6 Spatial Deconvolution

Spatial Deconvolution is an analysis technique that estimates the relative abundance of cell types for each ROI/AOI segment. The algorithm utilizes a pre-specified cell profile matrix generated from scRNA-seq data to deconvolute GeoMx® data. For additional details, please refer to the following publication and open-source algorithm.

For this analysis, we used Stewart et al., (2019) which contains 33 cell types.

We can analyze deconvolution results for a given cell type within an ROI/AOI segment either as an abundance beta estimate or as a proportion. The abundance beta can be thought of as a relative abundance score; the higher the beta for a given cell type, the more abundant that cell type is in that ROI/AOI segment relative to other segments. In contrast, the higher the proportion for a given cell type, the more abundant that cell type is relative to other cell types within that ROI/AOI segment.

6.1 Heatmap of Cell Abundance

We plot the abundance beta estimates using unsupervised hierarchical clustering (using Euclidean distances), displayed as a heatmap. Each column is an ROI/AOI segment and its relative abundances for the different cell types.

Analyst Notes: Sample clustering of beta estimates was performed using k-means partitioning (k =4). Clusters 1 and 2 consist of glomerulus regions and display an abundance of podocytes and glomerular endothelial cells. Qualitatively, cluster 1, which mostly contains DKD samples, shows increased fibroblast, pelvic epithelial, plasmacytoid dedritic, and NK cell abundances. Cluster 2 is largely comprised of normal samples. Cluster 4 includes PanCK+ tubules from both normal and DKD samples and shows relatively high connecting tubule, prinicipal cell, and Type B intercalated cell abundances.

6.2 Differential Abundance of Cell Types

In this section, we use the same overall approach as in differential gene expression analysis to quantify different cell type abundances for our experimental comparisons.

The figures in the tabs below can be used to compare differences between groups for each contrast. The Volcano tab shows the fold change between cell deconvolution abundance estimates (i.e., “beta” estimates) and the significance. The Abundance (Beta) tab shows beta estimates for each segment as a stacked bar plot with relevant annotations plotted to the side. The Proportion tab shows an analogous plot, but shows proportional cell deconvolution data. Finally, the cell types with the greatest difference between groups are displayed as a series of plots in the Violins tab.

Note that a high fold change in one group may be due to an increase in the number of cells in that group’s samples, or it may indicate a proportional decrease in the abundance of cells in the other group.

6.3 Glomeruli vs Tubules

We will compare glomeruli and tubules across the kidney dataset.

The following formula was used to model differences for the abundance (beta) of a given cell type:

\[ cell \sim region + (1+region|slide) \]

We adjust for the multiple sampling of ROI/AOI segments per tissue with the \(slide\) variable.

Analyst Notes: The volcano plot quantifies some of the qualitative observations made in Heatmap of Cell Abundance. That is, there is a higher abundance of podocytes and glomerular endothelial cells in glomeruli. Connecting tubules, principal cells, and proximal tubules are overall higher in tubule segments than in glomeruli segments. The three glomeruli segments with relatively low total beta values and low podocyte counts do not appear to be a result of low nuclei counts.

The stacked bar plot below shows the proportion estimates for cell types with the greatest fold change.

The stacked bar plot below shows the beta estimates for cell types with the greatest fold change.

The following table contains the fold change estimates and p-value/adjusted p-value (FDR, False Discovery Rate) for each cell type in this contrast.

6.4 Healthy vs Diseased (DKD) Glomeruli

We will now compare glomeruli from healthy and diseased (DKD) kidneys.

The following formula was used to model differences for the abundance of a given cell type:

\[ cell \sim class + (1|slide) \]

We adjust for the multiple sampling of ROI/AOI segments per tissue with the \(slide\) variable.

Analyst Notes: Podocytes and glomerular endothelium, both of which were overall higher in glomeruli than tubules, show evidence of decreased abundance in diabetic kidney diseased samples compared to normal tissues.

The stacked bar plot below shows the proportion estimates for the two significantly different cell types split by region.

The stacked bar plot below shows the beta estimates for the two significantly different cell types split by region.

The following table contains the fold change estimates and p-value/adjusted p-value (FDR, False Discovery Rate) for each cell type in this contrast.