Multi-modal Analysis Show Distinct Covariance Network Patterns Between FTLD Proteinopathies
Poster No:
1522
Submission Type:
Abstract Submission
Authors:
Hyung Seok Roh1, Daniel Ohm1, Jeffrey Phillips1, Noah Capp1, Alejandra Bahena1, Philip Sabatini1, Lauren Massimo1, David Wolk1, Edward Lee1, Murray Grossman1, Corey McMillan1, James Gee1, David Irwin1, Min Chen1
Institutions:
1University of Pennsylvania, Philadelphia, PA
First Author:
Co-Author(s):
Introduction:
Frontotemporal dementia (FTD) is a common form of early-onset dementia with diverse clinical syndromes [1] associated with Frontotemporal Lobar Degeneration (FTLD) mainly caused by tau (FTLD-Tau) or TDP-43 (FTLD-TDP) proteinopathies [2]. Despite evidence relating specific proteinopathies to FTD clinical syndromes, there is still considerable heterogeneity of clinical syndromes and proteinopathies [3]. We perform a multi-modal network-based analysis incorporating both antemortem MR structural imaging and histopathological sampling of FTLD. We construct structural covariance networks from MR volume measurements at regions with histopathological sampling to reveal distinct network patterns between FTLD-Tau and FTLD-TDP. These networks are then compared to postmortem covariance networks from histopathology measurements of disease burden.
Methods:
Digital Histopathology
Our cohort included 76 FTLD-Tau and 103 FTLD-TDP subjects. Each subject was sampled bilaterally at up to 20 regions in each hemisphere. Staining and percentage of area occupied (%AO) calculation of each pathology inclusions follow methods described in [4].
MR Data
56 autopsy-confirmed patients (26 FTLD-Tau, 30 FTLD-TDP), and 54 healthy controls (HCs) from Penn Integrated Neurodegenerative Disease Database were used. Volume was measured at cortical regions of interest from T1-weighted MR scans follow methods described in [5]. Volume measurements matching pathology regions were converted to W-scores [6], corrected for both age and sex, to account for healthy variation.
Covariance Networks
Postmortem log %AO and antemortem T1 MR volume W-score covariance networks for the FTLD-Tau and FTLD-TDP groups were constructed with nodes representing sampled regions and edges weighted by group-level Pearson's correlation coefficient. Group-level differences were evaluated using z-tests after Fisher's Z-transformation [7], focusing on the 5% most significant edges in the networks.
Node Strength
Node strength was calculated as sum of all connected edge weights at each node. To test if regions with greater pathology act as epicenters, correlation between node strength and log %AO of postmortem group difference networks was measured. To show that increased covariance reflects greater atrophy, we measured correlation between node strength and volume W-score of antemortem group difference networks. Lastly, to test that increased covariance reflects greater pathology, correlation between node strength of antemortem group difference networks and log %AO was measured.
Goodness of Fit
Graph distance metrics (Hamming Distance, Frobenius Distance, Jaccard Distance, and Graph Diffusion [8]) were used to measure the goodness of fit between postmortem and antemortem group-level difference networks.
Our cohort included 76 FTLD-Tau and 103 FTLD-TDP subjects. Each subject was sampled bilaterally at up to 20 regions in each hemisphere. Staining and percentage of area occupied (%AO) calculation of each pathology inclusions follow methods described in [4].
MR Data
56 autopsy-confirmed patients (26 FTLD-Tau, 30 FTLD-TDP), and 54 healthy controls (HCs) from Penn Integrated Neurodegenerative Disease Database were used. Volume was measured at cortical regions of interest from T1-weighted MR scans follow methods described in [5]. Volume measurements matching pathology regions were converted to W-scores [6], corrected for both age and sex, to account for healthy variation.
Covariance Networks
Postmortem log %AO and antemortem T1 MR volume W-score covariance networks for the FTLD-Tau and FTLD-TDP groups were constructed with nodes representing sampled regions and edges weighted by group-level Pearson's correlation coefficient. Group-level differences were evaluated using z-tests after Fisher's Z-transformation [7], focusing on the 5% most significant edges in the networks.
Node Strength
Node strength was calculated as sum of all connected edge weights at each node. To test if regions with greater pathology act as epicenters, correlation between node strength and log %AO of postmortem group difference networks was measured. To show that increased covariance reflects greater atrophy, we measured correlation between node strength and volume W-score of antemortem group difference networks. Lastly, to test that increased covariance reflects greater pathology, correlation between node strength of antemortem group difference networks and log %AO was measured.
Goodness of Fit
Graph distance metrics (Hamming Distance, Frobenius Distance, Jaccard Distance, and Graph Diffusion [8]) were used to measure the goodness of fit between postmortem and antemortem group-level difference networks.
Results:
Node Strength
Positive correlations between postmortem node strength and log %AO for FTLD-Tau and FTLD-TDP groups were observed. Statistically significant negative correlations were found between antemortem node strength and volume W-score for both groups. Positive correlations between antemortem node strength and log %AO were observed for both groups, with statistical significance in the FTLD-Tau group (Figure 1) establishing that increased covariance among nodes of both modalities was associated with greater neurodegeneration.
Goodness of Fit
Similar network characteristics within each FTLD-Tau and FTLD-TDP group were observed between postmortem and antemortem networks, with distinct patterns between the two groups. Quantitively, we consistently observe a smaller graph distance between postmortem and antemortem networks for both FTLD-Tau and FTLD-TDP groups than between pathology groups across all metrics (Figure 2).
Positive correlations between postmortem node strength and log %AO for FTLD-Tau and FTLD-TDP groups were observed. Statistically significant negative correlations were found between antemortem node strength and volume W-score for both groups. Positive correlations between antemortem node strength and log %AO were observed for both groups, with statistical significance in the FTLD-Tau group (Figure 1) establishing that increased covariance among nodes of both modalities was associated with greater neurodegeneration.
Goodness of Fit
Similar network characteristics within each FTLD-Tau and FTLD-TDP group were observed between postmortem and antemortem networks, with distinct patterns between the two groups. Quantitively, we consistently observe a smaller graph distance between postmortem and antemortem networks for both FTLD-Tau and FTLD-TDP groups than between pathology groups across all metrics (Figure 2).
Conclusions:
Our results suggest there exist distinct network patterns of disease between FTLD-Tau and FTLD-TDP that are consistent across different modalities. In future work, we plan to assess the goodness of fit through more intricate methodologies, such as motif analysis and stochastic block model.
Disorders of the Nervous System:
Neurodegenerative/ Late Life (eg. Parkinson’s, Alzheimer’s) 2
Modeling and Analysis Methods:
Connectivity (eg. functional, effective, structural) 1
Multivariate Approaches
Keywords:
Other - Frontotemporal Lobar Degeneration, Histopathology, MRI, Network Science, tauopathy, TDP-43
1|2Indicates the priority used for review
Provide references using author date format
[2] Mackenzie, I.R., Neumann, M., Bigio, E.H., Cairns, N.J., Alafuzoff, I., Kril, J., Kovacs, G.G., Ghetti, B., Halliday, G., Holm, I.E., Ince, P.G., Kamphorst, W., Revesz, T., Rozemuller, A.J., Kumar-Singh, S., Akiyama, H., Baborie, A., Spina, S., Dickson, D.W., Trojanowski, J.G., Mann, D.M. (2010). "Nomenclature and nosology for neuropathologic subtypes of frontotemporal lobar degeneration: an update." Acta Neuropathol, 119, 1–4.
[3] Murley, A.G., Coyle-Gilchrist, I., Rouse, M.A., Jones, P.S., Li, W., Wiggins, J., Lansdall, C., Rodríguez, P.V., Wilcox, A., Tsvetanov, K.A., Patterson, K., Lambon Ralph, M.A., Rowe, J.B. (2020). "Redefining the multidimensional clinical phenotypes of frontotemporal lobar degeneration syndromes." Brain, 143, 1555–1571.
[4] Chen, M., Ohm, D.T., Phillips, J.S., McMillan, C.T., Capp, N., Peterson, C., Xie, E., Wolk, D.A., Trojanowski, J.Q., Lee, E.B., Gee, J., Grossman, M., Irwin, D.J. (2022). "Divergent Histopathological Networks of Frontotemporal Degeneration Proteinopathy Subtypes." Journal of Neuroscience, 42(18), 3868–3877.
[5] Chen, M., Burke, S., Olm, C.A., Irwin, D.J., Massimo, L., Lee, E.B., Trojanowski, J.Q., Gee, J.C., Grossman, M. (2023). "Antemortem network analysis of spreading pathology in autopsy-confirmed frontotemporal degeneration." Brain Communications, 5(3), fcad147.
[6] La Joie, R., Perrotin, A., Barré, L., Hommet, C., Mézenge, F., Ibazizene, M., Camus, V., Abbas, A., Landeau, B., Guilloteau, D., de La Sayette, V., Eustache, F., Desgranges, B., Chételat, G. (2012). "Region-specific hierarchy between atrophy, hypometabolism, and β-amyloid (AΒ) load in Alzheimer’s disease dementia." The Journal of Neuroscience, 32(46), 16265–16273.
[7] Fisher, R.A. (1915). "Frequency distribution of the values of the correlation coefficient in samples from an indefinitely large population." Biometrika, 10, 507–521.
[8] McCabe, S., Torres, L., LaRock, T., Haque, S., Yang, C.-H., Hartle, H., Klein, B. (2021). "Netrd: A library for network reconstruction and graph distances." Journal of Open Source Software, 6(62), 2990.