Macaque cell type and gene expression correlates of neuroanatomy

887 

Abstract Submission 

Burke Rosen¹, Takuya Hayashi², David van Essen¹, Matthew Glasser¹

¹Washington University in St. Louis, St. Louis, MO, ²RIKEN Center for Biocystems Dynamics Research, Kobe, Hyogo

Burke Rosen  
Washington University in St. Louis
St. Louis, MO

Takuya Hayashi  
RIKEN Center for Biocystems Dynamics Research
Kobe, Hyogo

David van Essen  
Washington University in St. Louis
St. Louis, MO

Matthew Glasser, Dr.  
Washington University in St. Louis
St. Louis, MO

The morphology and physiology of every cell is largely determined by the relative transcription levels of its genes. Earlier cortical transcriptomic studies using bulk tissue assays found that the first principal component (PC1) of brain-enriched gene expression predicts the cortical T1w/T2w ratio, an indicator of myelin content [1] and hierarchical relationships [2], and a subset of these genes moderately predict human vs chimpanzee evolutionary expansion of cortex [3]. However, these studies did not identify cell types. Cell type composition is potentially more informative than aggregate expression because species differences in expression are sometimes constrained to specific cell types [4], the patterns of expression can be spatially opposed in different types [5], and because cells' gestalt transcription is statistically more robust than individual gene expression. A recent spatial transcriptomic survey of the entire macaque cortex [6] examined anterior-posterior (A-P) gradients of cell type composition but these data were not compared to cortical thickness, myelin, or evolutionary expansion. We performed these comparisons by mapping areal transcription [6] to the cortical surface.

Individual macaque cortical surfaces were reconstructed from T1 and T2 weighted MR Images [7]. Group average (n=32) surfaces and maps of cortical myelin and thickness were obtained. As described in a companion poster, human vs macaque evolutionary expansion was estimated. Cells in [6] are localized to the volumetric D99_v2 atlas [8]. We projected the atlas to the surface with Connectome Workbench [9]. For each parcel, the expression profile of all cells was averaged to yield a pseudo-bulk parcel x gene matrix (131x15929). Counts of cells were tallied to create a parcel x cell type matrix (131x258). First, in an exploratory analysis following [2], expression and composition PCs explaining the most variance across parcels were compared to brain measures. To determine which genes and cell types are most predictive of each brain measure, we fit elastic net regularized generalized linear models (GLMs) [10] by cross-validation.

PC1 of areal cell type composition explains considerable variance in myelin (69%), thickness (27%), and A-P position (26%) across parcels, whereas PC1 of pseudo-bulk gene expression explains only 0%, 1%, and 3%, respectively (Fig. 1). PC2 of expression was more predictive than cell type of the examined brain measures, though less predictive than cell type PC1. Expansion was not well captured by either factorization. Elastic net GLMs identified a subset of cell type or gene predictors that explain a large degree of variance in brain measures (Fig. 2). Models of cell type composition explain 95%, 84%, 96%, and 74% of variance in myelin, thickness, A-P position, and expansion vs 94%, 89%, 98%, and 40%, for expression. L2 and L4 glutamatergic neurons are prominent among predictors of cortical thickness, and evolutionary expansion, respectively. RELN-expressing inhibitory neurons are prominent predictors of cortical thickness.

The predictive power of cell type composition PC1 with respect to myelination is greater than previously reported for human bulk gene expression [2] and any pseudo-bulk gene expression PC, consistent with the hypothesis that cell type is more informative that individual gene expression. The elastic net GLM analysis is more ambiguous; the cell type GLM captures more variation in expansion but selects more predictors. Intriguingly, L4 pyramidal neurons, whose abundance was previously hypothesized to be associated with primate evolutionarily divergence [6], feature prominently and type L4.9 is a primate-specific neuron [6]. While macaque transcription was less able to explain evolutionary expansion than other brain measures, this is not unexpected as much of the variability in expansion is presumably results from human transcription [4]. Future interspecies comparisons are needed.

Transcriptomics ¹

Image Registration and Computational Anatomy

Cortical Anatomy and Brain Mapping ²

Cortical Cyto- and Myeloarchitecture

Cortex

Other - macaque;surface; myelin;thickness;evolution;cell type;spatial transcriptomics;single cell