Classifying Cells

We will now examine the results of our exploratory analysis and use a combination of techniques to classify different subsets of T and B cells in the MALT sample.

Exploratory Analysis Results

Double click the merged UMAP data node
Under Configure on the left, click Style, select the Graph-based cluster node, and color by the Graph-based attribute (Figure 1)

The 3D UMAP plot opens in a new data viewer session (Figure 2). Each point is a different cell and they are clustered based on how similar their expression profiles are across proteins and genes. Because a graph-based clustering task was performed upstream, a biomarker table is also displayed under the plot. This table lists the proteins and genes that are most highly expressed in each graph-based cluster. The graph-based clustering found 11 clusters, so there are 11 columns in the biomarker table.

Click and drag the 2D scatter plot icon from New plot onto the canvas (Figure 2)
Drop the 2D scatter plot to the right of the UMAP plot

Click Merged counts to use as data for the 2D scatter plot (Figure 3)

A 2D scatter plot has been added to the right of the UMAP plot. The points in the 2D scatter plot are the same cells as in the UMAP, but they are positioned along the x- and y-axes according to their expression level for two protein markers: CD3_TotalSeqB and CD4_TotalSeqB, respectively (Figure 4).

In Select & Filter, click Criteria to change the selection mode
Click the blue circle next to the Add rule drop-down menu (Figure 5)

Click Merged counts to change the data source
Choose CD3_TotalSeqB from the drop-down list (Figure 6)

Click and drag the slider on the CD3D_TotalSeqB selection rule to include the CD3 positive cells (Figure 7)

As you move the slider up and down, the corresponding points on both plots will dynamically update. The cells with a high expression for the CD3 protein marker (a marker for T cells) are highlighted and the deselected points are dimmed (Figure 8).

Click Merged counts in Get data on the left under Setup
Click and drag CD8a_TotalSeqB onto the 2D scatter plot (Figure 9)
Drop CD8_TotalSeqB onto the x-axis configuration option

The CD3 positive cells are still selected, but now you can see how they separate into CD4 and CD8 positive populations (Figure 10).

The simplest way to classifying cell types is to look for the expression of key marker genes or proteins. This approach is more effective with CITE-Seq data than with gene expression data alone as the protein expression data has a better dynamic range and is less sparse. Additionally, many cell types have expected cell surface marker profiles established using other technologies such as flow cytometry or CyTOF. Let's compare the resolution power of the CD4 and CD8A gene expression markers compared to their protein counterparts.

Click the duplicate plot icon above the 2D scatter plot (Figure 11)

Click Merged counts in the Get Data icon under Setup
Search for the CD4 gene
Click and drag CD4 onto the duplicated 2D scatter plot
Drop the CD4 gene onto the y-axis option
Search for the CD8A gene
Click and drag CD8A onto the duplicated 2D scatter plot
Drop the CD8A gene onto the x-axis option

The second 2D scatter plot has the CD8A and CD4 mRNA markers on the x- and y-axis, respectively (Figure 12). The protein expression data has a better dynamic range than the gene expression data, making it easier to identify sub-populations.

Manually select the cells with high expression of the CD4_TotalSeqB protein marker (Figure 13)

More than 2000 cells show positive expression for the CD4 cell surface protein.

Let's perform the same test on the gene expression data.

Click on a blank spot on the plot to clear the selection
Manually select the cells with high expression of the CD4 gene marker (Figure 14)

This time, only 500 cells show positive expression for the CD4 marker gene. This means that the protein data is less sparse (i.e. there fewer zero counts), which further helps to reliably detect sub-populations.

T cells

Based on the exploratory analysis above, most of the CD3 positive cells are in the group of cells in the right side of the UMAP plot. This is likely to be a group of T cells. We will now examine this group in more detail to identify T cell sub-populations.

Draw a lasso around the group of putative T cells (Figure 15)

Click and drag the plot to rotate it around

This group of putative T cells predominantly consists of cells assigned to graph-based clusters 3, 4, and 6, indicated by the colors. Examining the biomarker table for these clusters can help us infer different types of T cell.

Click and drag the bar between the UMAP plot and the biomarker table to resize the biomarker table to see more of it (Figure 17)

Cluster 6 has several interesting biomarkers. The top biomarker is CXCL13, a gene expressed by follicular B helper T cells (Tfh cells). Another biomarker is the PD-1 protein, which is expressed in Tfh cells. This protein promotes self-tolerance and is a target for immunotherapy drugs. The TIGIT protein is also expressed in cluster 6 and is another immunotherapy drug target that promotes self-tolerance.

Cluster 4 expresses several marker genes associated with cytotoxicity (e.g. NKG7 and GZMA) and both CD3 and CD8 proteins. Thus, these are likely to be cytotoxic cells.

We can visually confirm these expression patterns and assess the specificity of these markers by coloring the cells on the UMAP plot based on their expression of these markers.

Click the duplicate plot icon above the UMAP plot

We will color the cells on the duplicate by their expression of marker genes, while keeping the original plot colored by graph-based cluster assignment.

Click and drag the CXCL13 gene from the biomarker table onto the duplicate UMAP plot
Drop the CXCL13 gene onto the Green (feature) option (Figure 18)

Click and drag the NKG7 gene from the biomarker table onto the duplicate UMAP plot
Drop the NKG7 gene onto the Red (feature) option

The cells with higher CXCL13 and NKG7 expression are now colored green and red, respectively. By looking at the two UMAP plots side by side, you can see these two marker genes are localized in graph-based clusters 6 and 4, respectively (Figure 19).

Click the blue circle next to the Add criteria drop-down list
Search for Graph to search for a data source
Select Graph-based clustering (derived from the Merged counts > PCA data nodes)
Click the Add criteria drop-down list and choose Graph-based to add a selection rule (Figure 20)

In the Graph-based filtering rule, click All to deselect all cells
Click cluster 6 to select all cells in cluster 6
Using the Classify tool, click Classify selection
Label the cells as Tfh cells (Figure 21)
Click Save

Click cluster 4 to select all cells in cluster 4
In the Classify icon, click Classify selection
Label the cells as Cytotoxic cells
Click Save

We can classify the remaining cells as helper T cells, as they predominantly express the CD4 protein marker.

Click on the invert selection icon in either of the UMAP plots (Figure 22)

In Classify, click Classify selection
Label the cells as Helper T cells
Click Save

Let's look at our progress so far, before we classify subsets of B-cells.

Click the Clear filters link in Select & Filter
Select the duplicate UMAP plot (with the cell colored by marker genes)
Under Configure on the left, open Style and color the cells by New classifications (Figure 23)

B cells

In addition to T-cells, we would expect to see B lymphocytes, at least some of which are malignant, in a MALT tumor sample. We can color the plot by expression of a B cell marker to locate these cells on the UMAP plot.

In the Get data icon on the left, click Merged counts
Scroll down or use the search bar to find the CD19_TotalSeqB protein marker
Click and drag the CD19_TotalSeqB marker over to the UMAP plot on the right
Drop the CD19_TotalSeqB marker over the Color configuration option on the plot

The cells in the UMAP plot are now colored from grey to blue according to their expression level for the CD19 protein marker (Figure 24). The CD19 positive cells correspond to several graph-based clusters. We can filter to these cells to examine them more closely,

Lasso around the CD19 positive cells (Figure 25)

The plots will rescale to include the selected points. The CD19 positive cells include cells from graph-based clusters 1, 2 and 7 (Figure 26).

Find the CD3_TotalSeqB protein marker in the biomarker table
Click and drag the CD3_TotalSeqB onto the UMAP plot on the right
Drop the CD3_TotalSeqB protein marker onto the Color configuration option on the plot (Figure 27)

While these cells express T cell markers, they also group closely with other putative B cells and express B cell markers (CD19). Therefore, these cells are likely to be doublets.

Select either of the UMAP plots
Click on the Select & Filter
Find the CD3_TotalSeqB protein marker in the biomarker table
Click and drag CD3_TotalSeqB onto the Add criteria drop-down list in Select & Filter (Figure 28)
Set the minimum threshold to 3 in the CD3_TotalSeqB selection (Figure 29)
Click the Classify icon then click Classify selection
Label the cells as Doublets
Click Save

The biomarkers for clusters 1 and 2 also show an interesting pattern. Cluster 1 lists IGHD as its top biomarker, while cluster 2 lists IGHA1 as the fourth most significant. Both IGHD (Immunoglobulin Heavy Constant Delta) and IGHA1 (Immunoglobulin Heavy Constant Alpha 1) encode classes of the immunoglobulin heavy chain constant region. IGHD is part of IgD, which is expressed by mature B cells, and IGHA1 is part of IgA1, which is expressed by activated B cells. We can color the plot by both of these genes to visualize their expression.

Click, drag and drop IGHD from the biomarker table onto the Green (feature) configuration option on the UMAP plot on the right
Click, drag and drop IGHA1 from the biomarker table onto the Red (feature) configuration option on the UMAP plot on the right (Figure 30)

We can use the lasso tool to select and classify these populations.

Lasso around the IGHD positive cells (Figure 31)
In the Classify icon on the left, click Classify selection
Label the cells as Mature B cells
Click Save

Lasso around the IGHA1 positive cells (Figure 32)
In the Classify icon on the left, click Classify selection
Label the cells as Activated B cells
Click Save

We can now visualize our classifications.

Click the Clear filters link in the Select & Filter icon on the left
Select the duplicate UMAP plot (with the cell colored by marker genes)
Under Configure on the left, click the Style icon and color the cells by New classifications (Figure 33)