ABSTRACT: scRNA-seq procedure and analysis. ScRNA-seq was performed using the Massively Parallel Single-Cell RNA-sequencing technology (MARS-seq) as described by Jaitin et al., 2014 (PMID: 24531970) EpiSCs were seeded at ~6 x 103 cells/cm2 in 6-well cell-culture plates pre-coated with 15 µg/ml human Fibronectin, and differentiated to APSD. At desired time-points, cells were dissociated using Accutase (Thermo Fisher Scientific, 00-4555-56) for 3 min at 37°C and counted. 500,000 cells per condition were washed with ice-cold FACS buffer (10% (v/v) FBS in PBS) and resuspended into 1 ml ice-cold FACS buffer containing 1 µg/ml DAPI. For in vivo experiments, E6.5/E7.0/E7.75 Tg(Eomes::GFP) BAC transgenic embryos were individually dissociated into 300 μl warmed Trypsin 0.05% (w/v) EDTA for 10 min at 37°C. Digestion was stopped by addition of 600 μl ice-cold FACS buffer followed by centrifugation at 350 x g for 3 min. Cells were washed with 500 μl ice-cold FACS buffer, centrifuged again at 350 x g for 3 min and resuspended into 250 μl ice-cold FACS buffer containing 1 µg/ml DAPI. Single-cells from either in vitro-differentiated cells or transgenic embryos were sorted into Eppendorf Polypropylene U-shaped 384-well Twin Tec PCR Microplates (Thermo Fisher Scientific, 10573035), containing 2 μl of lysis solution (0.2% (v/v) Triton X-100) supplemented with 0.4 U/μl RNasin Ribonuclease Inhibitor (Promega, N2515) and 400 nM indexed RT primer from group 1 (1-96 barcodes) or group 2 (97-192 barcodes), as described in Jaitin et al., 2014. Additionally, 71 WT EpiSCs were sorted into each plate, as spike-in control for batch-effect correction. Capture plates were prepared on the Bravo automated liquid handling robot station (Agilent) using 384-filtered tips (Axygen, 302-82-101). Index sorting was performed using either a FACS Aria III cell sorter (BD Biosciences) or a SONY SH800S cell sorter (Sony Biotechnology) at the DanStem/reNEW Flow Cytometry Platform (University of Copenhagen, Copenhagen, Denmark), gating in SSC-A versus FSC-A to collect live cells, and then in FSC-W versus FSC-A to sort only singlets. For in vivo experiments, only GFPpos cells were sorted to capture Eomes-expressing cell types. Immediately after sorting, plates were spun down, snap-frozen on dry ice, and stored at -80°C until further processing. Semi-automated library preparation was performed as described by Jaitin et al., 2014, using 10-12 cycles of PCR amplification and AMPure XP beads (Agencourt) for purification. DNA concentration was measured with a Qubit Fluorometer (Thermo Fisher Scientific, Q32854), and fragment size was determined with a Fragment analyzer (Advanced Analytical). Libraries were paired-end sequenced on a Next-Seq 500 Sequencer (Illumina) at the DanStem/reNEW Genomics Platform (University of Copenhagen, Copenhagen, Denmark). Between 1,146 and 1,528 cells were sequenced per lane. R1 and R2 fastq files were generated using bcl2fastq (v2.19.1), and the pooling and well information were extracted from the sequence using umis (v1.0.3) [https://github.com/vals/umis] into a unique fastq file. The reads were then filtered based on the pooling barcodes with 1 mismatch allowed. The poly-Ts at the end of the reads were trimmed using Cutadapt (v1.18). The reads were mapped to the mouse genome (GRCm38/mm10 together with ERCC92) using HISAT2 (v2.1.0), the alignments were processed with Samtools (v1.7)118, and the reads were counted with featureCounts (Subread (v1.5.3)) using Ensembl v93, and the UMIs using UMI_tools (v1.0.0). Expression data were analyzed using Seurat (v3.1). Data filtering, normalization, and scaling were performed using the standard pre-processing workflow. Integration of different datasets was performed as described (PMID: 31178118). Spiked-in EpiSCs were used as a reference to correct the batch effect between integrated datasets. Marker genes of each cell cluster were outputted for GO-term analysis to define the cell type. The Monocle package (v2.16.0) was used to perform pseudotime analyses.
