT is due in part to decreased activation of the EGR2-driven transcriptional network, which includes NDRG1, as continued expression of EGR2 is required for myelin maintenance⁷⁰. Alternatively, the association may reflect the loss of myelinated fibers that occurs during progression, although one would have expected a commensurate loss of MPZ expression if that was the explanation. The correlation analysis performed at a 90% confidence interval also revealed a correlation of CMTES-R with TTYH1, which was one of the ASO-responsive genes in the recent treatment study of the C22 mouse model of CMT1A³.

Recent studies have employed a quantitative PCR analysis of human skin biomarkers based on gene expression studies of skin in rat model of CMT1A^{59, 71}. While this approach has been successful in measuring progression of CMT1A patients over a 2–3 year trial period, this analysis was not designed to detect SC-specific genes and PMP22 levels were also not incorporated. Therefore, some of the changes may occur in other cell types in response to CMT1A-associated denervation, and therefore may not be as immediately responsive to successful treatment. Our studies provide a complementary framework for gene expression analysis targeted to the Schwann cell compartment of skin biopsies, and provide an assay platform with enhanced precision to measure PMP22 levels that is designed to detect early changes in PMP22 and other CMT1A-associated genes. Further optimization of the Nanostring panel may enhance the ability to detect changes in CMT1A gene signatures.

Figure 6. CMTES-R neuropathy scores correlate with EGR2, NDRG1, and TTYH1.

Regression models (scatterplots) of significantly correlated genes to CMTES-R are shown, including the linear regression function. The correlation with TTYH1 was found only at the 90% confidence interval A) NDRG1, B) EGR2, and C) TTYH1