Disrupted brain subnetworks in early demyelinating clinically isolated syndrome
Poster No:
1518
Submission Type:
Abstract Submission
Authors:
Michael Foster1, Ferran Prados1,2, Sara Collorone1, Baris Kanber1, Indran Davagnanam1, Done-Helen Dogan1, Niamh Cawley1, Marios Yiannakas1, Lola Ogunbowale3, Ailbhe Burke3, Fulvia Palesi4, Frederik Barkhof1,5, Claudia Gandini Wheeler-Kingshott1,6, Olga Ciccarelli1, Wallace Brownlee1, Ahmed Toosy1
Institutions:
1University College London, London, United Kingdom, 2Universitat Oberta de Catalunya, Barcelona, Spain, 3Moorfields Eye Hospital NHS Foundation Trust, London, United Kingdom, 4Univerista di Pavia, Pavia, Italy, 5Amsterdam University Medical Centre, Amsterdam, Noord-Holland, 6Universita di Pavia, Pavia, Italy
First Author:
Co-Author(s):
Ferran Prados
University College London|Universitat Oberta de Catalunya
London, United Kingdom|Barcelona, Spain
University College London|Universitat Oberta de Catalunya
London, United Kingdom|Barcelona, Spain
Frederik Barkhof, MD, Ph. D
University College London|Amsterdam University Medical Centre
London, United Kingdom|Amsterdam, Noord-Holland
University College London|Amsterdam University Medical Centre
London, United Kingdom|Amsterdam, Noord-Holland
Claudia Gandini Wheeler-Kingshott
University College London|Universita di Pavia
London, United Kingdom|Pavia, Italy
University College London|Universita di Pavia
London, United Kingdom|Pavia, Italy
Introduction:
Here we assess the relationship of brain subnetwork connectomes with disability in the demyelinating clinically isolated syndrome (CIS).
Methods:
73 people with early CIS (pwCIS) were assessed for motor, visual and cognitive performance. Multi-shell diffusion-weighted imaging (DWI) was acquired alongside conventional brain MRI sequences. 28 healthy controls (HC) underwent the same protocol. Brain lesion masks were created with Jim software on PD/T2-weighted images. 3D T1-weighted images were filled [1] and segmented [2] into cortical, white and deep grey matter regions, and registered to diffusion space after distortion correction. Activation maps of motor, visual and default mode networks derived from task-based functional MRI meta-analysis [3] were applied to participants' native DWI space. Constrained spherical deconvolution was performed with MRtrix3 [4] and fibre orientation distributions generated with multi-shell multi-tissue response functions. Anatomically-constrained probabilistic tractography generated tractograms with 3x107 streamlines, re-weighted with SIFT2 [5] to produce connectomes. Graph metrics were calculated with Brain Connectivity Toolbox.[6,7]
Statistical analysis was performed with R 4.3.1. Significance levels were set at P<0.05. Linear regression models compared graph metrics in each network between pwCIS and HC. Using 'glmulti' [8] to identify linear regression models, graph metrics were independently and in combination assessed for association with clinical outcomes, and the combination of conventional MRI metrics that best correlated with clinical outcomes was identified. Graph metrics were added to the best MRI models. Resulting coefficients were examined to determine if the relationship of each graph metric to disability was preserved after adjusting for conventional MRI; log-likelihood ratio tests compared nested model performance.
Statistical analysis was performed with R 4.3.1. Significance levels were set at P<0.05. Linear regression models compared graph metrics in each network between pwCIS and HC. Using 'glmulti' [8] to identify linear regression models, graph metrics were independently and in combination assessed for association with clinical outcomes, and the combination of conventional MRI metrics that best correlated with clinical outcomes was identified. Graph metrics were added to the best MRI models. Resulting coefficients were examined to determine if the relationship of each graph metric to disability was preserved after adjusting for conventional MRI; log-likelihood ratio tests compared nested model performance.
Results:
Global efficiency (GE), local efficiency (LE), clustering coefficient (CC), transitivity (TS) and node strength (NS) were all increased (P=0.010, 0.028, 0.033, 0.047 and 0.020 respectively) in motor networks of pwCIS; the assortativity coefficient (AC) and betweenness centrality (BC) were preserved. No differences between pwCIS and HC were found in visual or cognitive networks.
No significant clinical relationships were seen in motor networks. In visual networks, low-contrast letter acuity (LCLA) 2.5% had an inverse relationship with TS (P=0.034), and Farnsworth-Munsell 100-hue (FM-100) outcomes had a positive association with NS (P=0.049). In cognitive networks, symbol-digit modality testing (SDMT) had an inverse relationship with AC (P=0.026).
Structural MRI models of clinical outcomes are in Table 1. For T25FW, addition of AC improved model performance (P=0.033, ΔR2 5.3%); no other MRI models of motor outcomes were improved by graph metrics. In visual networks, addition of GE and NS to the MRI model for LCLA 2.5% improved performance (P=0.001, ΔR2 15.1%), and GE, NS and BC together strengthened models of LCLA 1.25% (P<0.001, ΔR2 20.4%). In cognitive networks, adding CC and NS to a model for brief visuospatial memory testing (BVMT-R) appeared to improve the MRI model, but did not reach significance (P=0.078, ΔR2 4.5%).
No significant clinical relationships were seen in motor networks. In visual networks, low-contrast letter acuity (LCLA) 2.5% had an inverse relationship with TS (P=0.034), and Farnsworth-Munsell 100-hue (FM-100) outcomes had a positive association with NS (P=0.049). In cognitive networks, symbol-digit modality testing (SDMT) had an inverse relationship with AC (P=0.026).
Structural MRI models of clinical outcomes are in Table 1. For T25FW, addition of AC improved model performance (P=0.033, ΔR2 5.3%); no other MRI models of motor outcomes were improved by graph metrics. In visual networks, addition of GE and NS to the MRI model for LCLA 2.5% improved performance (P=0.001, ΔR2 15.1%), and GE, NS and BC together strengthened models of LCLA 1.25% (P<0.001, ΔR2 20.4%). In cognitive networks, adding CC and NS to a model for brief visuospatial memory testing (BVMT-R) appeared to improve the MRI model, but did not reach significance (P=0.078, ΔR2 4.5%).
Conclusions:
The motor network was most disrupted against HC; no differences were seen in visual or cognitive networks. However, the relationship of graph metrics with clinical outcomes was stronger in visual and cognitive networks compared with motor; worse clinical outcomes were associated with increased graph metric values. Additionally, graph metrics strengthened MRI models of T25FW and LCLA.
The results suggest motor network disruption occurs earlier in the demyelinating disease course; the increase in graph metrics indicates compensatory reorganisation, but clinical associations suggest it is either inadequate or maladaptive. Altogether, the results demonstrate the utility of graph metrics in varied brain networks in CIS.
The results suggest motor network disruption occurs earlier in the demyelinating disease course; the increase in graph metrics indicates compensatory reorganisation, but clinical associations suggest it is either inadequate or maladaptive. Altogether, the results demonstrate the utility of graph metrics in varied brain networks in CIS.
Disorders of the Nervous System:
Neurodegenerative/ Late Life (eg. Parkinson’s, Alzheimer’s)
Modeling and Analysis Methods:
Connectivity (eg. functional, effective, structural) 1
Neuroanatomy, Physiology, Metabolism and Neurotransmission:
White Matter Anatomy, Fiber Pathways and Connectivity
Novel Imaging Acquisition Methods:
Diffusion MRI 2
Keywords:
Cognition
Demyelinating
Motor
MRI
Neurological
STRUCTURAL MRI
Vision
1|2Indicates the priority used for review
Provide references using author date format
[2] Cardoso MJ, Modat M, Wolz R, et al. Geodesic Information Flows: Spatially-Variant Graphs and Their Application to Segmentation and Fusion. IEEE Trans Med Imaging. 2015;34:1976–88.
[3] Yarkoni T, Poldrack RA, Nichols TE, et al. Large-scale automated synthesis of human functional neuroimaging data. Nat Methods. 2011;8:665–70.
[4] Tournier J-D, Smith RE, Raffelt D, et al. MRtrix3: A fast, flexible and open software framework for medical image processing and visualisation. Neuroimage. 2019;202:116137.
[5] Smith RE, Tournier J-D, Calamante F, et al. SIFT2: Enabling dense quantitative assessment of brain white matter connectivity using streamlines tractography. Neuroimage. 2015;119:338–51.
[6] Rubinov M, Sporns O. Complex network measures of brain connectivity: Uses and interpretations. Neuroimage. 2010;52:1059–69.
[7] Humphries MD, Gurney K. Network ‘Small-World-Ness’: A Quantitative Method for Determining Canonical Network Equivalence. PLoS One. 2008;3:e0002051.
[8] Calcagno V, de Mazancourt C. glmulti: An R Package for Easy Automated Model Selection with (Generalized) Linear Models. J Stat Softw. 2010;34:1–29.