A Semi-Automated Software for Myelin G-Ratio Quantification

Research Article: Open Source Tools and Methods | Novel Tools and Methods MyelTracer: A semi-automated software for myelin g-ratio quantification https://doi.org/10.1523/ENEURO.0558-20.2021

Cite as: eNeuro 2021; 10.1523/ENEURO.0558-20.2021 Received: 23 December 2020 Revised: 14 June 2021 Accepted: 17 June 2021

This Early Release article has been peer-reviewed and accepted, but has not been through the composition and copyediting processes. The final version may differ slightly in style or formatting and will contain links to any extended data.

Alerts: Sign up at www.eneuro.org/alerts to receive customized email alerts when the fully formatted version of this article is published.

Copyright © 2021 Kaiser et al. This is an open-access article distributed under the terms of the Creative Commons Attribution 4.0 International license, which permits unrestricted use, distribution and reproduction in any medium provided that the original work is properly attributed.

1 Manuscript Title Page Instructions

2 Please list the following information on your separate title page in Word .DOC format in the order 3 listed below and upload as a Title Page file at submission

4 1. Manuscript Title (50 word maximum)

5 MyelTracer: A semi-automated software for myelin g-ratio quantification

6 2. Abbreviated Title (50 character maximum)

7 Myelin g-ratio software

8 3. List all Author Names and Affiliations in order as they would appear in the published article

9 Tobias Kaiser1,2,*, Harrison Allen1,3,*, Ohyoon Kwon1,2,*, Boaz Barak4, Jing Wang5, Zhigang He5, Minqing 10 Jiang1,2,#, Guoping Feng1,2,6,#

11 1McGovern Institute for Brain Research 12 2Department of Brain and Cognitive Sciences, Massachusetts Institute of Technology, Cambridge, MA 13 02139, USA 14 3Department of Electrical Engineering and Computer Science, Massachusetts Institute of Technology, 15 Cambridge, MA 02139, USA 16 Massachusetts Institute of Technology, Cambridge, MA 02139, USA 17 4The Sagol School of Neuroscience and The School of Psychological Sciences, Tel Aviv University, 18 Israel 19 5F.M. Kirby Neurobiology Center, Boston Children’s Hospital, and Department of Neurology and 20 Ophthalmology, Harvard Medical School, Boston, MA, USA 21 6Stanley Center for Psychiatric Research, Broad Institute of MIT and Harvard, Cambridge, MA 02142, 22 USA 23 24 *these authors contributed equally 25 #corresponding author 26

27 4. Author Contributions:

28 TK, HA, MJ, and GF designed the research. HA wrote the software. TK and MJ generated the electron 29 microscopy data. TK, MJ, and OK analyzed the data and benchmarked the software. BB, JW, ZH 30 contributed electron micrographs for additional analysis. TK, MJ, and GF wrote the paper with inputs 31 from all authors.

32 5. Correspondence should be addressed to (include email address)

33 Minqing Jiang; [email protected] 34 Guoping Feng; [email protected] 35

36 6. Number of Figures: 3 37 7. Number of Tables: 1 38 8. Number of Multimedia: 0 39 9. Number of words for Abstract: 159 40 10. Number of words for Significance Statement: 46 41 11. Number of words for Introduction: 562 42 12. Number of words for Discussion: 817

43 13. Acknowledgements

44 We thank all members of the Feng lab for helpful discussions. We would also like to acknowledge 45 Nicholas Sanders and Dongqing Wang for excellent technical support and Boaz Barak for sharing data 46 from Williams Syndrome mice. We further thank Maria Ericsson and Louise Trakimas at the Harvard 47 Electron Microscopy Core facility. TK is supported by the Tan-Yang center for Autism Research at MIT. 48 GF is supported by the National Institute of Mental Health (5R01MH097104), the Poitras Center for 49 Affective Disorders Research at MIT, Tan-Yang center for Autism Research at MIT, Stanley Center for 50 Psychiatric Research at Broad Institute of MIT and Harvard, Nancy Lurie Marks Family Foundation, 51 Simons Foundation Autism Research Initiative (SFARI) and Simons Center for the Social Brain at MIT.

52 14. Conflict of Interest

53 A. No (State ‘Authors report no conflict of interest’)

54 The authors report no conflict of interest.

55 B. Yes (Please explain)

56 15. Funding sources

57 Tan-Yang Center for Autism Research at MIT 58 Simons Foundation Autism Research Initiative 59 Simons Center for the Social Brain at MIT 60 NIMH (R01MH097104) 61 Poitras Center for Affective Disorders Research at MIT 62 Stanley Center for Psychiatric Research at Broad Institute of MIT and Harvard 63 Nancy Lurie Marks Family Foundation 64

65 MyelTracer: A semi-automated software for myelin g-ratio quantification

67 Abstract

69 In the central and peripheral nervous systems, the myelin sheath promotes neuronal

70 signal transduction. The thickness of the myelin sheath changes during development and in

71 disease conditions like multiple sclerosis. Such changes are routinely detected using electron

72 microscopy through g-ratio quantification. While g-ratio is one of the most critical measurements

73 in myelin studies, a major drawback is that g-ratio quantification is extremely laborious and

74 time-consuming. Here, we report the development and validation of MyelTracer, an installable,

75 stand-alone software for semi-automated g-ratio quantification based on the Open Computer

76 Vision Library (OpenCV). Compared to manual g-ratio quantification, using MyelTracer

77 produces consistent results across multiple tissues and animal ages, as well as in remyelination

78 after optic nerve crush, and reduces total quantification time by 40-60%. With g-ratio

79 measurements via MyelTracer, a known hypomyelination phenotype can be detected in a

80 Williams Syndrome mouse model. MyelTracer is easy to use and freely available for Windows

81 and Mac OS X [GitHub link, blinded for review].

85 Significance statement

86 An easy-to-use software suite for g-ratio quantification in myelin ultrastructure

87 micrographs is currently unavailable, but it is much needed to streamline the study of myelin in

88 homeostasis and disease-related conditions. We used computer vision libraries to develop a

89 freely available software to facilitate studies of myelination.

91 Introduction

93 The myelin sheath is essential for proper neuronal functions in central and peripheral

94 nervous systems. Oligodendrocytes and Schwann cells produce the myelin sheath through

95 lamellar enwrapping of axons (Simons and Trotter, 2007; Salzer, 2015; Hughes and Appel, 2016;

96 Stadelmann et al., 2019). The myelin sheath insulates axons, thereby preserving axonal integrity

97 and promoting neuronal signal transduction (Sherman and Brophy, 2005; Nave, 2010). Under

98 physiological conditions, changes in myelin contribute to behaviorally relevant neural plasticity

99 mechanisms and experience-dependent sensory adaptations (Gibson et al., 2014; McKenzie et

100 al., 2014; Hughes et al., 2018).

101 Abnormal changes in myelin are prominent features of many clinical pathologies.

102 Ultrastructural myelin abnormalities, such as hypomyelination, myelin degeneration, and

103 tomaculi, are cardinal features of prototypic white-matter diseases like multiple sclerosis,

104 Pelizaeus Merzbacher disease, and Charcot-Marie-Tooth disease (Krajewski et al., 2000; Sander

105 et al., 2000; Franklin and ffrench-Constant, 2008; Lin and Popko, 2009; Duncan and Radcliff,

106 2016). In addition, myelin abnormalities have recently been discovered as common features of

107 neurodevelopmental disorders, including Pitt Hopkins Syndrome, Rett Syndrome, autism

108 spectrum disorders (ASDs), and William’s Syndrome (Zhao et al., 2018; Barak et al., 2019; Phan

109 et al., 2020).

110 In both preclinical and clinical settings, pathological myelin abnormalities are often

111 identified by studying myelin ultrastructure. Specifically, researchers use g-ratio as a metric for

112 the relative thickness of the myelin sheath in cross-sectional micrographs (Rushton, 1951). Since

113 manual myelin tracing is a major bottleneck in g-ratio quantification, tools that streamline this

114 process will tremendously facilitate future studies investigating activity-dependent homeostatic

115 and pathological changes to the myelin sheath.

116 Several non-automated, semi-automated, and fully automated toolboxes are currently

117 available. They include G-ratio for ImageJ, AxonSeg, and AxonDeepSeg (Goebbels et al., 2010;

118 More et al., 2011; Bégin et al., 2014; Zaimi et al., 2016, 2018; Janjic et al., 2019). Collectively,

119 these tools have facilitated different quantitative measurements of myelin ultrastructure, such as

120 the number of axons and g-ratio. However, the current tools with more automation impose

121 difficult requirements on their users, such as pre-processing data or modifying parameters by

122 writing code. These drawbacks have impeded the tools’ widespread adoption by the scientific

123 community. We identified four major requirements of an open access toolbox that would

124 significantly facilitate myelin analyses: (1) an intuitive graphical user interface that can be

125 installed and used without running code, (2) consistent quantification results compared to manual

126 analysis, (3) significantly less time-consuming, and (4) well-organized output data files and

127 overlay images for publication and post-hoc quality control.

128 Here we report the development and validation of MyelTracer, an easy-to-use software

129 that fulfills these critical requirements. Built using OpenCV and PyQt5's GUI toolkit,

130 MyelTracer replaces the manual tracing of axons and myelin with semi-automated, threshold-

131 dependent outlining. Compared to manual quantification, semi-automated quantification of axon

132 diameters and g-ratio via MyelTracer produces consistent results across several tissues and

133 developmental time points while reducing total quantification time by 40-60% depending on the

134 region. Furthermore, using g-ratio measured with MyelTracer, we can detect a known

135 hypomyelination phenotype in a Williams Syndrome mouse model, as well as in remyelination

136 after optic nerve crush, indicating the applicability of MyelTracer for studies in disease contexts.

137 To ensure MyelTracer’s all-in-one output functionality, we included a manual tracing function

138 and counters for quantification of abnormally shaped sheaths and fraction of myelinated axons,

139 respectively. MyelTracer is a valuable tool for the study of myelin ultrastructure and is freely

140 available to the public as an installable software for Mac and PC [GitHub link, blinded for

141 review].

142 Materials and methods

143 Animal work

144 All animal procedures were performed in accordance with the [Author University] animal

145 care committee's regulations.

146

147 Code Accessibility

148 The code/software described in the paper is freely available online at [URL redacted for

149 double-blind review]. The code is available as Extended Data.

150

151 Software development

152 Outline extraction of axons is performed using an implementation of OpenCV in Python.

153 The processing is done as follows: First, the color space of the image is converted from RGB to

154 grayscale, so that all pixels have a scalar magnitude between 0 and 255.

155 cvtColor(src, COLOR_BGR2GRAY)

156 Then, a bilateral filter is applied to the source image to reduce noise and smooth edges.

157 bilateralFilter(src, d=9, sigmaColor=75, sigmaSpace=75)

158 A binary thresholding filter converts pixel values below the user defined threshold, t, to black,

159 and those above t to white.

160 threshold(src, thresh=t, maxval=255, type=THRESH_BINARY)

161 Lines drawn by the user are then applied to the thresholded image using the polylines function.

162 Contours are extracted as the boundaries between the black and white pixels of the thresholded

163 image.

164 findContours(src, mode=RETR_TREE, method=CHAIN_APPROX_SIMPLE)

165 Contours with fewer than five vertices are filtered out using the len function as well as contours

166 that don’t fall within the user-defined area constraints, Min Size and Max Size, calculated using

167 the contourArea function.

168 Once selected, contours for each feature—axon, inner myelin sheath, and outer myelin sheath—

169 are grouped together by checking if the features nest within one another using pointPolygonTest.

170 Exported data are written in a plain-text csv document for compatibility with most spreadsheet

171 softwares. The exported features are organized by the order in which the user selected the axon

172 feature. Area is calculated using contourArea, perimeter is calculated using arcLength, diameter

173 is calculated using 2 ⋅ /, and g-ratio is calculated using

174 /.

175 The extraction software is wrapped in a custom-designed PyQt5 GUI, and both Windows and

176 MacOS installers are generated using the fman Build System (fbs). MyelTracer has been tested

177 and proven to work on Windows 10 and MacOS Catalina 10.15.

178 Electron microscopy and g-ratio tracing

179 Sample preparation and imaging

180 One-month-old and two-month-old male and female C57BL/6J mice were deeply

181 anesthetized with isoflurane and transcardially perfused with 30 ml ice-cold saline solution

182 followed by 30 ml fresh ice-cold 2.5% glutaraldehyde + 2% PFA in 0.1 M sodium cacodylate

183 buffer (pH 7.4). Brains, optic nerves and sciatic nerves were dissected and kept in the fixation

184 solution overnight at 4 °C. Small pieces (1–2 mm cubes) of tissue samples from the dorsal corpus

185 callosum at the level of the fornix (Bregma -0.94 mm), optic nerve, and sciatic nerve were

186 postfixed for at least two hours at room temperature in the fixation solution. They were then

187 washed in 0.1 M sodium cacodylate buffer and postfixed with 1% osmium tetroxide/1.5%

188 potassium ferrocyanide for one hour before being washed in water three times and incubated in

189 1% aqueous uranyl acetate for one hour. This was followed by two washes in water and

190 subsequent dehydration in grades of alcohol (10 min each; 50, 70, 90 and 2 × 10 min 100%). The

191 samples were then put in propylene oxide for one hour and infiltrated overnight in a 1:1 mixture

192 of propylene oxide and epoxy resin mixture (TAAB Epon; Marivac Canada). The following day,

193 the samples were embedded in TAAB Epon and polymerized at 60 °C for 48 h. Ultra-thin

194 sections (approximately 60 nm) were cut on a Reichert Ultracut S microtome (Leica), picked up

195 on copper grids stained with lead citrate and examined in a JEOL 1200EX transmission electron

196 microscope (JEOL USA, Inc., USA). The images were recorded with a 2k CCD camera

197 (Advanced Microscopy Techniques).

198

199 Quantification of g-ratio

200 To compare manually quantified g-ratio and those quantified with MyelTracer, three

201 images for each region of interest were analyzed manually and using MyelTracer. Each image

202 was analyzed first with MyelTracer, which automatically numbered every quantified axon. For

203 manual g-ratio quantification, the same axons were manually numbered and then traced using

204 ImageJ. To compare the control and Gtf2i-knockout group, raw images were obtained from

205 Barak and coworkers (Barak et al., 2019), blinded of their genotypes and quantified using

206 MyelTracer. To test MyelTracer’s ability to quantify g-ratios in remyelinating tissues, raw

207 images of the optic nerve from control animals and animals 28 days post injury with Montelukast

208 and Pexidartinib treatment were obtained from Wang and coworkers (Wang et al., 2020). For

209 each experiment comparing manual tracing with MyelTracer, several micrographs were obtained

210 from preparations of two to three mice per anatomic region. Within these micrographs, at least

211 100-300 axons were analyzed depending on the anatomical region. For testing MyelTracer’s

212 capability to detect a hypomyelination phenotype in a Williams Syndrome model, several

213 micrographs from three mice per genotype with a total of 332 and 455 axons were analyzed for

214 control and Gtf2i fl/fl; Nex-Cre mice, respectively. For optic nerve, sciatic nerve, and corpus

215 callosum, three different researchers recorded image quantification times. Numbers denoting

216 measured axons in representative images were enlarged for easier visualization. MyelTracer was

217 run on a MacBook Pro (2.8 GHz, 16 GB RAM, Intel Iris Plus Graphics 655 1536 MB and

218 macOS Catalina Version 10.15.2.), Windows 10 (Intel Core i5-7300HQ CPU 2.50GHz, 8GB

219 RAM, NVIDIA GeForce GTX 1050 Ti) and MacBook Air (Intel Core i5 1.3 GHz CPU, 4GB

220 RAM, Intel HD Graphics 5000, macOS Catalina Version 10.15.2).

221

222 Results

223 MyelTracer is a semi-automated software using computer vision

224 G-ratio quantification requires measuring the cross-sectional areas of axons and myelin

225 sheaths, which is typically performed via manual tracing. Done manually, this process can be

226 immensely time-consuming and laborious. To facilitate g-ratio quantification, we created

227 MyelTracer by wrapping the powerful Python libraries provided by OpenCV in an easy-to-use

228 GUI using PyQt5. MyelTracer performs automatic detection of contours using dynamic

229 thresholding and separation of adjacent myelin sheaths based on user input. We packaged

230 MyelTracer as a readily installable software for Windows and Mac OS X operating systems.

231 Conceptually, MyelTracer takes a raw electron micrograph input image (Figure 1A), converts it

232 to grayscale, applies a bilateral filter, and thresholds it to a black and white binary image (Figure

233 1B). MyelTracer extracts boundaries between the black and white pixels on the binary image and

234 creates a contour image based on user-defined filtering parameters (Figure 1C). For g-ratio

235 plotting, MyelTracer is configured to measure the inner and outer myelin sheath as well as the

236 axon diameter to account for periaxonal space present in electron micrographs (Figure 1A).

237 Using the GUI (Figure 1G), the user can select the contours of axon, inner myelin, and outer

238 myelin while adjusting the threshold to closely match the displayed contours to the underlying

239 myelin ultrastructure. Once selected, MyelTracer groups the corresponding axons, inner myelin,

240 and outer myelin together and overlays those selections on top of the original image (Figure 1D).

241 The software automatically computes diameters of perfect circles from area measurements and

242 derives g-ratios by dividing inner myelin diameters by outer myelin diameters, which can then be

243 plotted against the axon diameter (Figure 1E, F).

244 To use MyelTracer efficiently, we recommend the following workflow (Figure 1H). First,

245 select all axons in the image by adjusting the threshold parameter to match contours to the axon

246 shape. Second, adjust the threshold to match myelin sheaths. Third, check for contiguous

247 contours of adjacent myelin sheaths as well as discontinuous contours around myelin sheaths. If

248 necessary, use the cut tool to separate adjacent myelin sheaths, or use the draw tool to connect

249 discontinuous myelin sheaths. Fourth, select the inner and outer myelin sheath contours,

250 adjusting the threshold parameter to fit. Fifth, save the progress and export the result, which

251 includes g-ratio measurements and an overlay (Figure 1D). MyelTracer also provides a manual

252 tracing option for otherwise untraceable features and a counting tool to calculate the percentage

253 of myelinated axons. Together, these functions make MyelTracer an all-in-one software suite for

254 studying myelin ultrastructure.

255

256 MyelTracer saves quantification time while maintaining accuracy

257 Myelin g-ratio quantification is routinely done by manually tracing the myelin sheath.

258 Manual tracing is very time-consuming but considered the standard practice for accurate results.

259 To compare the accuracy of MyelTracer to manual tracing in moderately myelinated tissues, we

260 measured g-ratios using both methods in the same electron micrographs of corpus callosum from

261 one-month-old mice (Figure 1I). Plotting the g-ratio as a function of axonal diameter, we did not

262 find a significant difference between MyelTracer and manual quantification results (Figure 1J,

263 MyelTracer: slope 0.061, y-intercept 0.77, n = 173, Manual: slope 0.057, y-intercept 0.77, n =

264 173; slopes p = 0.8996, intercept p = 0.7044a) and their averages (Figure 1K, MyelTracer mean ±

265 SD: 0.8072 ± 0.04417, n = 173, Manual: 0.8066 ± 0.04220, p = 0.9045b). Next, to examine the

266 amount of quantification time MyelTracer saves, the average amount of quantification time spent

267 per axon was compared between these two methods. Using MyelTracer, significantly less time

268 was spent per axon compared to manual tracing (Figure 1L, MyelTracer mean: 0.593, Manual:

269 mean 1.0, n = 173 axons, 3 images, p = 0.0069c). Further, to exemplify that MyelTracer can trace

270 myelin sheets containing folds, several such sheets were analyzed. Taking advantage of the cut

271 tool, MyelTracer readily detects myelin folds and correctly quantifies the myelin sheet without

272 including the area in the fold (Figure 1M-P). Together, these data show that MyelTracer

273 maintains a high level of accuracy, comparable to manual tracing, while saving about 40% of

274 time for the user.

275 Myelination is spatiotemporally regulated (Mayoral and Chan, 2016; Sun et al., 2018). As

276 a result, there are variations in myelin sheath thickness and spatial density of myelinated axons

277 across different types of tissues and developmental time points. To assess the feasibility of using

278 MyelTracer for different types of tissue samples, electron micrographs from optic nerves and

279 sciatic nerves were examined (Figure 2A, E). In the optic nerve of two-month-old mice, plotting

280 g-ratio as a function of axon diameter revealed no significant difference between the results from

281 using MyelTracer and those from manual quantification (Figure 2B, MyelTracer: slope 0.1046,

282 y-intercept 0.7372, n = 193, Manual: slope 0.1001, y-intercept 0.7427, n = 193, slopes p =

283 0.8311, intercept p = 0.4695d). We did not find a significant difference in g-ratio (Figure 2C,

284 MyelTracer mean ± SD: 0.7976 ± 0.04328, n = 193, Manual: 0.8018 ± 0.04244, p = 0.3406e), but

285 a significant reduction in time spent quantifying g-ratio (Figure 2D, MyelTracer mean: 0.4808,

286 Manual: mean 1.0, n = 193 axons, 3 images, p = 0.0046f). In the sciatic nerve of one-month-old

287 mice, regression relationships between g-ratio and axon diameter were not significantly different

288 between using MyelTracer and manual quantification (Figure 2F, MyelTracer: slope -0.0386, y-

289 intercept 0.627, n = 104, Manual: slope -0.030, y-intercept 0.615, n = 104, slopes p = 0.7437,

290 intercept p = 0.8997g). Comparing the results from using MyelTracer and those from manual

291 quantification in the sciatic nerve, there was no difference in g-ratio (Figure 2G, MyelTracer

292 mean ± SD: 0.5586 ± 0.07121, n = 104, Manual: 0.5576 ± 0.06833, p = 0.9146h), while there

293 was a significant reduction in time spent quantifying g-ratio (Figure 2H, MyelTracer mean

294 0.4054, Manual mean 1, n = 104 axons, 6 images, p < 0.0001i). Together, these findings

295 demonstrate that MyelTracer saves 40-60% user quantification time and offers accuracy levels

296 that are comparable to manual quantification.

297

298 Myelin thickness abnormalities can be detected using MyelTracer

299 Using MyelTracer for g-ratio quantification saves a significant amount of time and produces

300 accurate results across different types of tissues from wild-type mice (Figures 1, 2). In research

301 settings, however, most investigators seek a tool to detect differences between genotypes or

302 treatment paradigms. To test whether MyelTracer is suitable for detecting differences between

303 two experimental groups, its performance was tested using electron micrographs from control

304 (Gtf2ifl/fl, Nex-Cre-/-) and Gtf2ifl/fl; Nex-Cre+/- mice, a model for Williams Syndrome with a

305 known hypomyelination phenotype (Barak et al., 2019). With their genotypes blinded, electron

306 micrographs of corpus callosum from both groups at one-month of age were analyzed using

307 MyelTracer (Figure 3A, B). Regression relationships between g-ratio and axon diameter were

308 significantly different between control and Gtf2ifl/fl; Nex-Cre+/-mice (Figure 3C, Control: slope

309 0.1649, y-intercept 0.6447, n = 332 axons, n = 3 mice, Gtf2ifl/fl; Nex-Cre+/-: slope 0.1487, y-

310 intercept 0.7132, n = 787, n = 3 mice, slopes p = 0.1484, intercept p < 0.001J). The g-ratio

311 measurements were higher in Gtf2ifl/fl; Nex-Cre+/- mice than those in control mice, indicating a

312 reduction in myelin sheath thickness, a result consistent with the original report (Barak et al.,

313 2019). Together, these data demonstrate that MyelTracer is suitable for detecting differences in

314 myelin g-ratios across different genotypes.

315 316 MyelTracer can be used to quantify g-ratios in remyelination

317 MyelTracer readily recognizes well-developed myelin sheaths in micrographs of both central and

318 peripheral nerves and axon tracts. In contrast to the relatively thick and compact myelin in these

319 tissues, myelin in remyelinating tissues after injury is characterized by a thinner morphology

320 (Franklin and ffrench-Constant, 2008; Duncan et al., 2017). To test MyelTracer’s ability to

321 quantify g-ratios in tissues undergoing remyelination, we analyzed micrographs from an optic

322 nerve crush model 28 days after the injury (Figure 3D). Comparing the results from MyelTracer

323 and those from manual quantification in this model, there was no difference in g-ratio (Figure

324 3E, MyelTracer mean ± SD: 0.8212 ± 0.06843, n = 155, Manual: 0.8210 ± 0.06944, p =

325 0.9699k). Further, regression relationships between g-ratio and axon diameter were not

326 significantly different between MyelTracer and manual quantification (Figure 3F, MyelTracer:

327 slope 0.1484, y-intercept 0.7150, n = 155, Manual: slope 0.1282, y-intercept 0.7273, n = 155,

328 slopes p = 0.1511, intercept p = 0.8686l). To further test whether MyelTracer is suitable to detect

329 the thinned myelin sheets in remyelination compared to native myelin sheets, its performance

330 was tested using region-matched electron micrographs from remyelinated axons and axons from

331 matched native control animals (Figure 3G). Comparison of the average g-ratio per animal

332 revealed significantly increased g-ratios in the remyelinated group compared to the native control

333 group (Figure 3H, Native control mean ± SD: 0.7821 ± 0.01162, n = 3 mice, Remyelinated:

334 0.8324 ± 0.01942, n = 3 mice, p = 0.0183m). Regression relationships between g-ratio and axon

335 diameter were significantly different between the optic nerve axons in native control and

336 remyelinated mice (Figure 3I, Native control: slope 0.03137, y-intercept 0.7627, n = 183 axons,

337 Remyelinated: slope 0.1281, y-intercept 0.7449, n = 116, n = 3 mice, slopes p < 0.0001n).

338 Together, these results indicate that MyelTracer is suitable for g-ratio measurements in

339 remyelination.

340 Discussion

341 In this study, we show that MyelTracer can address the need for an easy-to-use, stand-

342 alone software for streamlining g-ratio quantification. OpenCV, PyQt5 and fbs were used in

343 Python to develop and package MyelTracer, an easy to install software with an intuitive GUI.

344 MyelTracer utilizes contrast in electron micrographs to generate contours around axons and

345 myelin sheaths. The user can select contours outlining myelin features of interest (axons, inner

346 myelin, and outer myelin), which then allow MyelTracer to group corresponding features,

347 measure their areas and calculate g-ratios. As an all-in-one platform for myelin analyses,

348 MyelTracer can also be used to calculate the percentage of myelinated axons. Using MyelTracer,

349 user quantification time is reduced by 40-60% depending on the region, and its accuracy is

350 comparable to manual tracing. Using MyelTracer, a known hypomyelination phenotype of the

351 Williams Syndrome mouse model can be detected, confirming its suitability for studies assessing

352 phenotypic changes in myelin sheaths. MyelTracer is freely available at [GitHub link, blinded

353 for review].

354 Myelin ultrastructure is highly heterogeneous with a wide range of axon calibers and

355 myelin sheath morphologies, presenting a challenge for automated analysis. MyelTracer uses

356 OpenCV to automatically contour cross-sections of axons and myelin sheaths in electron

357 micrographs (Figure 1A-C). The easy-to-use GUI allows users to intuitively interact with and

358 fine-tune the results generated by the underlying computer vision algorithms (Figure 1D, G).

359 MyelTracer decreases analysis time by dramatically accelerating axon and myelin sheath

360 contouring compared to manual tracing. This automated feature differentiates MyelTracer from

361 existing non-automated tools, such as GRatio for ImageJ, which is an ImageJ plugin that

362 facilitates grouping of traced structures and data export (Goebbels et al., 2010).

363 In addition to MyelTracer, several semi-automated and automated tools are available that

364 exhibit different strengths and weaknesses. AxonSeg and AxonDeepSeg are MATLAB-based

365 segmentation algorithms that are fully automated, enabling high-throughput measurements

366 (Zaimi et al., 2016, 2018). These tools are useful for processing large datasets but require image

367 pre-processing, adjustment of parameters in MATLAB, or additional training of the neural

368 network using manually labeled images from user-specific datasets, which can be hurdles for

369 implementation. Janjic et al. developed a high-fidelity tool using deep neural networks, which

370 suits researchers with a computer science background seeking an advanced g-ratio quantification

371 tool (Janjic et al., 2019). While there are other useful tools available, MyelTracer’s intuitive GUI

372 makes it a powerful tool allowing researchers to quantify myelin without computer science

373 knowledge.

374 G-ratio quantification using MyelTracer was demonstrated to be accurate across different

375 tissues, such as corpus callosum, optic nerve, and sciatic nerve (Figures 1, 2). For these types of

376 samples, there was a 40-60% reduction in time spent on quantification (Figure 1L, 2D, 2H). The

377 variation in the amount of time-reduction depended on the type of tissue. The smallest time-

378 reduction was for corpus callosum of one-month-old mice. This is likely due to the heterogeneity

379 in axonal caliber, axonal density, and the contrast level among the axon, periaxonal space, inner

380 tongue, and the myelin sheath (Figure 1A, I). In contrast, the greatest time-reduction was

381 observed for sciatic nerve, which has clearly defined myelinated axons with homogeneous

382 morphology, minimal periaxonal space and minimal myelin contiguity between adjacent axons

383 (Figure 2E, H). Comparing MyelTracer workflows for corpus callosum versus sciatic nerve, a

384 major difference was that the latter required less manual input to separate contiguous myelin

385 sheaths (Step 3, Figure 1H). Similar to manual tracing and other computational toolboxes,

386 MyelTracer also relies on the quality of input images and therefore shows less effective

387 performance for images with poor contrast levels, artifacts from inadequate perfusion or tissue

388 handling.

389 The analysis of electron micrographs from Williams Syndrome mice shows that

390 MyelTracer readily detects abnormalities in myelination (Figure 3A-C), demonstrating the

391 suitability of MyelTracer for detecting differences in myelin g-ratio across different genotypes or

392 treatment groups. In such studies, MyelTracer will decrease user quantification time, and

393 streamline data generation for publication purposes.

394 Extending the application for MyelTracer, analysis of remyelinating axons 28 days after

395 optic nerve crush shows that the software is suitable for g-ratio measurements in studies of

396 remyelination (Figure 3D-I). Myelin sheaths in remyelinating tissues are generally found to be

397 thinner (Franklin and ffrench-Constant, 2008; Duncan et al., 2017), and they can appear almost

398 discontinuous in a given sectioning plane. This presents a challenge to automation and may

399 require user input using MyelTracer’s Cut and Draw tools as extensively outlined in the Users’

400 Manual (Extended Data 2). Notwithstanding these challenges, MyelTracer is suitable for the

401 semi-automated tracing of micrographs in studies of remyelination provided that the myelin

402 sheaths and axons display good contrast.

403 As a user-friendly all-in-one software suite, MyelTracer will also allow researchers to

404 quantify the percentage of myelinated axons, which together with g-ratio, are the most

405 commonly reported metrics in myelin studies. Beyond existing tools and MyelTracer, advances

406 in computer vision and artificial intelligence may further advance data analysis methods and

407 produce more powerful toolboxes to streamline quantitative analysis of myelin ultrastructure.

408

409 410 Figure 1 MyelTracer semi-automatically traces axons and myelin and returns results

411 consistent with manual tracing. A-D, Electron micrograph of dorsal corpus callosum, software-

412 generated threshold overlays, and post-selection overlay (blue) show different steps of the image

413 analysis. * periaxonal space. E, Electron micrograph with MyelTracer overlay and outlines for

414 features including axon diameter (a), inner myelin diameter (d), and outer myelin diameter (D),

415 each computed from the respective areas. F, Schematic for g-ratio scatterplot illustrating how

416 MyelTracer calculates the g-ratio. G, Image of the main user interface of MyelTracer. H,

417 Schematic showing an example workflow using MyelTracer. I, Representative electron

418 micrograph of corpus callosum from one-month-old wild-type mice. J, g-ratio scatter plot from

419 manual and MyelTracer quantifications for the same axons. ANCOVA test p-value=0.8996

420 (slopes) and p=0.7044 (intercepts) , n=174 axons. K, g-ratios from manual and MyelTracer

421 quantification paired for the same axons. Student’s t-test p-value=0.9045, n=174 axons. L,

422 Comparison of time consumption between manual quantification and quantification using

423 MyelTracer. Individual data points represent the average time spent per axon per image. p-

424 value=0.0069; Student’s t-test. Scale bar, 1 μm. M-P, Illustration of MyelTracer’s ability to

425 quantify myelin sheets with folds (arrowheads). Folds require the user to draw a white line

426 (arrows) with the cut tool to connect the area inside the fold with the outside area for semi-

427 automated detection. Scale bar, 1 μm.

428

429

430

431 432 Figure 2 MyelTracer accurately returns g-ratios for myelinated tissues of varying axonal

433 density. A, Representative electron micrograph of the optic nerve at two months of age (2M)

434 with MyelTracer-generated overlays (blue). Scale bar, 1 μm. B, g-ratio scatter plot from manual

435 and MyelTracer quantifications for the same axons. ANCOVA test p-value=0.8311 (slopes) and

436 p=0.4695 (intercepts), n=193 axons. C, g-ratios from manual and MyelTracer quantifications

437 paired for the same axons. p=0.3406, n=193 axons. D, Comparison of time consumption between

438 manual quantification and quantification using MyelTracer. Individual data points represent the

439 average time spent per axon per image. p-value=0.0046. E, Representative electron micrograph

440 of the sciatic nerve at one month of age with MyelTracer-generated overlays (blue). Scale bar, 2

441 μm. F, g-ratio scatter plot from manual and MyelTracer quantifications for the same axons.

442 ANCOVA test p-value=0.7437 (slopes) and p=0.8997 (intercepts), n=104 axons. G, g-ratios

443 from manual and MyelTracer quantifications paired for the same axons. P=0.9146, n=104 axons.

444 H, Comparison of time consumption between manual quantification and quantification using

445 MyelTracer. Individual data points represent the average time spent per axon per image. p-

446 value<0.0001; Student’s t-test.

447

448

449

450

451 452 Figure 3 MyelTracer can be used to detect a myelin thickness abnormality in a Williams

453 Syndrome mouse model and to measure g-ratios in remyelination.

454 A-B, Representative electron micrographs of the corpus callosum in control and Gtf2i fl/fl; Nex-

455 Cre mice at one month of age with MyelTracer-generated overlays (blue). Scale bar, 1 μm. C, g-

456 ratio scatter plot from using MyelTracer for control and Gtf2i fl/fl; Nex-Cre mice. p-

457 value<0.0001; ANCOVA test; n=3 mice, 332 axons for control and n=3 mice, 455 axons for

458 Gtf2i fl/fl; Nex-Cre mice. D, Representative electron micrograph of the optic nerve 28 days after

459 optic nerve crush with MyelTracer-generated overlays (blue). Scale bar, 1 μm. E, g-ratios from

460 manual and MyelTracer quantifications paired for the same axons. P=0.9699, n=155 axons. F, g-

461 ratio scatter plot from manual and MyelTracer quantifications for the same axons. ANCOVA test

462 p-value=0.1511 (slopes) and p=0.8686 (intercepts), n=155 axons. G, Representative electron

463 micrograph of the native control optic nerve with MyelTracer-generated overlays (blue). H, g-

464 ratio group comparison of axons from native control and remyelinated axons from mice

465 following optic nerve crush. p-value<0.0183; t-test; n=3 mice per group. I, g-ratio scatter plot of

466 axons from native control and remyelinated axons from mice following optic nerve crush.

467 ANCOVA test p-value < 0.0001 (slopes), Native control n=183 axons, Remyelinated n=116

468 axons.

469 Extended Data 1

470

471 We recommend downloading MyelTracer directly from the GitHub repository, at [GitHub link,

472 blinded for review]. Download Extended Data 1 (containing source code).

473

474 Extended Data 2

475

476 Extended data 2 contains a Users’ Manual for MyelTracer and some illustrative images

477 regarding micrograph quality.

478

479

480 Table 1

481

482

Data Structure Type of Test Power

a Normal distribution ANCOVA Slopes:

Myeltracer 95% CI 0.02018 to 0.1021 vs manual 0.01402

to 0.1006

Intercepts:

Myeltracer 95% CI 0.7539 to 0.7976 vs manual 0.7521 to

0.7997

b Normal distribution Student’s t-test 95% CI of difference -0.009693 to 0.008578

c Normal distribution Student’s t-test 95% CI of difference -0.6277 to -0.1858

d Normal distribution ANCOVA Slopes:

Myeltracer 95% CI 0.07485 to 0.1343 vs manual 0.07089

to 0.1293

Intercepts:

Myeltracer 95% CI 0.7192 to 0.7552 vs manual 0.7246 to

0.7608

e Normal distribution Student’s t-test 95% CI of difference -0.01274 to 0.004416

f Normal distribution Student’s t-test 95% CI of difference -0.6713 to -0.3670

g Normal distribution ANCOVA Slopes:

Myeltracer 95% CI -0.06009 to -0.01153 vs manual -

0.05325 to -0.007339

Intercepts:

Myeltracer 95% CI 0.5788 to 0.6754 vs manual 0.5695 to

0.6612 h Normal distribution Student’s t-test 95% CI of difference -0.01804 to 0.02012 i Normal distribution Student’s t-test 95% CI of difference -0.6726 to -0.5166 j Normal distribution ANCOVA Slopes:

Control 95% CI 0.1481 to 0.1817 vs Gtf2i fl/fl; Nex-Cre

0.1340 to 0.1633

Intercepts:

Control 95% CI 0.6323 to 0.6572 vs Gtf2i fl/fl; Nex-Cre

0.7044 to 0.7220 k Normal distribution Student’s t-test 95% CI of difference -0.01511 to 0.01570

l Normal distribution ANCOVA Slopes:

Manual 95% CI 0.1009 to 0.1556 vs MyelTracer 0. 0.1355

to 0.1613

Intercepts:

Manual 95% CI 0. 0.7066 to 0.7481 vs MyelTracer 0.7068

to 0.7233

m Normal distribution Student’s t-test 95% CI of difference 0.01403 to 0.08659

n Normal distribution ANCOVA Slopes:

Native control 95% CI -0.001234 to 0.06396 vs

Remyelination 0.09512 to 0.1611

483

484 References

485

486 Barak B, Zhang Z, Liu Y, Nir A, Trangle SS, Ennis M, Levandowski KM, Wang D, Quast K, Boulting GL (2019) 487 Neuronal deletion of Gtf2i, associated with Williams syndrome, causes behavioral and myelin alterations 488 rescuable by a remyelinating drug. Nature neuroscience 22:700–708.

489 Bégin S, Dupont-Therrien O, Bélanger E, Daradich A, Laffray S, De Koninck Y, Côté DC (2014) Automated 490 method for the segmentation and morphometry of nerve fibers in large-scale CARS images of spinal cord 491 tissue. Biomedical optics express 5:4145.

492 Duncan ID, Marik RL, Broman AT, Heidari M (2017) Thin myelin sheaths as the hallmark of remyelination persist 493 over time and preserve axon function. Proceedings of the National Academy of Sciences 114:E9685– 494 E9691.

495 Duncan ID, Radcliff AB (2016) Inherited and acquired disorders of myelin: the underlying myelin pathology. 496 Experimental neurology 283:452–475.

497 Franklin RJM, ffrench-Constant C (2008) Remyelination in the CNS: from biology to therapy. Nature Reviews 498 Neuroscience 9:839–855.

499 Gibson EM, Purger D, Mount CW, Goldstein AK, Lin GL, Wood LS, Inema I, Miller SE, Bieri G, Zuchero JB 500 (2014) Neuronal activity promotes oligodendrogenesis and adaptive myelination in the mammalian brain. 501 Science 344.

502 Goebbels S, Oltrogge JH, Kemper R, Heilmann I, Bormuth I, Wolfer S, Wichert SP, Möbius W, Liu X, Lappe- 503 Siefke C (2010) Elevated phosphatidylinositol 3, 4, 5-trisphosphate in glia triggers cell-autonomous 504 membrane wrapping and myelination. Journal of Neuroscience 30:8953–8964.

505 Hughes EG, Appel B (2016) The cell biology of CNS myelination. Current opinion in neurobiology 39:93–100.

506 Hughes EG, Orthmann-Murphy JL, Langseth AJ, Bergles DE (2018) Myelin remodeling through experience- 507 dependent oligodendrogenesis in the adult somatosensory cortex. Nature neuroscience 21:696–706.

508 Janjic P, Petrovski K, Dolgoski B, Smiley J, Zdravkovski P, Pavlovski G, Jakjovski Z, Davceva N, Poposka V, 509 Stankov A (2019) Measurement-oriented deep-learning workflow for improved segmentation of myelin 510 and axons in high-resolution images of human cerebral white matter. Journal of neuroscience methods 511 326:108373.

512 Krajewski KM, Lewis RA, Fuerst DR, Turansky C, Hinderer SR, Garbern J, Kamholz J, Shy ME (2000) 513 Neurological dysfunction and axonal degeneration in Charcot–Marie–Tooth disease type 1A. Brain 514 123:1516–1527.

515 Lin W, Popko B (2009) Endoplasmic reticulum stress in disorders of myelinating cells. Nature Neuroscience 516 12:379–385.

517 Mayoral SR, Chan JR (2016) The environment rules: spatiotemporal regulation of oligodendrocyte differentiation. 518 Current opinion in neurobiology 39:47–52.

519 McKenzie IA, Ohayon D, Li H, De Faria JP, Emery B, Tohyama K, Richardson WD (2014) Motor skill learning 520 requires active central myelination. science 346:318–322.

521 More HL, Chen J, Gibson E, Donelan JM, Beg MF (2011) A semi-automated method for identifying and measuring 522 myelinated nerve fibers in scanning electron microscope images. Journal of Neuroscience Methods 523 201:149–158.

524 Nave K-A (2010) Myelination and support of axonal integrity by glia. Nature 468:244–252.

525 Phan BN et al. (2020) A myelin-related transcriptomic profile is shared by Pitt–Hopkins syndrome models and 526 human autism spectrum disorder. Nature Neuroscience 23:375–385.

527 Rushton WAH (1951) A theory of the effects of fibre size in medullated nerve. J Physiol 115:101–122.

528 Salzer JL (2015) Schwann cell myelination. Cold Spring Harbor perspectives in biology 7:a020529.

529 Sander S, Ouvrier R, McLeod J, Nicholson G, Pollard J (2000) Clinical syndromes associated with tomacula or 530 myelin swellings in sural nerve biopsies. Journal of Neurology, Neurosurgery & Psychiatry 68:483–488.

531 Sherman DL, Brophy PJ (2005) Mechanisms of axon ensheathment and myelin growth. Nature Reviews 532 Neuroscience 6:683–690.

533 Simons M, Trotter J (2007) Wrapping it up: the cell biology of myelination. Current opinion in neurobiology 534 17:533–540.

535 Stadelmann C, Timmler S, Barrantes-Freer A, Simons M (2019) Myelin in the central nervous system: structure, 536 function, and pathology. Physiological reviews 99:1381–1431.

537 Sun LO, Mulinyawe SB, Collins HY, Ibrahim A, Li Q, Simon DJ, Tessier-Lavigne M, Barres BA (2018) 538 Spatiotemporal control of CNS myelination by oligodendrocyte programmed cell death through the TFEB- 539 PUMA axis. Cell 175:1811–1826.

540 Wang J, He X, Meng H, Li Y, Dmitriev P, Tian F, Page JC, Lu QR, He Z (2020) Robust myelination of regenerated 541 axons induced by combined manipulations of GPR17 and microglia. Neuron 108:876–886.

542 Zaimi A, Duval T, Gasecka A, Côté D, Stikov N, Cohen-Adad J (2016) AxonSeg: open source software for axon 543 and myelin segmentation and morphometric analysis. Frontiers in neuroinformatics 10:37.

544 Zaimi A, Wabartha M, Herman V, Antonsanti P-L, Perone CS, Cohen-Adad J (2018) AxonDeepSeg: automatic 545 axon and myelin segmentation from microscopy data using convolutional neural networks. Scientific 546 reports 8:1–11.

547 Zhao C, Dong C, Frah M, Deng Y, Marie C, Zhang F, Xu L, Ma Z, Dong X, Lin Y (2018) Dual requirement of 548 CHD8 for chromatin landscape establishment and histone methyltransferase recruitment to promote CNS 549 myelination and repair. Developmental cell 45:753–768.

550

551

27 AB C D Raw Threshold Contours Overlay

* 2

E F GH g-ratio calculation and plotting GUI Workflow example

Step 1: Enter calibration factor

Step 2: Set a threshold and mark all axons to be quantified

Step 3: Use cut/draw tools to separate/connect myelin sheets

d Step 4: Change threshold and mark -ratio (d/D) inner and outer myelin boundaries at g D once for each axon a Axon diameter (a) Step 5: Save ROIs and export results

I J KL

1M WT Corpus Callosum

1.0 1.0 1.0 Manual 7 MyelTracer n.s. 1 23 0.8 ** 0.9 0.9

4 0.6 0.8 0.8 -ratio -ratio g g 0.4

5 0.7 0.7

8 relative time per axon 0.2 2 6 0.6 0.6 0.0 0.0 0.3 0.6 0.9 1.2 Manual MyelTracer Manual MyelTracer Axon diameter (μm)

MNMyelin fold Traced myelin fold O Myelin fold P Traced myelin fold

1 1 AB2M WT Optic Nerve C D

1.0 1.0 n.s. 1.0 1 20 Manual 6 14 24 12 19 MyelTracer 0.8 7 15 0.9 21 0.9 2 13 ** 8 0.6 9 16 3 25 0.8 0.8 -ratio -ratio g g 0.4

4 10 22 26 0.7 0.7

17 relative time per axon 0.2

5 11 18 23 27 0.6 0.6 0.0 0.0 0.5 1.0 1.5 Manual MyelTracer Manual MyelTracer Axon diameter ( m) EF1M WT Sciatic Nerve G H

0.8 0.8 n.s. 1.0 5 Manual 1 9 MyelTracer 2 0.7 0.7 0.8 6 3 10 14 21 0.6 *** 7 15 0.6 0.6 11

16 -ratio -ratio g

0.5 g 12 0.5 0.4 17 4 8 0.4

0.4 relative time per axon 0.2 13 18 22 19 0.3 20 01234 0.3 0.0 Manual MyelTracer Manual MyelTracer Axon diameter ( m) A Control BCGtf2i fl/fl; Nex-Cre 5 21 1 1.0 14 4 1 10 10 6 11 13 2 15 11 18 6 0.8 22 7 7 14 3

2 ratio -

12 g 16 23 15 0.6 Control 8 19 5 8 Gtf2i fl/fl; Nex-Cre 12 13 16 20 4 9 3 9 0.4 17 0.0 0.5 1.0 1.5 2.0 Axon diameter (μm) D Remyelination after optic nerve crush E F

1.0 6 1.0 n.s. 1 3 4 1 3 4 0.8 0.8 2 7 11 5 -ratio -ratio 2 g g 0.6 0.6 5 Manual

6 MyelTracer 8 0.4 0.4 7 9 Manual MyelTracer11 0.0 0.5 1.0 1.5 2.0 12 Axon diameter ( m) GHINative optic nerve (no crush) 10 13 14

1 38 15 16 1.0 16 39 37 1.0 4 7 2 36 * 3 6 18 17 40 35 0.9 0.9 33 34 5 15 41 32 31 0.8 0.8 19 29 30 -ratio -ratio g g 0.7 14 28 0.7 8 27 Native 10 13 20 26 0.6 0.6 23 Remyelinated 9 21 25 12 22 24 11 0.5 0.5 Native Remyelinated 0.0 0.5 1.0 1.5 2.0 Axon diameter ( m)

Table 1

Data Structure Type of Test Power a Normal distribution ANCOVA Slopes:

Myeltracer 95% CI 0.02018 to 0.1021 vs manual 0.01402

to 0.1006

Intercepts:

Myeltracer 95% CI 0.7539 to 0.7976 vs manual 0.7521 to

0.7997 b Normal distribution Student’s t-test 95% CI of difference -0.009693 to 0.008578 c Normal distribution Student’s t-test 95% CI of difference -0.6277 to -0.1858 d Normal distribution ANCOVA Slopes:

Myeltracer 95% CI 0.07485 to 0.1343 vs manual 0.07089

to 0.1293

Intercepts:

Myeltracer 95% CI 0.7192 to 0.7552 vs manual 0.7246 to

0.7608 e Normal distribution Student’s t-test 95% CI of difference -0.01274 to 0.004416 f Normal distribution Student’s t-test 95% CI of difference -0.6713 to -0.3670

g Normal distribution ANCOVA Slopes:

Myeltracer 95% CI -0.06009 to -0.01153 vs manual -

0.05325 to -0.007339

Intercepts:

Myeltracer 95% CI 0.5788 to 0.6754 vs manual 0.5695 to

Control 95% CI 0.1481 to 0.1817 vs Gtf2i fl/fl; Nex-Cre

0.1340 to 0.1633

Intercepts:

Control 95% CI 0.6323 to 0.6572 vs Gtf2i fl/fl; Nex-Cre

0.7044 to 0.7220 k Normal distribution Student’s t-test 95% CI of difference -0.01511 to 0.01570

l Normal distribution ANCOVA Slopes:

Manual 95% CI 0.1009 to 0.1556 vs MyelTracer 0. 0.1355

to 0.1613

Intercepts:

Manual 95% CI 0. 0.7066 to 0.7481 vs MyelTracer 0.7068

to 0.7233

m Normal distribution Student’s t-test 95% CI of difference 0.01403 to 0.08659

n Normal distribution ANCOVA Slopes:

Native control 95% CI -0.001234 to 0.06396 vs

Remyelination 0.09512 to 0.1611