snRNAseq mRNA library preparation and sequencingSingle-nuclei capture was performed following the 10x Chromium manufacturer’s instructions (Chromium Next GEM Single Cell 3' GEM, Library & Gel Bead Kit v3.1 Cat No 1000121, User Guide). Each sample was analyzed in a separate batch and the majority included one or more replicates. For each batch, 3’ snRNAseq libraries were prepared according to the manufacturer’s instructions. Libraries were sequenced on NextSeq500 (Illumina), NovaSeq6000 (Illumina) or DNBSEQ G400RS (MGI) machines. For NextSeq runs, denatured libraries were loaded on 75 cycle or 150 cycle HighOutput v2.5 cartridges at a final molarity of 1.8 pM. NovaSeq6000 runs were performed at the National Genomics Infrastructure (NGI) at SciLifeLab Stockholm, Sweden. For G400RS runs, 50ng of Illumina-style libraries were converted to circular ssDNA libraries using the MGIEasy Universal Library Conversion Kit (MGI) and employing 5 cycles of conversion PCR. Subsequently, 60 fmol of circular ssDNA was used to make DNA nanoballs (DNBs) using a custom rolling-circle amplification primer (5‘-TCGCCGTATCATTCA AGCAGAAGACG-3’) prior to sequencing. DNBs were loaded on FCL flow cells (MGI) and sequenced on PE100 cartridges with 26 cycles on read1, 160 bases on read 2 and 8 bases single index reads. The following custom primers were used in the sequencing cartridge: Read 1: 5’-ACACTCTTTCCCTACACGACGCTCTTCCGATCT-3’; MDA: 5’-CGTATGCCGTCTTCTGCTTGAATGATACGGCGAC-3’; Read 2: 5’-GTGACTGGAGTTCAGACGTGTGCTCTTCCGATCT-3’; i7 index 5’-CCGTATCATTCAAGCAGAAGACGGCATACGAGAT-3’. Primary data processingAll Illumina-based data were converted to fastq files using bcl2fastq (v2.20.0.422) with the ‘--create-fastq-for-index-reads’ flag. We next demultiplexed the data into one pair of fastq files for each 10x Genomics chip channel using deML (v1.1.3)(42) and allowing up to 2 mismatches. Demultiplexed reads were processed using zUMIs (v2.9.4f)(43), applying filtering and error correction (1 edit-distance) to barcode and UMI sequence. Within zUMIs, cDNA reads were aligned to the human genome (hg38) and gene-wise UMI-counts produced from intronic + exonic gene portions as annotated in Gencode (v35). To account for contaminating reads from ambient RNA in the droplet-based sample preparation, we exported counts from all droplets into Cellbender (v0.2)(44). Running the remove-background command using the CUDA implementation on a RTX3080 (Nvidia), we determined high-probability droplets containing nuclei and produced background subtracted count tables. The data obtained was filtered for nuclei that have between 200 and 15000 unique features (genes) detected and less than 20% mitochondrial feature.