We studied the global relationship between gene expression and neuroanatomical connectivity in the adult rodent brain. We utilized a large data set of the rat brain “connectome” from the Brain Architecture Management System (942 brain regions and over 5000 connections) and used statistical approaches to relate the data to the gene expression signatures of 17,530 genes in 142 anatomical regions from the Allen Brain Atlas. Our analysis shows that adult gene expression signatures have a statistically significant relationship to connectivity. In particular, brain regions that have similar expression profiles tend to have similar connectivity profiles, and this effect is not entirely attributable to spatial correlations. In addition, brain regions which are connected have more similar expression patterns. Using a simple optimization approach, we identified a set of genes most correlated with neuroanatomical connectivity, and find that this set of genes is enriched for genes involved in neuronal development and axon guidance. A number of the genes have been implicated in neurodevelopmental disorders such as autistic spectrum disorder. Our results have the potential to shed light on the role of gene expression patterns in influencing neuronal activity and connectivity, with potential applications to our understanding of brain disorders. Supplementary data are available at http://www.chibi.ubc.ca/ABAMS.
Full Publication at PLoS Computational Biology
Update – source code available at github.
Update – connection and nomenclature matrices added below.
Update – we have published a related analysis in Frontiers in Neuroinformatics.
Presented at Neuroscience 2010 in San Diego, Tuesday, Nov 16, 1-2pm, 636.13, B3
Supplement Figure 1:
Density Plot of Expression Correlation Between Region Pairs.
Mantel correlation between different matrices after controlling for proximity. The 141 regions with incoming connectivity information were used to generate the correlations.
Intersection of the top 10th percentile of brain region pairings for connectivity and gene expression correlations. Using outgoing proximity controlled connectivity. With expression correlation derived from only the top outgoing genes. Colors represent five major brain divisions: cerebellum (yellow), cerebrum (green), hindbrain (blue), interbrain (purple) and midbrain (red).
Heatmap produced by average linkage hierarchical clustering of the top outgoing gene list. Rows are normalized to show expression ranging from low (blue) to high (yellow) with grey representing missing values.
Gene(row) dendrogram from the hierarchical clustering used in Figure S4. Some of the genes mentioned in the text are highlighted in red, showing dispersed clustering.
Density Plot of Gene to Gene Correlations. Gene to gene correlations were computed within the “outgoing” gene list and all genes.
Pgrmc1 expression levels versus connectivity. For each region pair this plot shows the sum of the two regions expression in the context of their connectivity.
This file is formated for the Venn Master software tool.
The Dong.CA1 sets are from “Genomic-anatomic evidence for distinct functional domains in hippocampal field CA1” by Dong et al.
The Top1000 Validation sets are from “The Allen Brain Atlas: 5 years and beyond” by Jones et al.
The AutismDB set is from AutDB.
These matrix files follow the format of read.table in R (tab separated).
This matrix takes the form of a standard adjacency matrix – if a row and column pair is a 1 in this matrix then there is a reported BAMS connection from the row-region (source) to the column-region (target).
Outgoing connection matrix, created from the perspective of a column, all values of one in a column mean there is a connection from the column region to the row region, 942 by 112 regions
Both of the above are rectangular matrices and not the usual design of an adjacency matrix, columns have been filtered to remove regions with no connections (but not rows). The column names are the mapped Allen brain reference atlas region labels.
Full BAMS connection matrix from Swanson 1998 data, up propigated
The above matrices are directed (962 by 962 regions) and derived from the RDF converted Swanson-98.xml file by John Barkley. Updated connectivity data can be found at the Brain Architecture Management system (BAMS). If a row and column pair is a 1 in these full matrices then there is a reported BAMS connection from the row-region (source) to the column-region (target).
If a cell in the matrix has a value of one, then the row region encloses the column region. If the row and column regions are the same, the value is 0.