Posted by Friends of FSH Research on Nov 4, 2019
by Anna Pakula
Note that this supercedes the preliminary update previously posted, and references more fully analyzed data.
The goal of our studies was to characterize the gene expression levels in DUX4 (+) vs. DUX4 (-) human muscle cells to deepen the understanding of the initiation and progression of FSHD-1. We initially performed single cell sequencing of myoblasts cultivated in differentiation medium up to day 4 in June 2018, for 4 FSHD-1 and 4 Control samples. Each library consisted of ~3000 cells. For selected samples known to express DUX4 at higher level we sequenced ~6000 cells. However, several samples had low representation in the multiplexed libraries. To avoid biases in analysis due to skewed sample representation, selected backup libraries were sequenced in a second batch, in December 2018, and these displayed good balance between the multiplexed libraries.
During this time, Van den Heuvel et al. (Hum Mol Genet, 2019) published the results of a study quite similar to ours. We therefore adjusted the scope of our analysis to include a comparison to their study, with an overview of findings that are confirmed by our data and an exploration of reasons for discrepancies. Some discrepancies may be technical in origin, for example due to the use of different scRNA-seq platforms (10x Chromium in Van den Heuvel et al. vs. inDrops for us), different read lengths (114bp vs. 61bp),or different computational pipelines (Cell Ranger/Monocle vs. dropEst/Seurat). Other discrepancies may be biological in origin, for example due to differences in individual cell-lines, in the muscle types that cells were derived from, or in culture conditions.
As in Van den Heuvel et al. we detected DUX4 transcripts in a small number of FSHD cells and known DUX4 targets in a larger collection of FSHD cells, but the frequencies of such cells were about ten-fold lower in our study. We noticed that our cells did not display as high levels of MYH3 – a marker of late stage differentiation – as those in Van den Heuvel et al., despite being differentiated longer (4 days vs 3.5 days). We think that this is because our cells were plated at low confluency, with little cell-to-cell contact, whereas cells in Van den Heuvel et al. were plated at higher confluency and used EDTA to avoid cell fusion, allowing them to reach later stages of differentiation with higher DUX4 expression.
Because not every transcript in a cell is sequenced during scRNA-seq – a phenomenon known as “dropout” – our data likely underestimates the percentages of cells expressing DUX4 and its target genes. Using leftover funds from Amis FSH in France, we performed a pilot study using transcript-specific library preparations for scRNA-seq, with primers designed to amplify DUX4, ZSCAN4 and LEUTX. This increased their detection rates to ~1% of cells for DUX4 and ~10% for ZSCAN4 and LEUTX in the two FSHD samples sequenced, compared to rates of ~0.1% or less in the ordinary scRNA-seq data from the same subjects. Although not possible for our pilot transcript-specific study, as this used backup libraries, this strategy can be used to generate matched oridinaryand transcript-specific scRNA-seq datasets, in which cells are linked based on their barcodes. This should allow a more sensitive examination of global transcriptional changes associated with low levels of DUX4 expression.