Biomarkers predict outcome in Charcot Marie Tooth Disease 1A

Abstract

Background

Charcot Marie Tooth disease type 1A (CMT1A) is the most common inherited neuropathy, a debilitating disease without known cure. Among CMT1A patients, disease manifestation, progression and severity are strikingly variable, which poses major challenges for the development of new therapies. Hence, there is a strong need for sensitive outcome measures such as disease and progression biomarkers, which would add powerful tools to monitor therapeutic effects in CMT1A.

Methods

We established a pan-European and American consortium comprising 9 clinical centers including 311 patients with CMT1A in total. From all patients, the CMT neuropathy score (CMTNS) and secondary outcome measures were obtained and a skin biopsy collected. In order to assess and validate disease severity and progression biomarkers, we performed qPCR on a set of 16 animal model derived potential biomarkers in skin biopsy mRNA extracts.

Results

In 266 CMT1A patients, a cluster of eight cutaneous transcripts differentiates disease severity with a sensitivity and specificity of 90% and 76.1%, respectively. In an additional cohort of 45 CMT1A patients, from whom a second skin biopsy was taken after two to three years, the cutaneous mRNA expression of GSTT2, CTSA, PPARG, CDA, ENPP1 and NRG1-I is changing over time and correlates with disease progression.

Conclusions

In summary, we provide evidence that cutaneous transcripts in patients with CMT1A serve as disease severity and progression biomarkers and, if implemented into clinical trials, they could markedly accelerate the development of a therapy for CMT1A.

Introduction

Charcot-Marie-Tooth (CMT) diseases or hereditary motor and sensory neuropathies (HMSN) comprise a family of the most common inherited disorders of the peripheral nervous system with a prevalence of up to 1 in 1214 (1,2). So far, over 1000 different mutations have been discovered in more than 90 genes linked to hereditary neuropathy (3). In >60% of all cases, the genetic defect underlying CMT is duplication of the gene encoding the 22kDa peripheral myelin protein (PMP22) (4–9). Affected individuals develop a slowly progressive demyelinating neuropathy with distally pronounced muscle atrophy and sensory impairment, resulting in steppage gait, altered deep tendon reflexes and skeletal deformities (e.g. pes cavus)(10). CMT1A typically manifests within the first two decades of life, and walking disabilities, foot deformities and electrophysiological abnormalities in most patients present already in childhood (11,12). However, CMT1A disease onset, progression and severity are strikingly variable within families and even among monozygotic twins (13,14). Despite several promising trials in animal models, no causal treatment is available for any form of CMT yet (15,16). Neither trials of exercise and orthosis, nor pharmacological approaches with ganglioside, creatine and, more recently, oral administration of ascorbic acid showed beneficial effects in patients with CMT subtype 1A (17–21). For ascorbic acid, it was suggested that a treatment effect may have been missed because of an unexpectedly slow disease progression (21). Hence, insensitive outcome measures harbor the risk to mask therapeutic effects in clinical trials (21,22). Improvement of outcome measures is therefore essential to facilitate the development of a treatment for CMT1A disease (23,24). To date, only physical and electrophysiological examinations are available to determine disease severity in patients with CMT. The CMT neuropathy score (CMTNS) is a valid and reliable nine-item composite scale taking into account sensory and motor symptoms (25). In order to standardize patient assessment and to improve the scale’s sensitivity to change, a novel version of the CMTNS has been proposed (23,26). Surrogate biomarkers of disease severity and progression would add powerful tools to monitor therapeutic effects in clinical trials (15,21,27–29). Recently, non-invasive magnetic resonance imaging study of muscle water changes and intramuscular fat accumulation in the lower limbs were shown to be correlated to clinical impairment (30,31). While promising as a surrogate outcome measure for clinical trials, this approach is technically demanding and has so far only been tested in proof of principle in a small group of patients suffering from CMT and various other neuromuscular disorders. Molecular analysis of body fluids or tissue biopsies is a promising strategy for the development of clinically more practical surrogate biomarkers for disease severity in CMT1A.

The development of valid biomarker assays, however, requires a fit-for-purpose approach (32,33), that can be separated into four iterative phases: (I) The pre-validation phase defines the intended purpose of the biomarker, considering pre-analytical variables and bioanalytical method feasibility. (II) The exploratory validation phase assesses the basic assay performance. (III) The advanced validation phase characterizes the formal performance of the assay with regard to its intended use. (IV) The in-study validation phase that ensures that the assay method performs robustly across studies according to predefined specification and facilitates the establishment of definitive acceptance criteria (32,33).

Within an exploratory pre-validation phase (fit-for purpose step I), employing a Pmp22 transgenic rat model for CMT1A (CMT rat, (34)), we could previously show that transcriptional profiling in skin biopsies can be utilized to identify prognostic and disease severity biomarkers which correlate with clinical impairment (16). Importantly, in a translational proof of concept approach, we could validate rodent cutaneous mRNA biomarkers in skin biopsies from 46 patients with CMT1A (16). Building on these preliminary results, here we aimed at the exploratory validation (fit-for-purpose step II) of 16 CMT rat derived potential transcriptional biomarkers within a large pan-European and American consortium providing skin biopsies from 266 clinically well characterized genetically proven patients with CMT1A. Next to the assessment of clinical impairment, the ability to measure disease progression is an indispensable prerequisite for the successful conduction of therapeutic trials. Given the poor detectability of clinical impairment over time (21), disease progression biomarkers could markedly accelerate the development of a therapy. Hence, we aimed on the advanced validation phase of potential biomarkers with regard to disease progression (fit-for-purpose step III) and collected 45 paired skin biopsies from patients with CMT1A that were re-assessed after 2–3 years.

Methods

Patient recruitment

All clinical centers involved obtained prior ethical approval of their local institutional review board and their respective regional ethics committee. An overall description of the patient recruitment and a detailed listing of all assessed outcome measures outcome measures can be found in the supplemental material

Skin biopsy

Skin biopsies with a diameter of 3 mm were taken from the proximal, medial part of the index finger of the non-dominant hand. Sterile disposables of 3 mm diameter, “Biopsy Punch” by Stiefel were used. The skin biopsy was split into two equal parts, one placed in 4% paraformaldehyde in 1 phosphate-buffered saline, the other part was stored in RNAlater (Ambion).

RNA Isolation, Precipitation and cDNA Synthesis

For the gene expression measurements the total RNA was extracted from one half of the skin biopsies (in RNA later). The sample was homogenized in RLT buffer (Qiagen) using a rotor stator and processed using the Qiagen manufacture protocol for non-fatty tissue (RNeasy Mini Kit, Qiagen). Before cDNA synthesis the RNA was precipitated by adding 0.5 volumes ammoniumacetate (7M, RNA grade), 2μl pellet paint (EMD Millipore) and 2.5 volumes pure ethanol (RNA grade) and resuspended in RNAse free. The preparation and processing of all samples were performed in parallel by the same person using the same protocol and the quality of RNA was verified by integrity check (Agilent). Only samples with a RNA integrity number (RIN) higher than 7 were used for further analyses. The cDNA synthesis was carried out using the Superscript III RT kit (invitrogen). To reversely transcribe mRNA we used dT Primer (0.6 μM) and random nonamer primers (N9 Primers 120 μM) in a single polymerase chain reaction step. The obtained cDNA was diluted 1 to 10 before qPCR analyses.

Real-time semiquantative PCR with TaqMan and SYBRgreen

The qPCR was operated in the LightCycler 480 Systems (384 well format, Roche Applied Science) using 2μl of diluted cDNA from skin biopsies in each reaction. For all genes except NRG1-I, TaqMan qPCR assays were performed using the manufacturers protocol (TaqMan Universal PCR Master Mix, Applied Biosystems). Intron spanning primers (0.9 μM, each) and FAM-TAMRA tagged probes (0.25 μM) were designed by Microsynth (Switzerland) for each assay (see sequences in supplemental material). For NRG1-I no primer probe set could be designed and we designed primers amplifying the type I specific 5’ region of the NRG1 gene using Primer Express Software V.1.65 (Applied Biosystems, see sequences in supplemental material). For NRG1-I we performed qPCR using the SYBRGreen PCR master mix (Applied Biosystems).

For all qPCR assays, the reaction mix was prepared to the final volume of 10μl and all reactions were performed in four replicates. Due to the high number of samples, six 384 well plates for each assay were used and the plates were calibrated to each other by analysis of three calibrator samples on each plate for each gene. For quantification of expression levels, the threshold cycles (Ct) for each gene of interest was normalized against two stable housekeeping genes (B2M and RPLP0).

Data analysis and Statistics

Four technical replicates were available for each qPCR measurement which were summarized by their median for further analysis. The difference to the mean expression of the two housekeeper genes B2M and RPLP0 was calculated per patient. To remain on the Ct scale, for each plate the median of all housekeeper gene measurements was added. Three reference patients were quantified on each qPCR plate and their mean expression was subtracted to normalize for plate effects. Again, to remain on Ct scale, the median of all reference patient measurements was added. The resulting normalized expression data are ΔΔCt values rescaled to a Ct comparable scale.

The secondary clinical parameters were transformed into z scores using mean and standard deviation (SD) of healthy controls of corresponding age class, and gender. The z scores were further categorized into one of the levels normal, very mild, mild, moderate, severe (0,1,2,3,4).

Cohort specific mean and SD were calculated for age, BMI, all primary and secondary clinical parameters as well as different CMT scores.

The normalized Ct values from the PCR experiment were used for correlation based hierarchical clustering, where patients were clustered on the correlation of their expression profiles across the potential biomarkers. Similarly, the potential biomarkers were clustered based on the correlation of the expression profiles across patients, where both, positive and negative correlation, were interpreted as small distance. Additionally, the potential biomarkers were clustered by their pairwise correlations.

Supervised principal component analysis (PCA) was performed on the PCR data. To account for potential center and plate effects, linear mixed effects models were fit to the expression data controlling for CMTNS and estimating coefficients for the genes, the plates and the centers. The data was then corrected by the estimated plate and center effects. CMTNS values were genewise regressed onto the expression values and the expression values were scaled by the resulting coefficient so that the rescaled expression values were in CMTNS unit scale. A canonical PCA was performed on these data. CMT patients were classified as mild (CMTNS < 10), moderate (CMTNS between 10 and 20) or severe (CMTNS > 20). A support vector machine (SVM) classifier separating mild and severe CMT cases based solely on gene expression profiles was trained and evaluated using a 10 times repeated 10-fold cross validation.

The association between CMT scores, as well as primary and secondary sub-scores with the expression of the potential biomarkers was assessed by fitting linear models to the expression values with the score as predictor while controlling for center, plate, age, gender, and BMI.

Using the data from patients with measurements in two points in time linear mixed effect models were fit to the CMTNS quantifying the progression effect while controlling for BMI, gender, and age effects. Similar mixed effects repeated measures regression models were fit per gene to the expression data, where the effect of time was assessed controlling for BMI, gender, age, and center effects.

A predictive random forest model that classifies patients into progressive patients (CMTNS score increase) and non-progressive patients (CMTNS score does not increase) was trained on the genes with significant change in expression over time. Prediction performance was evaluated in a 10 times repeated 10-fold cross validation.

Gene set enrichment analysis was performed by the web based gene list enrichment analysis tool Enrichr (http://amp.pharm.mssm.edu/Enrichr/, (35))

Results

266 plus 45 CMT1A patients were examined in 9 centers around the world (95 from Germany, 56 from the Czech Republic, 50 from Spain, 26 from Belgium, 20 from the UK, 10 from Italy and 9 from France were examined once for the testing of disease severity biomarkers, and 16 from Germany, 14 from Italy and 15 from the USA were examined twice for disease progression biomarkers) (Fig. 1A). From all patients the CMTNS and related outcome measures were obtained and the patients were characterized according to standardized operating procedures (see methods) and the patients were further characterized for their gender, age and BMI (Tab. 1). The overall patient cohort (n=266) splits into 155 (58%) female and 111 (42%) male CMT1A patients, with a mean age of 43 years (SD 14, range 18–80). The CMTNS ranged from 4 to 29 with a median value of 14 (Tab.1). Importantly, the patient cohorts from the different centers display significant differences in a majority of the obtained parameters (Tab. 1). This inter-center variability may partially be explained by regional differences in the composition of the patient cohort. However, small cohort sizes in some centers or differences in routine clinical examination and assessment of the CMTNS, despite standardized operating procedures, are more likely to account for the inter-center variability. Indeed, e.g. the very disparate scores for the dynamometrical secondary outcome measures are in favor for this notion (Tab. 1). Therefore, to avoid disruptive influences of center effects in the clinical parameters on the biomarker analyses, the subsequent expression data analysis was controlled for the co-variables center, age, gender and BMI (Tab. 2).

Figure 1: Previously identified potential cutaneous mRNA biomarkers tested in skin biopsies of 266 patients with CMT1A.

A Participating centers in Europe and the USA with number of assessed patients and contributed skin biopsies for biomarker analyses (see Table 1 for further details). Next to 266 patients from Europe that were assessed once (in blue) additional 45 patients from Europe and the USA were sampled twice within 2–3 years giving information on the progression (in green). 37 healthy humans were included as controls from two centers in Germany (grey). B Heatmap displaying normalized Ct values from the qPCR analyses of all patients (columns) for all genes (rows). High values (in blue) correspond to low expression while low values (in red) indicate high expression. Both dimensions, patients and genes, are re-ordered by means of correlation based hierarchical clustering to group by expression profile similarities. The dendrogram on the top shows the clustering of patients, the dendrogram on the left shows the clustering of the genes. The biggest cluster of eight similarly regulated genes is highlighted in magenta. The two rows at the top show the two housekeeper genes for reference.

C Correlation of the gene expression profiles across the patients. Both, rows and columns show genes, the upper triangle shows the numerical values of the correlation coefficients the lower triangle visualizes the correlations coefficients where blue colored circles represent negative correlations and red colored circles represent positive correlations. The size of the circles indicates the strength of correlation. The genes are clustered by their correlation profile. Eight out of 16 genes cluster with correlation coefficients >0.7 (highlighted in magenta). Most notably, there is no negative correlation among the genes, except for the perfectly negatively correlated housekeeper genes.

Table 1: Patient characteristics.

This table holds descriptive statistics about the main clinical parameters of all the patients in the study cohort. The statistics are computed separately within each contributing clinical center. Grey: basic characteristics; white: CMT scores; green: subscores to the CMTNS; orange: secondary parameters.

parameter	level	Czech Republic	France	Italy	Belgium	Spain	UK	Germany/Munich	Germany/Gottingen	p value
n		56	9	10	26	50	20	40	55
Gender	female	37(66.1%)	5 (55.6%)	6 (60.0%)	17 (65.4%)	33 (66.0%)	−75.00%	15 (37.5%)	27 (49.1%)	0.04
	male	19 (33.9%)	4 (44.4%)	4 (40.0%)	9 (34.6%)	17 (34.0%)	5 (25.0%)	25 (62.5%)	28 (50.9%)
Age at Examination [years]	mean ± sd	40 ±12	46 ±18	54 ±16	51 ±16	39 ± 14	46 ±12	44 ±14	43 ±12	0.01
BMI [kg/(m2)]	mean ± sd	25 ±5.1	25 ± 4.8	24 ±3.1		25 ±4.4	27 ±5.6	24 ± 4.8	26 ±5.5	0.32
CMTNS [0/.../36]	mean ± sd	14 ±4.6	13 ±4.5	13 ±7.5	17 ±4.2	12 ±4.3	16 ±4.9	14 ±5.1	14 ±5	<0.01
CMTNSJull [0/.../60]	mean ± sd	27 ±8.4	22 ±8.7	25 ±11	28 ± 8.4	19 ± 6.7	36 ±11	24 ±8.4	26 ±6.6	<0.01
CMTNS_sig [0/.../48]	mean ± sd	22 ±6.2	20 ±6.3	20 ± 9.4	24 ± 6.4	16 ±5.5	25 ±6.2	20 ±6.7	22 ±5	<0.01
CMTNS_mod 0/.../32]	mean ± sd	14 ±4.6	12 ±5.1	12 ±5.6	16 ± 5.4	8.1 ±4.2	16 ±4.5	13 ±5.1	16 ±3.8	<0.01
Sensory Symptoms [0/1/2/3/4]	mean ± sd	1.1 ±1.1	0.78 ±0.83	0.56 ±0.73	1.3 ±1	0.58 ±1.1	2.1 ±0.64	1.3 ±0.99	1.7 ±0.83	<0.01
Motor Symptoms Legs [0/1/2/3/4]	mean ± sd	1.2 ±0.9	1.1 ±0.78	1.4 ±1.1	2 ±1.2	0.84 ±0.68	1.6 ±0.82	1.6 ± 1.1	1.3 ±0.96	<0.01
Motor Symptoms Arms [0/1/2/3/4]	mean ± sd	0.57 ±0.57	0.67 ±0.71	0.56 ±0.73	1.1 ±0.56	0.38 ±0.64	1.1 ±0.51	0.62 ±0.71	0.81 ±0.74	<0.01
Pin Sensibility [0/1/2/3/4]	mean ± sd	1.4 ±0.85	1.7 ±1	1.1 ±1.3	1.7 ±0.88	0.74 ±1.1	1.6 ±0.81	1.1 ±0.81	1.4 ±0.91	<0.01
Vibration [0/1/2/3/4]	mean ± sd	2.4 ±0.88	1.3 ±0.5	1.8 ±1.4	2.8 ±0.51	3.7 ±1	1.4 ±0.69	1.7 ±0.86	2 ± 1.2	<0.01
Strength of Legs[0/1/2/3/4]	mean ± sd	1.4 ±0.87	1.3 ±1	1.1 ±1.3	1.4 ±1	1.4 ±1.1	1.9 ±1	1.7 ±1.2	1.4 ±0.99	0.32
Strength of Arms [0/1/2/3/4]	mean ± sd	0.79 ±0.71	1 ± 0.87	1.2 ±1.3	0.81 ±0.69	0.2 ±0.4	1.3 ±0.86	0.67 ±0.87	0.48 ±0.78	<0.01
Ulnar CMAP (Median) [0/1/2/3/4]	mean ± sd	1.7 ± 1.1	1.4 ±1.3	1.7 ±1.3	2 ± 0.87	1.2 ±1.1	1.1 ± 1.2	2.2 ±1	2.1 ±1.1	<0.01
Ulnar SNAP (Median) [0/1/2/3/4]	mean ± sd	3.4 ±0.99	3.3 ±0.5	3.8 ±0.67	3.6 ±0.94	3.3 ±0.85	3.9 ± 0.45	3.5 ±0.55	2.9 ±1.1	<0.01
10m Walk Test [sec]	mean ± sd	7.8 ± 1.7	8 ±2.8	13 ± 6.4	16 ± 18	6.7 ±1.4	8.7 ±2.4	6.1 ±1.7	9.5 ±3.4	<0.01
9 Peg Hole Test (dominant hand) [sec]	mean ± sd	27 ±11	27 ±7.1	25 ±8.8	33 ±23	22 ± 4.4	32 ± 14	24 ±8.9	29 ± 14	<0.01
Fist Grip (dominant hand) [kg]	mean ± sd	128 ±80	197 ±121	105 ± 41	147 ± 79	169 ± 77	104 ±107	196 ± 77	139 ±75	<0.01
Three Point Grip (dominant hand) [kg]	mean ± sd	54 ±24	59 ±31	50 ±27	48 ±28	52 ±24	29 ±16	55 ±25	55 ±26	0.68
Foot Dorsal (dominant foot) [kg]	mean ± sd	51 ±35	105 ± 78	50 ±29	51 ±38	201 ±75	30 ±17	36 ±28	33 ±21	<0.01
Foot Plantar (dominant foot) [kg]	mean ± sd	75 ±46	244 ±118	75 ±37	113 ±50	263 ± 80	56 ±29	45 ±25	71 ±46	<0.01

Table 2: Regression coefficients for clinical parameters modeling biomarker expression.

Using multiple linear modeling, the effect of each of the CMT scores as well as each primary and secondary parameter (y axis) on the expression of each of the potental biomarkers (x axis) was assessed while controlling for effects from the co-variables center, plate, age, gender, and BMI. The table shows the regression coefficients from the individual models. The background color graphically represents the regression coefficients. The modeling was done on the Ct scale that is negative logarithmic to the gene expression, thus: blue shades: positive effect on gene expression, red shade negative effect on gene expression. Significant (on a non-adjusted 5% level) values are printed in black.

	SPRR1A	PPARG	NRG1-I	MUCL1	GSTT2	GSTT1	GSTA4	GRIA1	FN3KRP	FN1	ENPP1	CTSA	CRISP3	CDA	BGN	ANPEP
CMTNS	0,05	0,04	0,03	0,07	0,03	0,05	0,03	0,04	0,04	0,03	0,04	0,03	0,07	0,04	0,04	0,03
CMTNS_full	0,03	0,03	0,01	0,03	0	0,04	0,02	0,01	0,01	0,01	0,02	0,01	0,04	0,02	0,02	0,01
CMTNS_sig	0,04	0,04	0,02	0,05	0,02	0,08	0,03	0,03	0,03	0,02	0,03	0,03	0,05	0,03	0,03	0,02
CMTNS_mod	0,06	0,05	0,02	0,04	0,02	0,1	0,03	0,04	0,04	0,03	0,03	0,04	0,05	0,04	0,04	0,03
Sensory Symptoms	0,01	0,09	0,17	0,13	−0,01	0,3	0,04	0,08	0,01	0,11	0,11	0,04	0,11	0,04	0,08	0,07
Motor Symptoms Legs	0,22	0,24	0,12	0,16	0,13	0,52	0,13	0,22	0,19	0,08	0,12	0,16	0,17	0,24	0,16	0,1
Motor Symptoms Arms	0,4	0,13	0,02	0,11	0,07	0,3	0,1	0,07	0,14	0,01	0,16	0,05	0,24	0,08	0,13	0,05
Pin Sensibility	0,11	−0,02	0,08	0,06	−0,15	0,25	0,03	−0,01	−0,02	0,08	0,05	−0,11	0,04	−0,02	0,08	−0,05
Vibration	0,07	0,23	0,14	0,34	0,24	0,2	0,09	0,26	0,23	0,1	0,31	0,23	0,36	0,26	0,13	0,15
Strength of Legs	0,31	0,3	0,13	0,14	0,22	0,25	0,15	0,28	0,23	0,16	0,23	0,22	0,31	0,3	0,26	0,21
Strength of Arms	0,27	0,12	0,1	0,14	0,07	0,42	0,2	0,04	0,04	0,09	0,1	0	0,34	0,08	0,13	0,01
Ulnar CMAP (Median)	0,07	0,1	−0,03	0,12	0,13	−0,04	0,11	0,08	0,12	0,08	−0,04	0,07	0,08	0,05	0,02	0,05
Ulnar SNAP (Median)	0,01	−0,03	0,14	0,64	0,08	0,13	0,06	−0,07	0,02	0,11	0,02	−0,01	0,39	0,03	0,14	0,07
10m Walk Test	0,15	0,09	−0,04	−0,06	−0,05	0,2	0,03	0,08	0,1	0,01	0,07	0,06	0,06	0,07	0,03	0,03
9 Peg Hole Test (dominant hand)	0,12	0,04	0,08	0,21	0,03	0,32	0,07	0,01	0,05	0,11	0,05	0,06	0,16	0,05	0,13	0,08
Fist Grip (dominant hand)	0,15	0,12	0	0,15	−0,01	0,07	0,06	0,07	−0,01	0,1	0,14	0,05	0,21	0,07	0,12	0,07
Three Point Grip (dominant hand)	0,02	0,06	−0,06	0,15	−0,05	0,08	0,05	0,04	0	0,08	0,07	−0,02	0,14	0	0,03	0,03
Foot Dorsal (dominant foot)	0,12	0,18	0,04	0,27	0,11	−0,05	0,09	0,16	0,12	0,14	0,15	0,12	0,33	0,15	0,17	0,17
Foot Plantar (dominant foot)	0,19	−0,04	0,01	−0,11	−0,1	0,09	0,1	−0,11	−0,01	−0,02	−0,2	−0,09	0,12	−0,07	−0,05	−0,08

Distinct cutaneous gene expression profiles reflect disease severity in CMT1A

After clinical examination, the patients were asked to undergo a skin biopsy for biomarker analysis (Fig. 1A). From the collected skin biopsies the RNA was purified and qPCR performed for 16 transcripts derived from the previously identified disease severity biomarkers in CMT rats (16). The selected mRNA biomarkers were readily detectable and display a highly variable expression when compared to housekeeping genes (Fig. 1B). Of note, the expression values among potential biomarkers demonstrate a positive correlation with each other throughout the entire dataset (Fig. 1C). In detail, eight out of 16 tested biomarkers (CDA, CTSA, GRIA1, ENPP1, ANPEP, FN3KRP, GSTT2 and PPARG) show a similar expression pattern among patients with correlation coefficients >0.7 (Fig. 1A-B). Next, we asked whether we can identify the components in the expression of these eight biomarkers that provide information with respect to CMT1A disease severity. A supervised principal component analysis (PCA) indeed revealed a principal component 1 (PC1) that can explain 95% of the variance observed in the expression of the biomarker cluster (Fig.2A). Importantly, mapping the transcriptional profiles of the available samples to the first two principal components (PC1 and PC2) allows for the separation of the patient cohort according to the CMT status (Fig. 2B). In detail, PC1 not only separates patients with CMT1A from healthy controls, but also differentiates for disease severity within the patient group (mildly affected: CMTNS<10; moderately affected: CMTNS 10–20; severely affected: CMTNS>20) (Fig.2B). Of note, severely affected patients in PC1 cluster in close proximity to healthy controls, whereas mildly affected patients display a more distinct expression profile (Fig. 2B). A support vector machine (SVM) classifier, that was only trained on gene expression profiles of the identified eight biomarkers, separates mildly from severely affected patients (based on the CMTNS) with a sensitivity of 90% and a specificity of 76% (Fig. 2C). To test to which extent the expression of single genes correlates with disease severity, we next analyzed their relation to the CMTNS and individual primary and secondary outcome measures (Tab. 2). As mentioned above, to avoid disruptive influence of center effects on biomarker correlation, effects were controlled for the co-variables center, age, gender and BMI. Linear models were fit to the expression data to quantify the influence of each (sub)-score on the expression of each biomarker. The resulting regression coefficients (i.e. the factor by which the biomarker expression is changing when the assessed clinical parameter is changing for one unit) show a number of potential biomarkers significantly related to variants of CMTNS and one or more of the sub-scores (Tab. 2). Importantly, some primary clinical sub-scores, e.g. ‘Strength of Legs’ and ‘Vibration’, show numerous significant correlations when compared to the number of significant hits in the cumulative CMTNS variants (Tab. 2).

Figure 2: Cutaneous biomarker expression separates CMT1A disease severity and controls.

A Principal Component Analysis (PCA) was performed on the PCR data of the biomarker cluster identified in Fig.1B-C. A scree plot shows the percentage of the explained variance for each of the principal components (PCs). PC1 captures most of the information and explains 95% of the observed variance while PC2 explains already only 1% of the observed variance.

B Mapping of the samples on the first two principal components from A with PC1 on the x axis and PC2 on the y axis. All samples are color coded by their CMT disease severity status (red: healthy, green: mildly affected (CMTNS <= 10), blue: moderately affected (10 > CMTNS < 20), purple: severely affected (CMTNS >= 20). The ellipses are 70% probability ellipses assuming two dimensional normally distributed data drawn separately for each of the 4 groups.

C A support vector machine (SVM) classifier separating mild and severe CMT cases based solely on gene expression profiles of the eight biomarkers (Fig. 1B-C) was trained. Shown are the results from a 10 times repeated 10-fold cross validation. The presented ROC curve plots sensitivity (y axis) vs. specificity (x-axis). A perfect classifier would reach the top left corner (100% sensitivity and 100% specificity). The closest point (youden index) of the trained SVM reaches 90% sensitivity and 76.1% specificity.

The mRNA expression of single genes is changing with disease progression

To assess whether the previously identified biomarkers also harbor information about disease progression we analyzed mRNA expression in 45 sets of paired skin biopsies that were sampled from the very same patients over a 2–3 years’ timeframe (16 from Germany, 14 from Italy and 15 from the USA, Fig. 1A). In total, gene expression values were available for two time points for 45 patients, clinical data were available for 46 patients, with 30 CMT1A patients (mildly affected: n=3, moderately affected: n=23, severely affected: n=4) having both, clinical and gene expression values. We could confirm the low mean change in CMTNS of about ~0.7 points per year in our cohort of 46 patients with CMT1A (Fig. 3A-B), a value that matches previously published data (36). Mixed model linear regression on repeated measures analysis show a highly significant effect of ‘time’ (i.e. disease progression) on the CMTNS, controlling for gender, age and BMI, pointing to clinical detectability of disease progression in a 2–3 year’s timeframe (Fig. 3C). In 45 skin biopsies from CMT1A patients, we found six genes with significantly increased expression over time: GSTT2, CTSA, PPARG, CDA, ENPP1, NRG1-I (Fig. 4A). The extent of increase in gene expression per year ranges from 2^0.28=1.2 fold for NRG1-I to 2^0.66=1.6 fold for CTSA (Fig. 4B). A predictive random forest model was trained on these six genes. The 30 CMT1A patients from whom both, clinical and biomarker data were available, were classified into progressive (CMTNS increased, n=19) and non-progressive (CMTNS not increased, n=11), and the classifier had a specificity of 100% and a sensitivity of 63.2% in the detection of disease progression (Fig. 4C). Importantly, five out of the six disease progression biomarkers (GSTT2, CTSA, PPARG, CDA, ENPP1) are also part of the disease severity biomarker cluster of the 266 CMT1A patients cohort (Fig. 1B-C), further supporting their eligibility as biomarkers in CMT1A.

Figure 3: Disease progression is clinically detectable over a two to three year’s time period.

A Progression of CMT measured by the scores CMTNSv1, CMTNS_full, CMTNS_mod, and CMTNS_signif with patients (colored by contributing center) on the y-axis and the score difference on the x axis where positive difference correspond to higher scores at the second examination.

B The change of CMTNS is shown over time. The time between measurements ranges from 2 to 3 years (with one exception of only 1 year time difference). Each line represents one patient.

C Regression coefficients with 95% confidence intervals from the fit of a linear mixed effect model for CMTNS with fixed effects BMI, gender, age, and time and a random effect accounting for the repeated measures in the patients. The factor ‘time’ has a significant influence on the CMTNS with an estimated increase in CMTNS of 0.75 per year.

Figure 4: Cutaneous expression of selected biomarkers changes with disease progression.

A Mixed effects repeated measures regression models were fit per gene to the expression data. The effect of time was assessed controlling for BMI, gender, age, and center effects. Shown are the p values for the time effect (orange: adjusted p value significant) for all genes. B Shown is the progression data underlying the regression models from D displayed as Ct value (gene expression) as a function of time. Each colored line represents one patient and the black lines show the average patient at time 0 progressing by the modeled time effect.

C A predictive random forest model that classifies patients into progressive patients (CMTNS increase) and non-progressive patients (CMTNS does not increase) was trained on the six genes with significant change in expression over time. Data was available for 19 progressive and 11 non-progressive patients. The consensus ROC curve from a 10 times repeated 10-fold cross validation reveals an AUC of 0.74 and the sensitivity/specificity at the youden index (marked with a dot) are estimated to be 63.2% / 100%.

Thus, skin biopsy is a valuable source for the identification of biomarkers in CMT1A (Fig. 2, Tab. 2). Moreover, we identified six genes with changing expression over time that may serve as disease progression biomarkers and could directly be implemented into ongoing clinical trials (Fig. 3).

Discussion

Despite promising results from experimental therapies in animal models, CMT1A remains a disease without known cure. Clinical trials are hampered by the slow CMT1A disease progression and the availability of outcome measures with limited sensitivity to change (30,37,38). Next to improvement of clinical outcome measures, molecular biomarkers, when available, would add powerful tools to assess and track disease progression in clinical trials. Molecular biomarkers may ideally derive from body fluids or minimally invasively sampled tissue. Although it is not known so far whether CMT1A is reflected by alteration of body fluid characteristics which then could be harnessed for biomarker development, skin biopsies have proven powerful tools to study CMT1A (39–41). In a preliminary study we could identify CMT rat derived cutaneous disease severity biomarkers in skin biopsies from a small cohort of 46 patients with CMT1A. By applying a post hoc multiple linear regression model, in this study we were able to show that 47% of the observed variance in disease severity (as measured by the CMTNS) can be predicted by gene expression and age (16). The development of applicable biomarkers of disease severity in CMT1A, however, requires a four step fit-for-purpose approach (see introduction) (32,33). To validate the basic assay performance (fit-for-purpose step II), we initiated a worldwide prospective study and 9 centers in Europe and the US contributing clinical data and skin biopsies from 266 patients with CMT1A. The expression of 16 previously identified potential cutaneous disease severity biomarkers were tested in skin biopsies from the new patient cohort and we confirmed that disease severity is indeed reflected by cutaneous gene expression (Fig. 2). The number of individual genes, however, that significantly correlate with the cumulative CMTNS variants (maximum 7/16 genes) is relatively low compared to correlations of biomarker expression with selected single parameters (e.g. with ‘Strength of legs reached 13/16 hits, Tab. 2). The notion that some items of the outcome measures do in fact reflect the actual disease severity more sensitively than others has been reported previously (24) and could explain that biomarker correlations vary between CMTNS variants and sub-scores. However, cutaneous expression of a cluster of eight genes (CDA, CTSA, GRIA1, ENPP1, ANPEP, FN3KRP, GSTT2 and PPARG, correlates with disease severity (as assessed by the CMTNS) with very high sensitivity (90%, Fig. 2C) confirming that expression analyses is suitable to develop biomarkers (fit-for-purpose step II). A major caveat, however, is the difficulty to improve outcome sensitivity solely on the basis of a rather insensitive anchor as the CMTNS. To overcome this problem, we performed an analysis of change in gene expression over time (fit-for-purpose step III). From 16 tested genes, we found PPARG, GSTT2, CTSA, CDA, ENPP1 and NRG1-I to show a statistically significant increase in expression over time, all of which (except NRG1-I) were identified as a component of a biomarker cluster within the 266 CMT1A patients cohort (Fig. 1B-C, Fig. 4). Independent from the cluster analysis (Fig. 1), the expression of individual genes showed significant correlations with various CMTNS related items (Tab. 2). For example, the progression biomarker NRG1-I (Fig. 4A-B) correlates with the CMTNS was the only gene found to correlate with sensory symptoms (Tab. 2), although this gene was not part of the identified cluster. However, on the basis of the insensitive CMTNS, the potential disease progression biomarkers display a sensitivity of 63.2% (Fig. 4C). Here, the only moderate sensitivity of progression biomarkers towards clinical impairment may directly result from the insensitive CMTNS anchor. The maximum specificity of 100%, in turn, demonstrates that non-progressive patients were easily detectable by biomarker expression in the examined patient cohort. As a consequence, patients which respond to a therapy (and thereby display a halt in disease progression) could be identified more easily with the help of the here reported disease progression biomarkers. The validation of these biomarkers in the disease progression of further CMT1A patients in natural history studies (also by next generation sequencing approaches) may show whether a higher sensitivity can be achieved and may reveal further evidence for the potential practical use of these biomarkers.

Enrichment analyses of the eight clustered genes (Tab. 3) revealed an involvement in metabolism (Reactome pathway), negative regulation of cell growth (Gene Ontology, biological process), vacuolar/lysosomal membrane (Gene Ontology, cellular component) and exopeptidase activity (Gene Ontology, molecular function). This functional annotation may fit to the reported impact of the duplicated disease gene PMP22 on lysosomal alteration upon overexpression in CMT1A (42).

Table 3: Gene enrichment analysis and functional annotation of CDA, CTSA, GRIA1, ENPP1, ANPEP, FN3KRP, GSTT2 and PPARG.

Enriched terms are shown for the databases a) Reactome Pathway, b) Gene Ontology (GO), biological process, c) GO, cellular component and d) GO, molecular function. The maximum 20 significant hits are displayed with term, database identifier, gene set overlap, adjusted p-value, Z-score, combined score and genes involved (only hits with more than one gene).

Reactome pathway
Term	Identifier	Overlap	Adjusted P-value	Z-score	Combined Score	Genes
Metabolism	R-HSA-1430728	5/1908	0,01167	−2,25	9,99	CTSA;CDA;ENPP1;GSTT2;PPARG
Metabolism of proteins	R-HSA-392499	4/1074	0,01167	−2,14	9,54	CTSA;GRIA1;ANPEP;FN3KRP
Post-translational protein modification	R-HSA-597592	3/521	0,01428	−2,18	9,25	CTSA;GRIA1;FN3KRP
Asparagine N-linked glycosylation	R-HSA-446203	2/259	0,03554	−2,08	6,95	CTSA;GRIA1
GO biological process
Term	Identifier	Overlap	Adjusted P-value	Z-score	Combined Score	Genes
negative regulation of cell growth	GO:0030308	3/150	0,00619	−2,25	11,44	CDA;ENPP1;PPARG
negative regulation of growth	GO:0045926	3/221	0,00984	−2,31	10,68	CDA;ENPP1;PPARG
regulation of cell growth	GO:0001558	3/322	0,01999	−2,34	9,17	CDA;ENPP1;PPARG
receptor metabolic process	GO:0043112	2/71	0,02359	−2,12	7,94	GRIA1;PPARG
regulation of fat cell differentiation	GO:0045598	2/87	0,02832	−2,32	8,26	ENPP1;PPARG
cellular response to peptide hormone stimulus	GO:0071375	2/261	0,03832	−2,39	7,80	ENPP1;PPARG
cellular response to peptide	GO:1901653	2/273	0,03832	−2,38	7,76	ENPP1;PPARG
cellular response to insulin stimulus	GO:0032869	2/195	0,03832	−2,36	7,68	ENPP1;PPARG
response to insulin	GO:0032868	2/246	0,03832	−2,33	7,60	ENPP1;PPARG
protein maturation	GO:0051604	2/222	0,03832	−2,29	7,46	ANPEP;PPARG
receptor-mediated endocytosis	GO:0006898	2/152	0,03832	−2,27	7,41	GRIA1;ENPP1
protein processing	GO:0016485	2/201	0,03832	−2,24	7,32	ANPEP;PPARG
endocytosis	GO:0006897	2/323	0,03832	−2,24	7,30	GRIA1;ENPP1
cellular response to organonitrogen compound	GO:0071417	2/411	0,03955	−2,40	7,75	ENPP1;PPARG
cellular response to nitrogen compound	GO:1901699	2/438	0,03955	−2,37	7,64	ENPP1;PPARG
response to peptide	GO:1901652	2/384	0,03955	−2,31	7,47	ENPP1;PPARG
response to peptide hormone	GO:0043434	2/364	0,03955	−2,31	7,47	ENPP1;PPARG
cellular response to hormone stimulus	GO:0032870	2/462	0,03955	−2,29	7,39	ENPP1;PPARG
negative regulation of phosphate metabolic process	GO:0045936	2/458	0,03955	−2,26	7,30	CDA;ENPP1
anion transport	GO:0006820	2/443	0,03955	−2,14	6,91	ENPP1;PPARG
GO cellular component
Term	Identifier	Overlap	Adjusted P-value	Z-score	Combined Score	Genes
vacuolar part	GO:0044437	3/285	0,00730	−2,32	11,42	CTSA;ANPEP;ENPP1
lysosomal membrane	GO:0005765	2/196	0,04628	−2,26	6,93	ANPEP;ENPP1
vacuolar membrane	GO:0005774	2/208	0,04628	−2,25	6,93	ANPEP;ENPP1
GO molecular function
Term	Identifier	Overlap	Adjusted P-value	Z-score	Combined Score	Genes
exopeptidase activity	GO:0008238	2/111	0,03047	−2,49	8,69	CTSA;ANPEP
zinc ion binding	GO:0008270	4/1256	0,03047	−2,31	8,05	CDA;ANPEP;ENPP1;PPARG

Taken together, the validity of the identified biomarkers was tested twice, first as disease severity biomarkers in a cohort including 266 CMT1A patients and 37 healthy controls and, second, as progression markers in an additional longitudinal cohort including 45 patient with CMT1A. We confirmed that gene expression analyses of the skin is suitable for the assessment of the disease severity, and that the cutaneous expression of individual genes is changing over time in patients with CMT1A. We suggest, that the expression of five genes that were part of an identified biomarker cluster and correlated with disease progression over time (CDA, CTSA, ENPP1, GSTT2, PPARG) is a valid set of biomarkers in CMT1A. When implemented and successfully validated in ongoing clinical trials, the here reported biomarkers could markedly accelerate the development of a therapy for CMT1A.

APPENDIX

The CMT-TRIAAL Group (all in Italy)

IRCCS Foundation, C. Besta Neurological Institute, Milan: D.Pareyson, D. Calabrese, C. Marchesi, E. Salsano, L. Nanetti, C. Marelli, V. Scaioli, C. Ciano, M. Rimoldi, G. Lauria, G. Ferrari, E. Rizzetto, F. Camozzi;

Department of Neurology, Ophthalmology and Genetics, University of Genoa, Genoa: A. Schenone, E. Narciso, M. Grandis, M. Monti-Bragadin, L. Nobbio;

Department of Neurological, Neuropsychological, Morphological and Motor Sciences, University of Verona, Verona: G.M. Fabrizi, T. Cavallaro, A. Casano, L. Bertolasi, I. Cabrini, K. Corrà, N. Rizzuto;

Department of Neurological Sciences, Federico II University of Naples, Naples: L. Santoro, F. Manganelli, C. Pisciotta; Department of Neurology, “Salvatore Maugeri” Foundation, IRCCS, Telese Terme: M. Nolano;

Department of Neurosciences, University of Messina, Messina: G. Vita, A. Mazzeo, R. Di Leo, G. Majorana, M. Russo;

Magna Graecia University, Neurology Clinic, and Neuroimaging Research Unit, National Research Council, Catanzaro: A. Quatrone, P. Valentino, R. Nisticò, D. Pirritano, A. Lucisano, M. Canino;

Institute of Neurology, Department of Neurosciences, Sacro Cuore Catholic University, and Don Gnocchi Foundation, Rome: L. Padua, C. Pazzaglia, G. Granata, M. Foschini;

Department of Neurosciences, University of Parma, Parma: F. Gemignani, F. Brindani, F. Vitetta; I. Allegri;

Department of Pharmacological Sciences, School of Pharmacy, University of Milan, Milan: F. Visioli, P. Bogani.