Characterising structural similarity networks in the marmoset using Morphometric Inverse Divergence
Poster No:
433
Submission Type:
Abstract Submission
Authors:
Ed Hutchings1, Stephen Sawiak1, Richard Bethlehem2, Angela Roberts1, Edward Bullmore3
Institutions:
1University of Cambridge, Cambridge, Cambridgeshire, 2Autism Research Centre, Department of Psychiatry, University of Cambridge, Cambridge, United Kingdom, 3University of Cambridge, Cambridge, United Kingdom
First Author:
Co-Author(s):
Richard Bethlehem
Autism Research Centre, Department of Psychiatry, University of Cambridge
Cambridge, United Kingdom
Autism Research Centre, Department of Psychiatry, University of Cambridge
Cambridge, United Kingdom
Introduction:
Many psychiatric disorders are in part neurodevelopmental, motivating the need to characterise healthy brain development. MRI provides a non-invasive, non-ionising, and high detail way to track brain development. MRI has been used with network approaches, such as graph theory, which models regions as nodes and connections as edges [1]. Graph theoretic properties, such as hubs and modules, are thought to play specific roles in information processing and have been implicated in various mental disorders [1].
Structural similarity has increasingly been used to generate networks from structural MRI images, in which edges reflect genetic similarity as well as axonal connectivity [2]. Regions with high structural covariance (a type of similarity) tend to develop together as coordinated units, and disturbances in this coordination may play a role in the aetiology of psychiatric disorders [3]. This technique generates group level networks, and later approaches have aimed to generate networks within individuals. One such approach is Morphometric Inverse Divergence (MIND), which has been biologically validated in humans and macaques [4].
Understanding how structural similarity in early life relates to brain morphology and behaviour across development of individuals requires longitudinal imaging. We turned to the common marmoset as an animal model due to their short life history yet cortical and behavioural complexity [5]. As a preliminary analysis, we generated MIND networks from a mean MTsat image (a measure of myelin [6]) to assess biological validity and characterise network properties.
Structural similarity has increasingly been used to generate networks from structural MRI images, in which edges reflect genetic similarity as well as axonal connectivity [2]. Regions with high structural covariance (a type of similarity) tend to develop together as coordinated units, and disturbances in this coordination may play a role in the aetiology of psychiatric disorders [3]. This technique generates group level networks, and later approaches have aimed to generate networks within individuals. One such approach is Morphometric Inverse Divergence (MIND), which has been biologically validated in humans and macaques [4].
Understanding how structural similarity in early life relates to brain morphology and behaviour across development of individuals requires longitudinal imaging. We turned to the common marmoset as an animal model due to their short life history yet cortical and behavioural complexity [5]. As a preliminary analysis, we generated MIND networks from a mean MTsat image (a measure of myelin [6]) to assess biological validity and characterise network properties.
Methods:
N=119 marmosets were scanned longitudinally (Fig 1A) using a 9.4T system. Three 3D multi-gradient echo sequences (PDw/MTw/T1w) were acquired. Estimation of MTsat parameter maps followed previously published methods [6]. Preprocessing was performed using SPM12 in Matlab. The SPMMouse toolbox [7] was used with DARTEL to generate population templates. These were warped to create a mean MTsat map across all animals, and a cortical parcellation consisting of 232 regions was applied [8]. MIND networks were generated using code from [4].
Results:
Building and validating the network
Highest edge weights were found between homotypic interhemispheric regions (Fig. 1B, D). To assess the extent to which similarity was driven by distance between regions, we correlated the raw MIND matrix with a matrix formed from Euclidian distance between region centroids (Fig. 1C). There was a small negative correlation (r = -0.16, p = 1.022e-5), indicating a slight decay in similarity with distance.
Network analysis
Node strength distribution of the raw matrix was negatively skewed (Fig 2A). Hubs of the network (top 20 node strengths) concentrated in frontal and paracentral areas (Fig 2B). Lowest node strengths were found in the occipital lobe, with a decreasing gradient from V3 to V1. We clustered the network to see if we could identify modules with distinct myeloarchitecture. Optimal Louvain clustering found two modules (modularity = 0.0603), containing mainly superior temporal cortex (Fig. 2C). Hierarchical clustering found a smaller cluster, containing only visual and auditory parabelt regions (Fig. 2D). We performed a principal component analysis on edge weights. The first principal component explained 40% of the total variation in edge weights (Fig. 2E) separating auditory and visual regions from frontal and temporo-parietal association areas.
Highest edge weights were found between homotypic interhemispheric regions (Fig. 1B, D). To assess the extent to which similarity was driven by distance between regions, we correlated the raw MIND matrix with a matrix formed from Euclidian distance between region centroids (Fig. 1C). There was a small negative correlation (r = -0.16, p = 1.022e-5), indicating a slight decay in similarity with distance.
Network analysis
Node strength distribution of the raw matrix was negatively skewed (Fig 2A). Hubs of the network (top 20 node strengths) concentrated in frontal and paracentral areas (Fig 2B). Lowest node strengths were found in the occipital lobe, with a decreasing gradient from V3 to V1. We clustered the network to see if we could identify modules with distinct myeloarchitecture. Optimal Louvain clustering found two modules (modularity = 0.0603), containing mainly superior temporal cortex (Fig. 2C). Hierarchical clustering found a smaller cluster, containing only visual and auditory parabelt regions (Fig. 2D). We performed a principal component analysis on edge weights. The first principal component explained 40% of the total variation in edge weights (Fig. 2E) separating auditory and visual regions from frontal and temporo-parietal association areas.
·Figure 1: Building the network and testing validity
·Figure 2: Network properties
Conclusions:
MTsat MIND networks show strong interhemispheric similarity and decay with distance, suggesting they are biologically valid [4]. Principal component analysis identified a primary sensory to association gradient anchored in visual and auditory cortex at one end and fronto-temporal cortex at the other, accounting for 40% of variance in edge weights. This finding echoes the sensory-association gradient of myeloarchitecture found in human [9] and may suggest conserved organisational principals between species.
Disorders of the Nervous System:
Neurodevelopmental/ Early Life (eg. ADHD, autism) 1
Psychiatric (eg. Depression, Anxiety, Schizophrenia)
Lifespan Development:
Normal Brain Development: Fetus to Adolescence
Modeling and Analysis Methods:
Connectivity (eg. functional, effective, structural)
Novel Imaging Acquisition Methods:
Anatomical MRI 2
Keywords:
Cortex
Data analysis
Development
Morphometrics
MRI
Myelin
Psychiatric
Psychiatric Disorders
STRUCTURAL MRI
Other - Structural similarity networks
1|2Indicates the priority used for review
Provide references using author date format
2. Evans, Alan C. 2013. ‘Networks of Anatomical Covariance’. NeuroImage 80 (October): 489–504
3. Alexander-Bloch, Aaron, Jay N. Giedd, and Ed Bullmore. 2013. ‘Imaging Structural Co-Variance between Human Brain Regions’. Nature Reviews Neuroscience 14 (5): 322–36
4. Sebenius, Isaac, Jakob Seidlitz, Varun Warrier, Richard A. I. Bethlehem, Aaron Alexander-Bloch, Travis T. Mallard, Rafael Romero Garcia, Edward T. Bullmore, and Sarah E. Morgan. 2023. ‘Robust Estimation of Cortical Similarity Networks from Brain MRI’. Nature Neuroscience 26 (8): 1461–71
5. Okano, Hideyuki, Keigo Hikishima, Atsushi Iriki, and Erika Sasaki. 2012. ‘The Common Marmoset as a Novel Animal Model System for Biomedical and Neuroscience Research Applications’. Seminars in Fetal and Neonatal Medicine 17 (6): 336–40
6. Sawiak, Stephen J., Nigel I. Wood, Guy B. Williams, A. Jennifer Morton, and T. Adrian Carpenter. 2013. ‘Voxel-Based Morphometry with Templates and Validation in a Mouse Model of Huntington’s Disease’. Magnetic Resonance Imaging 31 (9): 1522–31
7. Weiskopf, Nikolaus, John Suckling, Guy Williams, Marta M. Correia, Becky Inkster, Roger Tait, Cinly Ooi, Edward T. Bullmore, and Antoine Lutti. 2013. ‘Quantitative Multi-Parameter Mapping of R1, PD*, MT, and R2* at 3T: A Multi-Center Validation’. Frontiers in Neuroscience 7
8. Paxinos, George and Watson, Charles and Petrides, Michael and Rosa, Marcello and Tokuno, Hironobu. 2012. The Marmoset Brain in Stereotaxic Coordinates. San Diego: Elsevier Academic Press.
9. Huntenburg, Julia M., Pierre-Louis Bazin, and Daniel S. Margulies. 2018. ‘Large-Scale Gradients in Human Cortical Organization’. Trends in Cognitive Sciences 22 (1): 21–31