Cell clustering and Differential Expression (DE)¶

Motivation:

Spermatogenesis goes through different stages, starting from SpermatogoniaA cells, going into clonal expansion while keeping cells connected through cytoplasmic bridges (SpermatogoniaB), and then continuing with the meiotic process (Spermatocites I and II). Finally, cells become Round spermatids, which then elongate to become Elongated spermatids and sperm.

Detecting those cell types is essential to answer biological questions such as

• which genes are most expressed for each cell type (beyond well known ones)?
• in which proportion is every cell type present?
• are there unknown cell types that I can identify?

Learning objectives:

• Identify potential cell clusters by visualizing marker genes on the UMAP plot
• Understanding and applying differential gene expression analysis to verify cluster identities
• Performing an analysis of subclusters in the dataset

Execution time: 45-60 minutes

*Import packages*

Read the data integrated with glmpca and bbknn

Identification through marker genes¶

we try to identify clusters of cells by looking at the expression of relevant marker genes. This requires a previous biological knowledge of those cell types, such that we can input the markers. Below, we define a dectionary, where for each cell type we define a list of markers. Then we will plot every list of markers on the UMAP plot

We can see how many markers easily identify groups of cells by plotting the expression

Sertoli are often not possible to identify. They are big in size, meaning they are often not isolated successfully. Many of their markers are in common with other somatic cells. Also, their function as nurse cells for germ cells of the testis means that their marker genes are also expressed. We can see that CTSL is expressed in some germ cells, but not in other clusters, while VIM is expressed in a likely somatic cluster (but it is common to other somatic cell types)

Macrophage cells seem to be absent

There is a little endothelial cluster

and also a myoid cluster

Leydig cells are likely to be missing as well

Now we create some clusters, and try to get the same division we saw by plotting markers. We can tune the number of clusters by changing the resolution parameter. We will be able to give the same name to more clusters, so it is fine to create a fine-grained clustering.

Write the names in the dictionary new_names. You should be able to give a name for each cell type. Below is an example, but the names are not in the right position. If there is more than one cluster with same cell type, just write the name followed by a dot . and a number. For example, by writing for example .1 and .2 at the end of the names. We will remove the numbers afterwords.

we apply the new names

we remove the numbers from same cell types

save the clusters in the sample and plot the new ones

We can look at markers in a heatmap or a dotplot

Differential Expression (DE) analysis¶

We can do differential expression (DE) analysis to double check which genes are differentially expressed in each cluster. A gene is differentially expressed in a cluster when its expression in the cells of that cluster is statistically bigger than in all other cells. This is verified through a statistical test.

Together with the gene names we also get p-values from the test, and a factor (log-fold change) telling the magnitude of how much the expression is larger than in other cells.

Apply the differential expression tool on the clusters for the top ten genes of each cluster. Save the results in .uns[DE_clusters]

Access the list of names

Access the table including p-values (with suffix _P in each column) and log-fold change (with suffix _L in each column)

We can easily save the table in csv format. This can be opened in Excel.

Subclustering the data¶

We find subclusters of cells using markers for cell types that are found between late spermatogonia and spermatocites.

Let's look at the markers plottes only over the cells of the clusters SpermatogoniaB, SpermatocitesI and SpermatocitesII

We want to create new clusters in our dataset by starting from the old clustering. To do this we use the option restrict_to, where we write the name of the old clustering, and the name of which clusters we want to subset. Try to tune the resolution to have a proper number of clusters to rename.

Let's look at the new clustering of spermatogoniaB and spermatocites I/II

Names are very long. We keep only the numbers at the end

Before renaming, we also want to look at the differentially expressed genes. We should be able to find at least some of the markers used in the plots. However, it can be that those do not appear because there are many other coexpressed genes with high expression values.

We can find some of the marker genes. For example SCML1 for leptotene cells, SYCP1 and LY6K for zygotene, CCDC112 and PIWIL1 for pachitene, CCNA1 and AURKA for zygotene.

We can again look at p-values and log-fold changes

and save the table

We can look at only the columns of a cluster from the large table so it is more readable

We rename the new clusters. Write the names in the dictionary. Some of the clusters might still be SpermatogoniaB or SpermatocitesII as before.

Just a plot of the two clustering side by side

It isn't really clear how to rename clusters in somatic cells. We let them be called somatic without further specification

You can plot the correlation matrix of the clusters with a dendrogram tree on the left side. Note how round and elongated spermatids are largely separated from the rest of the data. Spermatogonias are very similar to each other and could maybe reduced into a single cluster.

Look at the proportion of each cell type in the data

finally, save the data

Wrapping up¶

We have been showing how to simply identify potential cell clusters. At least in this dataset, the cells change from one type to another in a continuous process, so such a hard clustering does not completely reflect biological reality. However, it is a good approximation, as it is illustrated by the differentially expressed genes we could check in each cluster. We introduced how to perform differential expression, and what are the useful values that we get from it (p-value of the test, magnitude of the gene expression compared to all other clusters). Finally, we subsetted the data into a more fine grained cell identification.