jackman/wholeBrain
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

res and disc work

Browse Source
This commit is contained in:
ackman678
2014-08-14 09:52:36 -04:00
parent c4039831dd
commit 30ab8ce178
1 changed files with 15 additions and 12 deletions
Download Patch File Download Diff File Expand all files Collapse all files

27
wholeBrain_main.md
View File

@@ -2,7 +2,7 @@ Author: James B. Ackman
Date: 2014-06-05 00:38:46
Tags: paper, draft, manuscript, literature, research, #results, retinal waves, spontaneous activity, development, calcium domains
# Structured population activity across developing neocortex
# Structured dynamics of neural activity across developing neocortex
**Authors and Affiliated Institutions:**
James B. Ackman¹, Hongkui Zeng², and Michael C. Crair¹
@@ -65,18 +65,22 @@ We monitored motor movements simultaneously with cortical activity during our fM
The complex activity dynamics among nascent neocortical networks suggests that spatiotemporal correlations between areas may exist that provide information for development of intra- and inter-hemispheric connections. Indeed, recent work has suggested that neural activity is required for the migration of some interneuron subtypes (fishell work; ZJ. Huang work) and the development of callosal connections (kir2.1 eporation paper; olavarria work?). Since the timing of neural activity patterns is thought to be important for various aspects of circuit development it remains crucial to understand the correlational structure of ongoing cortical activity between brain regions. Thus to achieve a better understanding of the early activity patterns that may regulate interactions between cortical regions we first looked at correlation between the hemispheres. Cortical activity exhibited high temporal correlation between the hemispheres ([SupplementaryFig-symmetry.ai](../wholeBrain_blob/SupplementaryFig-symmetry.ai)). In addition this activity was highly correlated in the spatial dimension. We found that activity was correlated in anterior-posterior and medial-lateral directions. It exhibited mirror symmetric and non-mirror symmetric patterns. For example epochs of time would exhibit high correlation in the medial-lateral dimension or in the rostral-caudal dimension. This strength of correlation temporally and spatially increased between the hemsipheres with a function of age.
We computed a correlation matrix for each recording based on the pixel active fraction timecourses for each pair of cortical parcellations. Community structure among cortical nodes in the pairwise association matrix at P12-13 was detected with hierarchical clustering based on optimization of the graph modularity score, which measures the number of edges that fall within groups minus the number expected by chance [#Newman:2004]. The resulting community dendrogram was used to order the mean correlation matrices for each age group (Fig. 4a) (N=). This revealed a non-random network organization and a grossly similar correlational structure present across development (Fig. 4a). There were 4 primary subnetworks detected at P12-13-- motor-parietal (green), body-parietal (red), medial-cingulate-occipital (blue), and temporal (purple). Indeed the 4 network modules detected at 12-13 display as clusters along the diagonal in the correlation matrix and these are largely apparent in the the earlier age groups as well. However, functional correlation between cortical regions was highly dynamic across development, with both increases and decreases in correlation depending on brain area and age. Remarkably, detection of community structure in each age group independently showed similar vertex memberships in the 3 largest modules, with developmental switches in module membership occurring in PPC and M2, two brain regions known to integrate multimodal sensory and motor planning information (Fig. 4b).
We computed a correlation matrix for each recording based on the pixel active fraction timecourses for each pair of cortical parcellations. Community structure among cortical nodes in the pairwise association matrix at P12-13 was detected with hierarchical clustering based on optimization of the graph modularity score, which measures the number of edges that fall within groups minus the number expected by chance [#Newman:2004]. The resulting community dendrogram was used to order the mean correlation matrices for each age group (Fig. 4a) (N=). This revealed a non-random network organization and a grossly similar correlational structure present across development (Fig. 4a). There were 4 primary subnetworks detected at P12-13-- motor-S1-face (green), S1-body (red), medial--visual (blue), and auditory (purple). Indeed the 4 network modules detected at 12-13 display as clusters along the diagonal in the correlation matrix and these are apparent in the earlier age groups as well. However, functional correlation between cortical regions was highly dynamic across development, with both increases and decreases in correlation depending on brain area and age. Remarkably, detection of community structure in each age group independently showed similar vertex memberships in the 3 largest modules, with developmental switches in module membership occurring in PPC and M2, two brain regions known to integrate multimodal sensory and motor planning information (Fig. 4b).
The strongest correlations at each age typically occurred at symmetrical regions between the hemispheres in RSA, M1, M2, and T (Fig. 4a,b). In primary sensory areas such as A1, V1, and S1-barrel, inter-hemispheric correlation was initially weak and became strong at P12-13. At P8-9, correlations among cortical regions became more spread out, with strong correlations becoming stronger and weak correlations becoming weaker. There was increased positive correlation within modules and stronger negative correlation between modules at P12-13. For example, visual regions exhibited increased negative correlation with the body-parietal subnetwork, A1 showed increased negative correlations with multiple areas, and motor-parietal regions exhibited increased intra-module correlations.
![ **Figure 4.** Functional architecture of developing neocortex. **a** Group averaged correlation matrices of domain activity among cortical areas. Colormap indicates Pearson's r correlation coefficient values. Dendrogram and vertex order from community structure detected with hierarchical clustering in the P12-13 group. **b** Map of cortical area associations for r > 0.15. Vertex colors represent cortical communities detected with clustering within each age group. Edge width indicates the squared connection weight (r^2). Note both similarities in module membership and increased connection strength with age.](Figure4.png)
![ **Figure 4.** Functional architecture of developing neocortex. **a** Group averaged correlation matrices of domain activity among cortical areas. Colormap indicates Pearson's r correlation coefficient values. Dendrogram and node order from community structure detected with hierarchical clustering in the P12-13 group. **b** Map of cortical area associations for r > 0.15. Node colors represent cortical communities detected with clustering within each age group. Edge width indicates the squared connection weight (r^2). Note both similarities in module membership and increased connection strength with age.](Figure4.png)
We next analysed the topological properties of developing mouse cortical networks with graph theoretic measures as used in studies of large-scale functional connectivity derived from fMRI time-series [#Bullmore:2009].
We analysed the topological properties of developing mouse cortical networks with graph theoretic measures used in fMRI studies of large-scale functional connectivity [#Bullmore:2009]. Graphs of mean functional connectivity illustrated decreased randomness and tighter clustering within cortical modules during network development (Fig. 5a).
Indeed, the global clustering coefficient-- calculated as the mean of local clustering coefficients that reflect how highly interconnected each node's neighbors are-- significantly increased during development (Fig. 5c). In contrast, the average network diameter and shortest path changed marginally over the course of development. Both of these measures were significantly different than that of equivalent random networks at each age group. Networks having both high clustering and short average path lengths are key features that distinguish small world networks from random or lattice based networks-- , the clustering coefficient and average path length in the mean graphs from each age group. While the average path length decreased only slightly over the course of development, the clustering coefficient increased significantly with age.
The mean degree (number of edges per node) was higher at P12-13 and the degree distribution shifted during development to match the node order seen at P12-13 (Fig. 5b) (Supplementary Fig graph metrics). The mean node strength--the sum of all edge weights per node-- was higher at P12-13 and the rank order of node strengths also changed to that of the P12-13 distribution during devleopment. The 4 cortical areas having both the highest degree and node strength were M1, M2, PPC, and V2M (Fig. 5b). Measures of node centrality were computed to identify potential network hubs. Betweeness centrality measures the fraction of all shortest paths in a network that pass through a node, therefore identifing important throughputs in the network. M1, M2, PPC, and RSA had the highest betweeness centrality scores, which generally decreased during development. Eigenvector centrality is proportional to the sum of centralities for a node's connections, giving high scores to nodes that are linked to many other highly connected nodes. High scores (near the top of the linear fit) indicated hub nodes which included M1 and M2. Deviations from betweeness-eigenvector linearity can indicate whether a node is potentially more important as a network throughput (higher betweeness) or a driver node (higher eigenvector centrality). These results indicate that greater local connectivity together with higher connection strengths are key features in development of global network architecture of the cerebral cortex. Furthermore, the developing mouse cerebral cortex shares features of network topology that are similar to those seen in resting states networks in human infant.
![ **Figure 5.** Dynamics of functional connectivity in developing neocortex. **a** Graph of functional connections for r > 0.15. Vertex colors represent cortical communities detected with clustering within each age group. **b** Boxplots of degree (number of links) and vertex strength (sum of connection weights) by cortical area. The distributions become increasingly ordered like the P12-13 group with age. **c** Boxplots of clustering coefficient and average path length by recording. **d** Scatterplots of mean network centrality scores by cortical area.](Figure5.png)
![ **Figure 5.** Dynamics of functional connectivity in developing neocortex. **a** Graph of functional connections for r > 0.15. Node colors represent cortical communities detected with clustering within each age group. **b** Boxplots of degree (number of links) and node strength (sum of connection weights) by cortical area. The distributions become increasingly ordered like the P12-13 group with age. **c** Boxplots of clustering coefficient and average path length by recording. **d** Scatterplots of mean network centrality scores by cortical area.](Figure5.png)
@@ -84,15 +88,14 @@ We next analysed the topological properties of developing mouse cortical network
# Conclusions
* Neural population activity constitutes discrete spatial and temporal activations among developing cortical areas
* Cortical activity exhibits symmetrical spatial and temporal activations across the hemispheres
* Cortical activity is coordinated with motor behavioral state in an areal dependent fashion
* Developing cerebral networks comprise distinct functional modules among cortical areas
We have described for the first time how neural population activity constitutes discrete spatial and temporal activations among developing cortical areas in vivo. In contrast to classical in vitro imaging studies (Yuste, Katz)
Inter hemispheric functional connectivity, importance for autism, schizophrenia. Maybe an activity-dependent mechanism for commisural connectivity. olavarria work, evidence for inter hemispheric activity dependence
Cortical activity was coordinated with motor behavioral state in an areal dependent fashion.
* **Ongoing activity in developing cortex is not random** -- it is coordinated in space and time within and between the hemispheres among cortical areas
* Functional mesoscale imaging technique as a template for assessing altered functional dynamics in models for neurological disorders
Cortical activity exhibited symmetrical spatial and temporal activations across the hemispheres. This is consistent with the stronger interhemsipheric resting state connectivity between homotopic cortical regions in the human infant. Increased intra-hemispheric connectivity and modularity is thought to be key features of functional connectivity development in large scale brain networks.
Developing cerebral networks comprised distinct functional modules among cortical areas. This indicated that functional network identities may be shaped early in development. The inter-module dynamics we found may have important relevance for cortical plasticity seen after sensory deprivation or traumatic injury early in development-- such as that seen in enucleated monkeys (H. kennedy experimetns), children born blind or deaf, or epilepsy patients undergoing hemispherectomies. Given that altered inter- and intra- hemispheric functional connectivity is thought to be relevant in autism and schizophrenia extension of the work performed in this study to mouse models for these neurological diseases will be interesting.
This study demonstrates that ongoing activity in developing cortex is not random -- it is coordinated in space and time between hemispheric networks in the neocortex. Furthermore, our functional mesoscale imaging technique should be a useful tool for assessing potentially altered functional connectivity dynamics in animal models for neurological disorders.
# Methods Summary